dataset string | model_name string | model_links list | paper_title string | paper_date timestamp[ns] | paper_url string | code_links list | metrics string | table_metrics list | prompts string | paper_text string | compute_hours float64 | num_gpus int64 | reasoning string | trainable_single_gpu_8h string | verified string | modality string | paper_title_drop string | paper_date_drop string | code_links_drop string | num_gpus_drop int64 | dataset_link string | time_and_compute_verification string | link_to_colab_notebook string | run_possible string | notes string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
PDBbind | BAPULM | [] | BAPULM: Binding Affinity Prediction using Language Models | 2024-11-06T00:00:00 | https://arxiv.org/abs/2411.04150v1 | [
"https://github.com/radh55sh/BAPULM"
] | {'RMSE': '0.898±0.0172'} | [
"RMSE"
] | Given the following paper and codebase:
Paper: BAPULM: Binding Affinity Prediction using Language Models
Codebase: https://github.com/radh55sh/BAPULM
Improve the BAPULM model on the PDBbind dataset. The result
should improve on the following metrics: {'RMSE': '0.898±0.0172'}. You must use only the code... | BAPULM: Binding Affinity Prediction using Language Models Radheesh Sharma Meda†and Amir Barati Farimani∗,‡,¶,†,§ †Department of Chemical Engineering, Carnegie Mellon University, 15213, USA ‡Department of Mechanical Engineering, Carnegie Mellon University, 15213, USA ¶Department of Biomedical Engineering, Carnegie Mello... | 1 | 1 | The model uses ProtT5-XL-U50 and MolFormer architectures, which are large transformer-based models. Given that training on an Nvidia RTX 2080 Ti took approximately 4 minutes, and assuming training occurs over a reduced dataset with 100k sequences, with a complex architecture having a moderate number of parameters, a si... | yes | Yes | Bioinformatics | BAPULM: Binding Affinity Prediction using Language Models | 2024-11-06 0:00:00 | https://github.com/radh55sh/BAPULM | 1 | https://huggingface.co/datasets/radh25sh/BAPULM/resolve/main/prottrans_molformer_tensor_dataset100k.json?download=true | 16sec * 60 epochs = 16 minutes | https://colab.research.google.com/drive/1--rNlCN01wUgN_6cTTuiVcusqSP9vGlG?usp=sharing | Yes | -- no pdbind dataset.Specifices to use prottrans malformer |
Digital twin-supported deep learning for fault diagnosis | DANN | [] | A domain adaptation neural network for digital twin-supported fault diagnosis | 2025-05-27T00:00:00 | https://arxiv.org/abs/2505.21046v1 | [
"https://github.com/JialingRichard/Digital-Twin-Fault-Diagnosis"
] | {'Accuray': '80.22'} | [
"Accuray"
] | Given the following paper and codebase:
Paper: A domain adaptation neural network for digital twin-supported fault diagnosis
Codebase: https://github.com/JialingRichard/Digital-Twin-Fault-Diagnosis
Improve the DANN model on the Digital twin-supported deep learning for fault diagnosis dataset. The result
... | A domain adaptation neural network for digital twin-supported fault diagnosis Zhenling Chen CentraleSupélec, Université Paris-Saclay, Gif-sur-Yvette, 91190, FranceHaiwei Fu CentraleSupélec, Université Paris-Saclay, Gif-sur-Yvette, 91190, France Zhiguo Zeng Chair on Risk and Resilience of Complex Systems, Laboratoie Gen... | 2 | 1 | The DANN model employs a CNN architecture with two convolutional layers. Given the specified batch size of 32 and 250 training epochs on a dataset with 3,600 samples (360 samples per class for 9 distinct labels, plus a significantly smaller test set of 90 samples), the total iterations required for training would be (3... | yes | Yes | Time Series | A domain adaptation neural network for digital twin-supported fault diagnosis | 2025-05-27T00:00:00.000Z | [https://github.com/JialingRichard/Digital-Twin-Fault-Diagnosis] | 1 | Included in Repo | 3 Hours | Copy of train_ai_pytorch_DANN.ipynb | Yes | It starts and runs successfully |
MNIST | GatedGCN+ | [] | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00 | https://arxiv.org/abs/2502.09263v1 | [
"https://github.com/LUOyk1999/GNNPlus"
] | {'Accuracy': '98.712 ± 0.137'} | [
"Accuracy"
] | Given the following paper and codebase:
Paper: Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence
Codebase: https://github.com/LUOyk1999/GNNPlus
Improve the GatedGCN+ model on the MNIST dataset. The result
should improve on the following metrics: {'Accur... | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence Yuankai Luo1 2Lei Shi*1Xiao-Ming Wu*2 Abstract Message-passing Graph Neural Networks (GNNs) are often criticized for their limited expres- siveness, issues like over-smoothing and over- squashing, and challenges in captu... | 4 | 1 | The GNN models (GCN, GIN, and GatedGCN) enhanced with GNN+ have approximately 500K parameters each, which is moderate for graph neural networks. The datasets used involve a variety of sizes, but the mentioned ones have a maximum of around 500K graphs (like the OGB datasets). Given the average training time of these mod... | yes | Yes | Graph | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00.000Z | [https://github.com/LUOyk1999/GNNPlus] | 1 | https://data.pyg.org/datasets/benchmarking-gnns/MNIST_v2.zip | 9 hour approx - ( 200 epochs * avg 157.2 sec) | https://drive.google.com/file/d/1Y7jMNhNybbdgrUJa_MxcOrbwpJNkDPav/view?usp=sharing | Yes | null |
ogbg-molhiv | GatedGCN+ | [] | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00 | https://arxiv.org/abs/2502.09263v1 | [
"https://github.com/LUOyk1999/GNNPlus"
] | {'Test ROC-AUC': '0.8040 ± 0.0164', 'Validation ROC-AUC': '0.8329 ± 0.0158', 'Number of params': '1076633', 'Ext. data': 'No'} | [
"Test ROC-AUC",
"Ext. data",
"Validation ROC-AUC",
"Number of params"
] | Given the following paper and codebase:
Paper: Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence
Codebase: https://github.com/LUOyk1999/GNNPlus
Improve the GatedGCN+ model on the ogbg-molhiv dataset. The result
should improve on the following metrics: {... | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence Yuankai Luo1 2Lei Shi*1Xiao-Ming Wu*2 Abstract Message-passing Graph Neural Networks (GNNs) are often criticized for their limited expres- siveness, issues like over-smoothing and over- squashing, and challenges in captu... | 4 | 1 | The paper describes training across 14 well-known graph-level datasets with a mean parameter count of approximately 500K for classic GNNs, which is manageable for modern GPUs. Assuming training occurs over 2000 epochs, the time per epoch for the enhanced GNNs is reported to be less than that for SOTA GTs, suggesting it... | yes | Yes | Graph | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00.000Z | [https://github.com/LUOyk1999/GNNPlus] | 1 | http://snap.stanford.edu/ogb/data/graphproppred/csv_mol_download/hiv.zip | approx 40 min - ( 100 epochs * 22.8s) | https://drive.google.com/file/d/1Y7jMNhNybbdgrUJa_MxcOrbwpJNkDPav/view?usp=sharing | Yes | null |
Fashion-MNIST | Continued fraction of straight lines | [] | Real-valued continued fraction of straight lines | 2024-12-16T00:00:00 | https://arxiv.org/abs/2412.16191v1 | [
"https://github.com/grasshopper14/Continued-fraction-of-straight-lines/blob/main/continued_fraction_reg.py"
] | {'Accuracy': '84.12', 'Trainable Parameters': '7870', 'NMI': '74.4'} | [
"Percentage error",
"Accuracy",
"Trainable Parameters",
"NMI",
"Power consumption"
] | Given the following paper and codebase:
Paper: Real-valued continued fraction of straight lines
Codebase: https://github.com/grasshopper14/Continued-fraction-of-straight-lines/blob/main/continued_fraction_reg.py
Improve the Continued fraction of straight lines model on the Fashion-MNIST dataset. The result... | Real-valued continued fraction of straight lines Vijay Prakash S Alappuzha, Kerala, India. prakash.vijay.s@gmail.com Abstract In an unbounded plane, straight lines are used extensively for mathematical analysis. They are tools of conve- nience. However, those with high slope values become unbounded at a faster rate tha... | 4 | 1 | The model is trained on the Fashion-MNIST dataset, which consists of 60,000 training images and 10,000 testing images, each with a size of 28x28 pixels (784 input features). The training procedure described in the paper involves mini-batch gradient descent with 100 batches of 600 samples each for 50 iterations (or epoc... | yes | Yes | CV | Real-valued continued fraction of straight lines | 2024-12-16T00:00:00.000Z | [https://github.com/grasshopper14/Continued-fraction-of-straight-lines/blob/main/continued_fraction_reg.py] | 1 | https://github.com/zalandoresearch/fashion-mnist | 20 min | https://colab.research.google.com/drive/1LNMCRLMIWN5U_9WDeRxYmcbnAgaNadSd?usp=sharing | Yes | Yes Everythng is running successfully |
Traffic | GLinear | [] | Bridging Simplicity and Sophistication using GLinear: A Novel Architecture for Enhanced Time Series Prediction | 2025-01-02T00:00:00 | https://arxiv.org/abs/2501.01087v3 | [
"https://github.com/t-rizvi/GLinear"
] | {'MSE ': '0.3222'} | [
"MSE "
] | Given the following paper and codebase:
Paper: Bridging Simplicity and Sophistication using GLinear: A Novel Architecture for Enhanced Time Series Prediction
Codebase: https://github.com/t-rizvi/GLinear
Improve the GLinear model on the Traffic dataset. The result
should improve on the following metrics... | IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTATIONAL INTELLIGENCE JOURNAL, 2025 1 Bridging Simplicity and Sophistication using GLinear: A Novel Architecture for Enhanced Time Series Prediction Syed Tahir Hussain Rizvi1 , Neel Kanwal1 , Muddasar Naeem2 , Alfredo Cuzzocrea3∗and Antonio Coronato2 1Department of Electrica... | 4 | 1 | The GLinear model, being a simplified architecture without complex components like Transformers, should have a relatively low parameter count compared to complex models. The datasets used are of manageable sizes, with the largest having around 50,000 time steps and multiple channels, which is well within the processing... | yes | Yes | Time Series | Bridging Simplicity and Sophistication using GLinear: A Novel Architecture for Enhanced Time Series Prediction | 2025-01-02 0:00:00 | https://github.com/t-rizvi/GLinear | 1 | Inside the repo in dataset folder | 193 sec * 4 = 12.9 minutes | https://colab.research.google.com/drive/1sI72VSxjN4cyQR7UrueWfBXwoFi9Y9Qr?usp=sharing | Yes | -- Training on all data set is included inside the scripts/EXP-LookBackWindow_\&_LongForecasting/Linear_LookBackWindow.sh fiLE . But to run only traffic dataset. I have included the conda command. |
BTAD | URD | [] | Unlocking the Potential of Reverse Distillation for Anomaly Detection | 2024-12-10T00:00:00 | https://arxiv.org/abs/2412.07579v1 | [
"https://github.com/hito2448/urd"
] | {'Segmentation AUROC': '98.1', 'Detection AUROC': '93.9', 'Segmentation AUPRO': '78.5', 'Segmentation AP': '65.2'} | [
"Detection AUROC",
"Segmentation AUROC",
"Segmentation AP",
"Segmentation AUPRO"
] | Given the following paper and codebase:
Paper: Unlocking the Potential of Reverse Distillation for Anomaly Detection
Codebase: https://github.com/hito2448/urd
Improve the URD model on the BTAD dataset. The result
should improve on the following metrics: {'Segmentation AUROC': '98.1', 'Detection AUROC':... | Unlocking the Potential of Reverse Distillation for Anomaly Detection Xinyue Liu1, Jianyuan Wang2*, Biao Leng1, Shuo Zhang3 1School of Computer Science and Engineering, Beihang University 2School of Intelligence Science and Technology, University of Science and Technology Beijing 3Beijing Key Lab of Traffic Data Analys... | 4 | 1 | The proposed method utilizes a WideResNet50 architecture as a teacher network which typically has about 68 million parameters. Given the dataset size of around 5354 images with a training batch size of 16, the model is expected to go through multiple epochs for convergence, likely around 100 epochs based on common prac... | yes | Yes | CV | Unlocking the Potential of Reverse Distillation for Anomaly Detection | 2024-12-10 0:00:00 | https://github.com/hito2448/urd | 1 | https://www.mydrive.ch/shares/38536/3830184030e49fe74747669442f0f282/download/420938113-1629952094/mvtec_anomaly_detection.tar.xz; https://www.robots.ox.ac.uk/~vgg/data/dtd/download/dtd-r1.0.1.tar.gz | 8 hours for one folder. There are 11 folders. | https://drive.google.com/file/d/1OLbo3FifM1a7-wbCtfpjZrZLr0K5bS87/view?usp=sharing | Yes | -- Just need to change the num_workers in train.py according to system |
York Urban Dataset | DT-LSD | [] | DT-LSD: Deformable Transformer-based Line Segment Detection | 2024-11-20T00:00:00 | https://arxiv.org/abs/2411.13005v1 | [
"https://github.com/SebastianJanampa/DT-LSD"
] | {'sAP5': '30.2', 'sAP10': '33.2', 'sAP15': '35.1'} | [
"sAP5",
"sAP10",
"sAP15",
"FH"
] | Given the following paper and codebase:
Paper: DT-LSD: Deformable Transformer-based Line Segment Detection
Codebase: https://github.com/SebastianJanampa/DT-LSD
Improve the DT-LSD model on the York Urban Dataset dataset. The result
should improve on the following metrics: {'sAP5': '30.2', 'sAP10': '33.2... | DT-LSD: Deformable Transformer-based Line Segment Detection Sebastian Janampa The University of New Mexico sebasjr1966@unm.eduMarios Pattichis The University of New Mexico pattichi@unm.edu Abstract Line segment detection is a fundamental low-level task in computer vision, and improvements in this task can im- pact more... | 4 | 1 | The proposed DT-LSD model has a relatively small batch size of 2 and uses a single Nvidia RTX A5500 GPU, which has sufficient memory (24 GB) to handle the model's parameters and intermediate activations. With a total of 24 epochs and leveraging the efficient Line Contrastive Denoising training technique, the training t... | yes | Yes | CV | DT-LSD: Deformable Transformer-based Line Segment Detection | 2024-11-20 0:00:00 | https://github.com/SebastianJanampa/DT-LSD | 1 | script to download is provided in colab file. | uses cpu to trainf or some reason 8hr per epoch | https://colab.research.google.com/drive/1XPiW-hDq6q8HNZ4yVP0oAn-3a1_ay5rG?usp=sharing | Yes | -- Trains but uses cpu for some reason |
UCR Anomaly Archive | KAN | [] | KAN-AD: Time Series Anomaly Detection with Kolmogorov-Arnold Networks | 2024-11-01T00:00:00 | https://arxiv.org/abs/2411.00278v1 | [
"https://github.com/issaccv/KAN-AD"
] | {'AUC ROC ': '0.7489'} | [
"Average F1",
"AUC ROC "
] | Given the following paper and codebase:
Paper: KAN-AD: Time Series Anomaly Detection with Kolmogorov-Arnold Networks
Codebase: https://github.com/issaccv/KAN-AD
Improve the KAN model on the UCR Anomaly Archive dataset. The result
should improve on the following metrics: {'AUC ROC ': '0.7489'}. You must... | KAN-AD: Time Series Anomaly Detection with Kolmogorov–Arnold Networks Quan Zhou*, Changhua Pei, Haiming Zhang, Gaogang Xie, Jianhui Li† Computer Network Information Center Chinese Academy of Science zhouquan,chpei,hai,xie,lijh@cnic.cnFei Sun Institution of Computing Technology Chinese Academy of Science sunfei@ict.ac.c... | 4 | 1 | The KAN-AD model is based on a novel architecture that leverages Fourier series for anomaly detection in time series, which would imply a moderate computational overhead given the 1D CNN architecture with stacked layers for coefficient learning. The training dataset size varies per dataset, with the largest (KPI) conta... | yes | Yes | Time Series | KAN-AD: Time Series Anomaly Detection with Kolmogorov-Arnold Networks | 2024-11-01 0:00:00 | https://github.com/issaccv/KAN-AD | 1 | Downloaded when running prepeare_env.sh from repository & uses UTS dataset, https://github.com/CSTCloudOps/datasets | There are 5 folders. May take around 2 hours or more no idea as time was not specified and traing was happening fast. | https://colab.research.google.com/drive/1sE1mKwy3n9yameE-JG27Oa_HI-q8lFn9?usp=sharing | Yes | -- After the installation of environment.sh. I changed a line of code to run matplot lib on colab and need to fix the typo on .bin file which i have mentioned in colab file. It takes 10 min to install env on colab with requirements. |
Chameleon | CoED | [] | Improving Graph Neural Networks by Learning Continuous Edge Directions | 2024-10-18T00:00:00 | https://arxiv.org/abs/2410.14109v1 | [
"https://github.com/hormoz-lab/coed-gnn"
] | {'Accuracy': '79.69±1.35'} | [
"Accuracy"
] | Given the following paper and codebase:
Paper: Improving Graph Neural Networks by Learning Continuous Edge Directions
Codebase: https://github.com/hormoz-lab/coed-gnn
Improve the CoED model on the Chameleon dataset. The result
should improve on the following metrics: {'Accuracy': '79.69±1.35'}. You mus... | Preprint IMPROVING GRAPH NEURAL NETWORKS BY LEARN - INGCONTINUOUS EDGE DIRECTIONS Seong Ho Pahng1, 2& Sahand Hormoz3, 2, 4 1Department of Chemistry and Chemical Biology, Harvard University 2Department of Data Science, Dana-Farber Cancer Institute 3Department of Systems Biology, Harvard Medical School 4Broad Institute o... | 4 | 1 | The proposed CoED GNN is a graph neural network architecture that utilizes a complex-valued Laplacian with directed edges. Given the nature of GNNs and from insights into existing literature, the complexity of the model estimates that a typical training session could be reasonably completed in under 8 hours. The model ... | yes | Yes | Graph | Improving Graph Neural Networks by Learning Continuous Edge Directions | 2024-10-18 0:00:00 | https://github.com/hormoz-lab/coed-gnn | 1 | specify on the classification.py and it handles itself | 2 min | https://colab.research.google.com/drive/1FiCFbVmQhjIqcCdViYynfEb9mWtJkB09?usp=sharing | Yes | -- I have put the best parameter with advice of "Gemini" Can change accordingly. |
California Housing Prices | Binary Diffusion | [] | Tabular Data Generation using Binary Diffusion | 2024-09-20T00:00:00 | https://arxiv.org/abs/2409.13882v2 | [
"https://github.com/vkinakh/binary-diffusion-tabular"
] | {'Parameters(M)': '1.5', 'RF Mean Squared Error': '0.39', 'LR Mean Squared Error': '0.55', 'DT Mean Squared Error': '0.45'} | [
"Parameters(M)",
"RF Mean Squared Error",
"DT Mean Squared Error",
"LR Mean Squared Error"
] | Given the following paper and codebase:
Paper: Tabular Data Generation using Binary Diffusion
Codebase: https://github.com/vkinakh/binary-diffusion-tabular
Improve the Binary Diffusion model on the California Housing Prices dataset. The result
should improve on the following metrics: {'Parameters(M)': ... | Tabular Data Generation using Binary Diffusion Vitaliy Kinakh Department of Computer Science University of Geneva Geneva, Switzerland vitaliy.kinakh@unige.chSlava Voloshynovskiy Department of Computer Science University of Geneva Geneva, Switzerland Abstract Generating synthetic tabular data is critical in machine lear... | 4 | 1 | The proposed Binary Diffusion model has fewer than 2 million parameters, making it lightweight compared to contemporary models that often exceed 100 million parameters. Given its focus on binary data, the model architecture is likely simpler, which will lead to faster training times. The training is performed on benchm... | yes | Yes | Tabular | Tabular Data Generation using Binary Diffusion | 2024-09-20 0:00:00 | https://github.com/vkinakh/binary-diffusion-tabular | 1 | inside the project repo | around 2 hours | https://drive.google.com/file/d/154F-06anE1dsOik9zkn3uBqcw9t3Lz53/view?usp=sharing | Yes | -- i put some line of code in colab to make sure it runs. Please check the colab file for more info. |
Kvasir-SEG | Yolo-SAM 2 | [] | Self-Prompting Polyp Segmentation in Colonoscopy using Hybrid Yolo-SAM 2 Model | 2024-09-14T00:00:00 | https://arxiv.org/abs/2409.09484v1 | [
"https://github.com/sajjad-sh33/yolo_sam2"
] | {'mean Dice': '0.866', 'mIoU': '0.764'} | [
"mean Dice",
"Average MAE",
"S-Measure",
"max E-Measure",
"mIoU",
"FPS",
"F-measure",
"Precision",
"Recall"
] | Given the following paper and codebase:
Paper: Self-Prompting Polyp Segmentation in Colonoscopy using Hybrid Yolo-SAM 2 Model
Codebase: https://github.com/sajjad-sh33/yolo_sam2
Improve the Yolo-SAM 2 model on the Kvasir-SEG dataset. The result
should improve on the following metrics: {'mean Dice': '0.8... | SELF-PROMPTING POLYP SEGMENTATION IN COLONOSCOPY USING HYBRID YOLO-SAM 2 MODEL Mobina Mansoori†, Sajjad Shahabodini†, Jamshid Abouei††, Konstantinos N. Plataniotis‡, and Arash Mohammadi† †Intelligent Signal & Information Processing (I-SIP) Lab, Concodia University, Canada ‡Edward S. Rogers Sr. Department of Electrical ... | 4 | 1 | The YOLOv8 medium model has 25 million parameters and the SAM 2 large model has 224.4 million parameters. With a batch size of 64 and an input image size of 680, it is fairly demanding but feasible on a single GPU, especially since the paper states they used an A100 GPU (40 GB). Given the datasets involved (around 5,00... | yes | Yes | CV | Self-Prompting Polyp Segmentation in Colonoscopy using Hybrid Yolo-SAM 2 Model | 2024-09-14 0:00:00 | https://github.com/sajjad-sh33/yolo_sam2 | 1 | downloaded from kaggle https://www.kaggle.com/datasets/debeshjha1/kvasirseg | 40sec * 50 epoch = 33.33 minutes | https://colab.research.google.com/drive/1_iOHO7njejU5yFtKPoF2477d_H0Cw4tf?usp=sharing | Yes | -- Fine tuning the model. I have patched the code and also put instuctions on how to prepare data and fix the python file for Kvasir dataset. |
Office-31 | EUDA | [] | EUDA: An Efficient Unsupervised Domain Adaptation via Self-Supervised Vision Transformer | 2024-07-31T00:00:00 | https://arxiv.org/abs/2407.21311v1 | [
"https://github.com/a-abedi/euda"
] | {'Accuracy': '92'} | [
"Accuracy",
"Avg accuracy"
] | Given the following paper and codebase:
Paper: EUDA: An Efficient Unsupervised Domain Adaptation via Self-Supervised Vision Transformer
Codebase: https://github.com/a-abedi/euda
Improve the EUDA model on the Office-31 dataset. The result
should improve on the following metrics: {'Accuracy': '92'}. You ... | PREPRINT 1 EUDA: An Efficient Unsupervised Domain Adaptation via Self-Supervised Vision Transformer Ali Abedi, Graduate Student Member, IEEE, Q. M. Jonathan Wu, Senior Member, IEEE, Ning Zhang, Senior Member, IEEE, Farhad Pourpanah, Senior Member, IEEE Abstract —Unsupervised domain adaptation (UDA) aims to mitigate the... | 4 | 1 | The EUDA framework utilizes a frozen DINOv2 feature extractor (self-supervised Vision Transformer) and incorporates a bottleneck of fully connected layers. Given the efficiency improvements stated (42% to 99.7% fewer parameters than prior ViT models), it is likely in the range of hundreds of millions of parameters, sim... | yes | Yes | CV | EUDA: An Efficient Unsupervised Domain Adaptation via Self-Supervised Vision Transformer | 2024-07-31 0:00:00 | https://github.com/a-abedi/euda | 1 | https://drive.usercontent.google.com/download?id=0B4IapRTv9pJ1WGZVd1VDMmhwdlE&export=download&authuser=0&resourcekey=0-gNMHVtZfRAyO_t2_WrOunA | 2000 steps × 2.05 sec/step = 4100 seconds ≈ 68 minutes ≈ 1 hour 8 minutes | https://drive.google.com/file/d/1woeCrW4aU_I6LUR6K2N7bh_uUPn5rkAK/view?usp=sharing | Yes | --Need to fix some line of code which i included in the colab file. |
WiGesture | CSI-BERT | [] | Finding the Missing Data: A BERT-inspired Approach Against Package Loss in Wireless Sensing | 2024-03-19T00:00:00 | https://arxiv.org/abs/2403.12400v1 | [
"https://github.com/rs2002/csi-bert"
] | {'Accuracy (% )': '93.94'} | [
"Accuracy (% )"
] | Given the following paper and codebase:
Paper: Finding the Missing Data: A BERT-inspired Approach Against Package Loss in Wireless Sensing
Codebase: https://github.com/rs2002/csi-bert
Improve the CSI-BERT model on the WiGesture dataset. The result
should improve on the following metrics: {'Accuracy (% ... | Finding the Missing Data: A BERT-inspired Approach Against Package Loss in Wireless Sensing Zijian Zhao∗†, Tingwei Chen∗, Fanyi Meng∗‡, Hang Li∗, Xiaoyang Li∗, Guangxu Zhu∗ ∗Shenzhen Research Institute of Big Data †School of Computer Science and Engineering, Sun Yat-sen University ‡School of Science and Engineering, Th... | 4 | 1 | The CSI-BERT model has approximately 2.11 million parameters, similar in scale to other models like BERT-base, which has around 110 million parameters. Given that the dataset entails wireless Channel State Information (CSI) samples collected at 100Hz, with an average of 14.51% loss rate, we estimate the dataset is mana... | yes | Yes | Signal Processing | Finding the Missing Data: A BERT-inspired Approach Against Package Loss in Wireless Sensing | 2024-03-19 0:00:00 | https://github.com/rs2002/csi-bert | 1 | http://www.sdp8.net/Dataset?id=5d4ee7ca-d0b0-45e3-9510-abb6e9cdebf9 | around 2 hours estimated. | https://colab.research.google.com/drive/1ijfudC_ZodlZSMvHtHgcLEvLwfWVF6-i?usp=sharing | Yes | -- Login and download the dataset or inside the repo it is present. |
Astock | SRL&Factors | [] | FinReport: Explainable Stock Earnings Forecasting via News Factor Analyzing Model | 2024-03-05T00:00:00 | https://arxiv.org/abs/2403.02647v1 | [
"https://github.com/frinkleko/finreport"
] | {'Accuray': '69.48', 'F1-score': '69.28', 'Recall': '69.41', 'Precision': '69.54'} | [
"Accuray",
"F1-score",
"Recall",
"Precision"
] | Given the following paper and codebase:
Paper: FinReport: Explainable Stock Earnings Forecasting via News Factor Analyzing Model
Codebase: https://github.com/frinkleko/finreport
Improve the SRL&Factors model on the Astock dataset. The result
should improve on the following metrics: {'Accuray': '69.48',... | FinReport: Explainable Stock Earnings Forecasting via News Factor Analyzing Model Xiangyu Li∗ 65603605lxy@gmail.com South China University of TechnologyXinjie Shen∗ frinkleko@gmail.com South China University of TechnologyYawen Zeng yawenzeng11@gmail.com ByteDance AI Lab Xiaofen Xing† xfxing@scut.edu.cn South China Univ... | 4 | 1 | The model includes multiple modules (news factorization, return forecasting, risk assessment) but seems to utilize established architectures like RoBERTa for the news factorization, which could have a manageable parameter count. The dataset, Astock, has a significant amount of historical data over more than three years... | yes | Yes | NLP | FinReport: Explainable Stock Earnings Forecasting via News Factor Analyzing Model | 2024-03-05 0:00:00 | https://github.com/frinkleko/finreport | 1 | isndie the repo . | under 5 minutes | https://colab.research.google.com/drive/1G6z0MNnOdpYGIu6F2wPc69cd-fbUWjsr?usp=sharing | Yes | -- Just run this colab file. I ahave include the dataextraction process from the repo and passed into path corrctly. This ipynb file is downloaded from repo itself. |
Fashion-MNIST | ENERGIZE | [] | Towards Physical Plausibility in Neuroevolution Systems | 2024-01-31T00:00:00 | https://arxiv.org/abs/2401.17733v1 | [
"https://github.com/rodriguesGabriel/energize"
] | {'Percentage error': '9.8', 'Accuracy': '0.902', 'Power consumption': '71.92'} | [
"Percentage error",
"Accuracy",
"Trainable Parameters",
"NMI",
"Power consumption"
] | Given the following paper and codebase:
Paper: Towards Physical Plausibility in Neuroevolution Systems
Codebase: https://github.com/rodriguesGabriel/energize
Improve the ENERGIZE model on the Fashion-MNIST dataset. The result
should improve on the following metrics: {'Percentage error': '9.8', 'Accurac... | arXiv:2401.17733v1 [cs.NE] 31 Jan 2024Towards Physical Plausibility in Neuroevolution Systems Gabriel Cortês[0000 −0001 −6318 −8520], Nuno Lourenço[0000 −0002 −2154 −0642], and Penousal Machado[0000 −0002 −6308 −6484] University of Coimbra, CISUC/LASI – Centre for Informatics and Systems of the University of Coimbra, D... | 4 | 1 | The study utilizes Fast-DENSER on the Fashion-MNIST dataset, which has 60,000 training images. Given the detailed architecture modifications for training two separate models simultaneously, I estimate the training time based on the complexity of multiple evolutionary computations and a default training time of 10 minut... | yes | Yes | CV | Towards Physical Plausibility in Neuroevolution Systems | 2024-01-31 0:00:00 | https://github.com/rodriguesGabriel/energize | 1 | downlaoded by training script | Max runtime = generations × population_size × train_time_per_individual
= 150 × 4 × 300 seconds
= 180,000 seconds
= 50 hours (plus overhead for evaluation, logging, mutation, etc.)
| https://drive.google.com/file/d/1ToU-VDe6i5AXDihxb_T3v7gNC6iEP9ng/view?usp=sharing | Yes | -- Straight forward just change -d while calling the train script. I have included the arguments for the train file in colab |
Fashion-MNIST | GECCO | [] | A Single Graph Convolution Is All You Need: Efficient Grayscale Image Classification | 2024-02-01T00:00:00 | https://arxiv.org/abs/2402.00564v6 | [
"https://github.com/geccoproject/gecco"
] | {'Percentage error': '11.91', 'Accuracy': '88.09'} | [
"Percentage error",
"Accuracy",
"Trainable Parameters",
"NMI",
"Power consumption"
] | Given the following paper and codebase:
Paper: A Single Graph Convolution Is All You Need: Efficient Grayscale Image Classification
Codebase: https://github.com/geccoproject/gecco
Improve the GECCO model on the Fashion-MNIST dataset. The result
should improve on the following metrics: {'Percentage erro... | A SINGLE GRAPH CONVOLUTION IS ALL YOU NEED: EFFICIENT GRAYSCALE IMAGE CLASSIFICATION Jacob Fein-Ashley†, Sachini Wickramasinghe†, Bingyi Zhang†, Rajgopal Kannan∗, Viktor Prasanna† †University of Southern California,∗DEVCOM Army Research Office ABSTRACT Image classifiers for domain-specific tasks like Synthetic Aperture... | 4 | 1 | The GECCO model is lightweight with a relatively low number of parameters (approx. 5.08M) and uses simple architecture elements (single GCN layer and MLP). The MSTAR dataset consists of 2747 training samples and 2425 testing samples of 128x128 pixels, and the CXR dataset consists of 5216 training samples and requires l... | yes | Yes | CV | A Single Graph Convolution Is All You Need: Efficient Grayscale Image Classification | 2024-02-01 0:00:00 | https://github.com/geccoproject/gecco | 1 | downloaded by training script | 20s * 1000 epoch = 5.5 hr approx | https://drive.google.com/file/d/1b72abDo06zMcoMYcEnbhx-eryDxMP2G0/view?usp=sharing | Yes | -- Need to make some fixes for fashion mnsit . I have included the changes in colab file please follow that |
MNIST | rKAN | [] | rKAN: Rational Kolmogorov-Arnold Networks | 2024-06-20T00:00:00 | https://arxiv.org/abs/2406.14495v1 | [
"https://github.com/alirezaafzalaghaei/rkan"
] | {'Accuracy': '99.293'} | [
"Percentage error",
"Accuracy",
"Trainable Parameters",
"Cross Entropy Loss",
"Epochs",
"Top 1 Accuracy"
] | Given the following paper and codebase:
Paper: rKAN: Rational Kolmogorov-Arnold Networks
Codebase: https://github.com/alirezaafzalaghaei/rkan
Improve the rKAN model on the MNIST dataset. The result
should improve on the following metrics: {'Accuracy': '99.293'}. You must use only the codebase provided.... | rKAN: Rational Kolmogorov-Arnold Networks Alireza Afzal Aghaei Independent Researcher Email: alirezaafzalaghaei@gmail.com June 21, 2024 Abstract The development of Kolmogorov-Arnold networks (KANs) marks a significant shift from traditional multi-layer perceptrons in deep learning. Initially, KANs employed B-spline cur... | 4 | 1 | The model described (rKAN) is similar to existing neural network architectures in complexity. It has a manageable architecture (1-10-1 for regression tasks) and a batch size of 512 for the MNIST classification task. Training on the MNIST dataset (60,000 training images) for 30 epochs with a relatively simple architectu... | yes | Yes | CV | rKAN: Rational Kolmogorov-Arnold Networks | 2024-06-20 0:00:00 | https://github.com/alirezaafzalaghaei/rkan | 1 | In Code | 1 | cnn.ipynb | Yes | null |
Tiny ImageNet Classification | MANO-tiny | [] | Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics | 2025-07-03T00:00:00 | https://arxiv.org/abs/2507.02748 | [
"https://github.com/AlexColagrande/MANO"
] | {'Validation Acc': '87.52'} | [
"Validation Acc"
] | Given the following paper and codebase:
Paper: Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics
Codebase: https://github.com/AlexColagrande/MANO
Improve the MANO-tiny model on the Tiny ImageNet Classification dataset. The result
should improve on the followin... | arXiv:2507.02748v1 [cs.CV] 3 Jul 2025Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics Alex Colagrande1, Paul Caillon1, Eva Feillet1, Alexandre Allauzen1,2 1Miles Team, LAMSADE, Universit ´e Paris Dauphine-PSL, Paris, France 2ESPCI PSL, Paris, France {name}.{surname }@dauphine... | 5 | 1 | The model described has approximately 28 million parameters, which is comparable to other lightweight vision transformer models known to train in a reasonable timeframe. Considering the architecture's efficiency with linear complexity for attention, and the fact it utilizes a modified Swin Transformer backbone, a typic... | yes | Yes | CV | Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics | 2025-07-03T00:00:00.000Z | [https://github.com/AlexColagrande/MANO] | 1 | Code Downloads Dynamically upon naming | same | Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics.ipynb | Yes | It starts and runs successfully |
Food-101 | MANO-tiny | [] | Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics | 2025-07-03T00:00:00 | https://arxiv.org/abs/2507.02748 | [
"https://github.com/AlexColagrande/MANO"
] | {'Accuracy (%)': '82.48'} | [
"Accuracy (%)",
"Accuracy"
] | Given the following paper and codebase:
Paper: Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics
Codebase: https://github.com/AlexColagrande/MANO
Improve the MANO-tiny model on the Food-101 dataset. The result
should improve on the following metrics: {'Accurac... | arXiv:2507.02748v1 [cs.CV] 3 Jul 2025Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics Alex Colagrande1, Paul Caillon1, Eva Feillet1, Alexandre Allauzen1,2 1Miles Team, LAMSADE, Universit ´e Paris Dauphine-PSL, Paris, France 2ESPCI PSL, Paris, France {name}.{surname }@dauphine... | 5 | 1 | The MANO model is based on the 'Tiny' version of the Swin Transformer V2, which has approximately 28.47M parameters, leading to a manageable memory footprint on modern GPUs. Given that the training is conducted on the ImageNet-1k dataset and several other benchmarks for a total of 50 epochs, the expected total training... | yes | Yes | CV | Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics | 2025-07-03T00:00:00.000Z | [https://github.com/AlexColagrande/MANO] | 1 | Code Downloads Dynamically upon naming | same | Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics.ipynb | Yes | It starts and runs successfully |
Gowalla | RLAE-DAN | [] | Why is Normalization Necessary for Linear Recommenders? | 2025-04-08T00:00:00 | https://arxiv.org/abs/2504.05805v2 | [
"https://github.com/psm1206/dan"
] | {'Recall@20': '0.1922', 'nDCG@20': '0.1605'} | [
"nDCG@20",
"Recall@20",
"HR@10",
"HR@100",
"PSP@10",
"nDCG@10",
"nDCG@100"
] | Given the following paper and codebase:
Paper: Why is Normalization Necessary for Linear Recommenders?
Codebase: https://github.com/psm1206/dan
Improve the RLAE-DAN model on the Gowalla dataset. The result
should improve on the following metrics: {'Recall@20': '0.1922', 'nDCG@20': '0.1605'}. You must u... | Why is Normalization Necessary for Linear Recommenders? Seongmin Park Sungkyunkwan University Suwon, Republic of Korea psm1206@skku.eduMincheol Yoon Sungkyunkwan University Suwon, Republic of Korea yoon56@skku.eduHye-young Kim Sungkyunkwan University Suwon, Republic of Korea khyaa3966@skku.eduJongwuk Lee∗ Sungkyunkwan ... | 5 | 1 | The model described in the paper is a linear autoencoder (LAE) based on existing LAE architectures, which typically have a smaller parameter count compared to non-linear models. The datasets used for training (like ML-20M, Netflix, and others) are substantial but manageable for a single GPU. Given the simpler architect... | yes | Yes | Graph | Why is Normalization Necessary for Linear Recommenders? | 2025-04-08T00:00:00.000Z | [https://github.com/psm1206/dan] | 1 | https://github.com/psm1206/DAN/tree/main/data | 15 min | https://colab.research.google.com/drive/1euiNcqVAl4SgDK75YJJEP_DGBXQzxd08?usp=sharing | Yes | Everthing is working fine. |
Weather (192) | xPatch | [] | xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition | 2024-12-23T00:00:00 | https://arxiv.org/abs/2412.17323v2 | [
"https://github.com/stitsyuk/xpatch"
] | {'MSE': '0.189', 'MAE': '0.227'} | [
"MSE",
"MAE",
"Accuracy"
] | Given the following paper and codebase:
Paper: xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition
Codebase: https://github.com/stitsyuk/xpatch
Improve the xPatch model on the Weather (192) dataset. The result
should improve on the following metrics: {'MSE': '0.189... | xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition Artyom Stitsyuk1, Jaesik Choi1,2 1Korea Advanced Institute of Science and Technology (KAIST), South Korea 2INEEJI, South Korea {stitsyuk, jaesik.choi }@kaist.ac.kr Abstract In recent years, the application of transformer-based mod... | 5 | 1 | The xPatch model employs a dual-stream architecture leveraging MLP and CNN components for time series forecasting. Given the nature of time series data, an estimated dataset size around 100,000 to 1,000,000 samples (typical for LTSF tasks) is reasonable. The model is likely to have a moderate parameter count, roughly c... | yes | Yes | Time Series | xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition | 2024-12-23T00:00:00.000Z | [https://github.com/stitsyuk/xpatch] | 1 | https://drive.usercontent.google.com/download?id=1NF7VEefXCmXuWNbnNe858WvQAkJ_7wuP&export=download&authuser=0 | 30 min | https://colab.research.google.com/drive/1JaT0PQUcJJLSUpemXlylsIULjRvMCW1G?usp=sharing | Yes sussessfully run | It is successfully run and working fine |
CNN/Daily Mail | Claude Instant + SigExt | [] | Salient Information Prompting to Steer Content in Prompt-based Abstractive Summarization | 2024-10-03T00:00:00 | https://arxiv.org/abs/2410.02741v2 | [
"https://github.com/amazon-science/SigExt"
] | {'ROUGE-1': '42', 'ROUGE-L': '26.6'} | [
"ROUGE-1",
"ROUGE-2",
"ROUGE-L"
] | Given the following paper and codebase:
Paper: Salient Information Prompting to Steer Content in Prompt-based Abstractive Summarization
Codebase: https://github.com/amazon-science/SigExt
Improve the Claude Instant + SigExt model on the CNN/Daily Mail dataset. The result
should improve on the following ... | Salient Information Prompting to Steer Content in Prompt-based Abstractive Summarization Lei Xu1, Mohammed Asad Karim2†, Saket Dingliwal1, Aparna Elangovan1 1Amazon AWS AI Labs 2Carnegie Mellon University {leixx, skdin, aeg}@amazon.com mkarim2@cs.cmu.edu Abstract Large language models (LLMs) can generate fluent summari... | 5 | 1 | The paper describes SigExt, which uses a fine-tuned Longformer with 433M parameters. The model is trained on a dataset consisting of 1000 to 10000 examples (for the general-purpose variant), with an unspecified batch size, but given the model's size, it can be assumed that a moderate batch size like 16-32 can be used e... | yes | Yes | NLP | Salient Information Prompting to Steer Content in Prompt-based Abstractive Summarization | 2024-10-03 0:00:00 | https://github.com/amazon-science/SigExt | 1 | run the script to download and process data inside the repo | 10 min * 10 = 1hr 40 min | https://colab.research.google.com/drive/1Wzlo_ybMDNuEVs4wDC4GJq93kFwX6rMJ?usp=sharing | Yes | -- Justneed to change the argument while calling the python and need to add some line of code on data process script. I have included all on the colab file |
ETTh1 (720) Multivariate | SparseTSF | [] | SparseTSF: Modeling Long-term Time Series Forecasting with 1k Parameters | 2024-05-02T00:00:00 | https://arxiv.org/abs/2405.00946v2 | [
"https://github.com/lss-1138/SparseTSF"
] | {'MSE': '0.426'} | [
"MSE",
"MAE"
] | Given the following paper and codebase:
Paper: SparseTSF: Modeling Long-term Time Series Forecasting with 1k Parameters
Codebase: https://github.com/lss-1138/SparseTSF
Improve the SparseTSF model on the ETTh1 (720) Multivariate dataset. The result
should improve on the following metrics: {'MSE': '0.426... | SparseTSF: Modeling Long-term Time Series Forecasting with 1kParameters Shengsheng Lin1Weiwei Lin1 2Wentai Wu3Haojun Chen1Junjie Yang1 Abstract This paper introduces SparseTSF, a novel, ex- tremely lightweight model for Long-term Time Series Forecasting (LTSF), designed to address the challenges of modeling complex tem... | 5 | 1 | The SparseTSF model has less than 1,000 parameters, making it significantly lighter than most deep learning models typically trained on time series data. Given the dataset sizes (up to 26,304 timesteps for the Electricity dataset with multiple channels), it's reasonable to estimate that training may require about 5 hou... | yes | Yes | Time Series | SparseTSF: Modeling Long-term Time Series Forecasting with 1k Parameters | 2024-05-02 0:00:00 | https://github.com/lss-1138/SparseTSF | 1 | https://drive.google.com/drive/folders/1ZOYpTUa82_jCcxIdTmyr0LXQfvaM9vIy | 2min | https://colab.research.google.com/drive/1OgVLdCqFrODVu7AdHeI2_N-1qGZe2AcD?usp=sharing | Yes | -- Just download the dataset and run. |
Peptides-struct | GCN+ | [] | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00 | https://arxiv.org/abs/2502.09263v1 | [
"https://github.com/LUOyk1999/GNNPlus"
] | {'MAE': '0.2421 ± 0.0016'} | [
"MAE"
] | Given the following paper and codebase:
Paper: Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence
Codebase: https://github.com/LUOyk1999/GNNPlus
Improve the GCN+ model on the Peptides-struct dataset. The result
should improve on the following metrics: {'... | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence Yuankai Luo1 2Lei Shi*1Xiao-Ming Wu*2 Abstract Message-passing Graph Neural Networks (GNNs) are often criticized for their limited expres- siveness, issues like over-smoothing and over- squashing, and challenges in captu... | 6 | 2 | The model architectures described in the paper (GNN+, GCN, GIN, GatedGCN) are enhanced versions of classic GNNs with parameter counts estimated around 100K to 500K. Given that they are trained on 14 datasets with various sizes—each containing multiple graphs with hundreds of nodes and edges—it's reasonable to estimate ... | yes | Yes | Graph | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00.000Z | [https://github.com/LUOyk1999/GNNPlus] | 1 | https://www.dropbox.com/s/ol2v01usvaxbsr8/peptide_multi_class_dataset.csv.gz?dl=1, https://www.dropbox.com/s/j4zcnx2eipuo0xz/splits_random_stratified_peptide.pickle?dl=1 | ETA - Under 1 hour as per model desc approx 0.8 hour | https://drive.google.com/file/d/1Y7jMNhNybbdgrUJa_MxcOrbwpJNkDPav/view?usp=sharing | Yes | - Clone the repo and run requuirements, dataset is downloaded from dropbox on the train code execution. |
Tiny-ImageNet | PRO-DSC | [] | Exploring a Principled Framework for Deep Subspace Clustering | 2025-03-21T00:00:00 | https://arxiv.org/abs/2503.17288v1 | [
"https://github.com/mengxianghan123/PRO-DSC"
] | {'Accuracy': '0.698', 'NMI': '0.805'} | [
"Accuracy",
"NMI",
"ARI"
] | Given the following paper and codebase:
Paper: Exploring a Principled Framework for Deep Subspace Clustering
Codebase: https://github.com/mengxianghan123/PRO-DSC
Improve the PRO-DSC model on the Tiny-ImageNet dataset. The result
should improve on the following metrics: {'Accuracy': '0.698', 'NMI': '0.8... | Published as a conference paper at ICLR 2025 EXPLORING A PRINCIPLED FRAMEWORK FOR DEEP SUBSPACE CLUSTERING Xianghan Meng†, Zhiyuan Huang†& Wei He Beijing University of Posts and Telecommunications, Beijing 100876, P.R. China {mengxianghan,huangzhiyuan,wei.he }@bupt.edu.cn Xianbiao Qi & Rong Xiao Intellifusion, Shenzhen... | 6 | 1 | The paper describes a framework for deep subspace clustering with potentially large models considering the complexity of the tasks (e.g., high-dimensional images from datasets like CIFAR and ImageNet). Given that experiments were conducted on multiple datasets, it's reasonable to estimate that training may require inte... | yes | Yes | CV | Exploring a Principled Framework for Deep Subspace Clustering | 2025-03-21T00:00:00.000Z | [https://github.com/mengxianghan123/PRO-DSC] | 1 | Dataset found at: [https://drive.google.com/drive/folders/1C4qlqYOW4-YulIwgkNfqMM7dZ2O5-BK_], [https://drive.google.com/drive/folders/1L9jH8zRF3To6Hb_B0UZ6PbknhgusWm5_] | 20 | https://colab.research.google.com/drive/1D4PwvmROZazdEKuhZj7QkfBKOqY9Jb0r?usp=sharing | YES! SUCCESSFULLY RUN | All things fine! successfully run |
CIFAR-100 | PRO-DSC | [] | Exploring a Principled Framework for Deep Subspace Clustering | 2025-03-21T00:00:00 | https://arxiv.org/abs/2503.17288v1 | [
"https://github.com/mengxianghan123/PRO-DSC"
] | {'Accuracy': '0.773', 'NMI': '0.824'} | [
"Accuracy",
"NMI",
"ARI",
"Train Set",
"Backbone"
] | Given the following paper and codebase:
Paper: Exploring a Principled Framework for Deep Subspace Clustering
Codebase: https://github.com/mengxianghan123/PRO-DSC
Improve the PRO-DSC model on the CIFAR-100 dataset. The result
should improve on the following metrics: {'Accuracy': '0.773', 'NMI': '0.824'}... | Published as a conference paper at ICLR 2025 EXPLORING A PRINCIPLED FRAMEWORK FOR DEEP SUBSPACE CLUSTERING Xianghan Meng†, Zhiyuan Huang†& Wei He Beijing University of Posts and Telecommunications, Beijing 100876, P.R. China {mengxianghan,huangzhiyuan,wei.he }@bupt.edu.cn Xianbiao Qi & Rong Xiao Intellifusion, Shenzhen... | 6 | 1 | The proposed framework is based on deep learning techniques which typically require significant computational resources. The paper mentions extensive experiments on multiple datasets including CIFAR-10, CIFAR-20, and CIFAR-100, which are standard benchmarks in the field, often used in deep learning model training. Give... | yes | Yes | CV | Exploring a Principled Framework for Deep Subspace Clustering | 2025-03-21T00:00:00.000Z | [https://github.com/mengxianghan123/PRO-DSC] | 1 | Dataset found at: [https://drive.google.com/drive/folders/1C4qlqYOW4-YulIwgkNfqMM7dZ2O5-BK_], [https://drive.google.com/drive/folders/1L9jH8zRF3To6Hb_B0UZ6PbknhgusWm5_] | 20 | https://colab.research.google.com/drive/1D4PwvmROZazdEKuhZj7QkfBKOqY9Jb0r?usp=sharing | YES! SUCCESSFULLY RUN | All things fine! successfully run |
FB15k-237 | DaBR | [] | Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation | 2024-12-05T00:00:00 | https://arxiv.org/abs/2412.04076v2 | [
"https://github.com/llqy123/dabr"
] | {'MRR': '0.373', 'Hits@10': '0.572', 'Hits@3': '0.410', 'Hits@1': '0.247', 'MR': '83'} | [
"Hits@1",
"Hits@3",
"Hits@10",
"MRR",
"MR",
"training time (s)",
"Hit@1",
"Hit@10"
] | Given the following paper and codebase:
Paper: Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation
Codebase: https://github.com/llqy123/dabr
Improve the DaBR model on the FB15k-237 dataset. The result
should improve on the following metrics: {'MRR': '0.373', 'Hits@10': '0... | Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation Weihua Wang1,2,3, *, Qiuyu Liang1, Feilong Bao1,2,3, Guanglai Gao1,2,3 1College of Computer Science, Inner Mongolia University, Hohhot, China 2National and Local Joint Engineering Research Center of Intelligent Information Processing Tec... | 6 | 1 | The DaBR model has a unique architecture involving quaternion embeddings and bidirectional rotations, likely making it smaller than multi-layered transformer models but larger than simple embeddings. The paper does not mention exact parameter counts, but it's implied that the embedding size can vary (300-500) and relat... | yes | Yes | Graph | Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation | 2024-12-05T00:00:00.000Z | [https://github.com/llqy123/dabr] | 1 |
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
| 20 min | https://colab.research.google.com/drive/1nML0U1finrLk-EkU2gHBF3GiLpUh6rLK?usp=sharing | Yes | we fixes some issues and it runs successfully |
MM-Vet | FlashSloth-HD | [] | FlashSloth: Lightning Multimodal Large Language Models via Embedded Visual Compression | 2024-12-05T00:00:00 | https://arxiv.org/abs/2412.04317v1 | [
"https://github.com/codefanw/flashsloth"
] | {'GPT-4 score': '49.0', 'Params': '3.2B'} | [
"GPT-4 score",
"Params"
] | Given the following paper and codebase:
Paper: FlashSloth: Lightning Multimodal Large Language Models via Embedded Visual Compression
Codebase: https://github.com/codefanw/flashsloth
Improve the FlashSloth-HD model on the MM-Vet dataset. The result
should improve on the following metrics: {'GPT-4 score... | FlashSloth : Lightning Multimodal Large Language Models via Embedded Visual Compression Bo Tong1, Bokai Lai1, Yiyi Zhou1*, Gen Luo3, Yunhang Shen2, Ke Li2, Xiaoshuai Sun1, Rongrong Ji1 1Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, 361005, P.... | 6 | 2 | The FlashSloth model is based on a smaller-scale architecture with about 2-3 billion parameters (similar to other tiny MLLMs mentioned such as Qwen2-VL-2B). Given the complexity of multimodal tasks, it may require more time than training single-modality models. The dataset size is assumed to be large as it is optimized... | yes | Yes | Multimodal | FlashSloth: Lightning Multimodal Large Language Models via Embedded Visual Compression | 2024-12-05T00:00:00.000Z | [https://github.com/codefanw/flashsloth] | 2 | https://github.com/codefanw/FlashSloth/tree/main/scripts/eval | 20 min | https://colab.research.google.com/drive/1EbXpI0FmQ27nGKgRtKQCVz3m1EpBKiDY?usp=sharing | Yes | Successfully run |
CIFAR-10 | ABNet-2G-R0 | [] | ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities | 2024-11-28T00:00:00 | https://arxiv.org/abs/2411.19213v1 | [
"https://github.com/dvssajay/New_World"
] | {'Percentage correct': '94.118'} | [
"Percentage correct",
"Top-1 Accuracy",
"Accuracy",
"Parameters",
"Top 1 Accuracy",
"F1",
"Cross Entropy Loss"
] | Given the following paper and codebase:
Paper: ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities
Codebase: https://github.com/dvssajay/New_World
Improve the ABNet-2G-R0 model on the CIFAR-10 dataset. The result
should improve on the following metrics: {'Percentage correct': '9... | ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities Venkata Satya Sai Ajay Daliparthi Blekinge Institute of Technology Karlskrona, Sweden venkatasatyasaiajay.daliparthi@bth.se Abstract Inspired by Many-Worlds Interpretation (MWI), this work introduces a novel neural network architecture that spl... | 6 | 1 | The ANDHRA Bandersnatch architecture implemented with a branching factor of 2 at three levels results in 8 heads, with a total of 15 convolutional layers based on the geometric formula presented in the paper. Given that the model is intended to be used on the CIFAR-10/100 datasets, which consist of 60,000 (for CIFAR-10... | yes | Yes | CV | ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities | 2024-11-28T00:00:00.000Z | [https://github.com/dvssajay/New_World] | 1 | dataset or example for training or testing found at: [https://github.com/dvssajay/New_World] | 20 | https://colab.research.google.com/drive/16oyFcqCzN797OOwZbD6L9uZm818KurD6?usp=sharing | YES, Successfully run on | But it run on just training set complete successfully, but on testing side need to change in code or some thing is missing like model |
FB15k-237 | DaBR | [] | Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation | 2024-12-05T00:00:00 | https://arxiv.org/abs/2412.04076v2 | [
"https://github.com/llqy123/dabr"
] | {'MRR': '0.373', 'Hits@10': '0.572', 'Hits@3': '0.410', 'Hits@1': '0.247', 'MR': '83'} | [
"Hits@1",
"Hits@3",
"Hits@10",
"MRR",
"MR",
"training time (s)",
"Hit@1",
"Hit@10"
] | Given the following paper and codebase:
Paper: Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation
Codebase: https://github.com/llqy123/dabr
Improve the DaBR model on the FB15k-237 dataset. The result
should improve on the following metrics: {'MRR': '0.373', 'Hits@10': '0... | Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation Weihua Wang1,2,3, *, Qiuyu Liang1, Feilong Bao1,2,3, Guanglai Gao1,2,3 1College of Computer Science, Inner Mongolia University, Hohhot, China 2National and Local Joint Engineering Research Center of Intelligent Information Processing Tec... | 6 | 1 | The DaBR model has a unique architecture involving quaternion embeddings and bidirectional rotations, likely making it smaller than multi-layered transformer models but larger than simple embeddings. The paper does not mention exact parameter counts, but it's implied that the embedding size can vary (300-500) and relat... | yes | Yes | Graph | Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation | 2024-12-05 0:00:00 | https://github.com/llqy123/dabr | 1 | Dataset inside benchmark folder. | 4 and half days. 10000 epochs and each take 42sec. | https://drive.google.com/file/d/1XLeWvyV4sdoLDoVBzAMB6czlAbOAhB0W/view?usp=sharing | Yes | -- Straight forward clone and just run the train_FB15k-237 file. |
CIFAR-10 | ABNet-2G-R0 | [] | ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities | 2024-11-28T00:00:00 | https://arxiv.org/abs/2411.19213v1 | [
"https://github.com/dvssajay/New_World"
] | {'Percentage correct': '94.118'} | [
"Percentage correct",
"Top-1 Accuracy",
"Accuracy",
"Parameters",
"Top 1 Accuracy",
"F1",
"Cross Entropy Loss"
] | Given the following paper and codebase:
Paper: ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities
Codebase: https://github.com/dvssajay/New_World
Improve the ABNet-2G-R0 model on the CIFAR-10 dataset. The result
should improve on the following metrics: {'Percentage correct': '9... | ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities Venkata Satya Sai Ajay Daliparthi Blekinge Institute of Technology Karlskrona, Sweden venkatasatyasaiajay.daliparthi@bth.se Abstract Inspired by Many-Worlds Interpretation (MWI), this work introduces a novel neural network architecture that spl... | 6 | 1 | The ANDHRA Bandersnatch architecture implemented with a branching factor of 2 at three levels results in 8 heads, with a total of 15 convolutional layers based on the geometric formula presented in the paper. Given that the model is intended to be used on the CIFAR-10/100 datasets, which consist of 60,000 (for CIFAR-10... | yes | Yes | CV | ANDHRA Bandersnatch: Training Neural Networks to Predict Parallel Realities | 2024-11-28 0:00:00 | https://github.com/dvssajay/New_World | 1 | embedded inside the file to download CIFAR-10 | 200 epochs * 75 sec = 4.2 hours | https://drive.google.com/file/d/1RvV1o-KRUtLVHUwzpcTIPZYB6vpTVmHy/view?usp=sharing | Yes | --Run New_World/mainAB2GR0_10_1.py file.Each model has own code. |
5-Datasets | CODE-CL | [] | CODE-CL: Conceptor-Based Gradient Projection for Deep Continual Learning | 2024-11-21T00:00:00 | https://arxiv.org/abs/2411.15235v2 | [
"https://github.com/mapolinario94/CODE-CL"
] | {'Average Accuracy': '93.32', 'BWT': '-0.25'} | [
"Average Accuracy",
"BWT"
] | Given the following paper and codebase:
Paper: CODE-CL: Conceptor-Based Gradient Projection for Deep Continual Learning
Codebase: https://github.com/mapolinario94/CODE-CL
Improve the CODE-CL model on the 5-Datasets dataset. The result
should improve on the following metrics: {'Average Accuracy': '93.32... | CODE-CL: Co nceptor-Based Gradient Projection for De ep Continual L earning Marco P. E. Apolinario Sakshi Choudhary Kaushik Roy Elmore Family School of Electrical and Computer Engineering Purdue University, West Lafayete, IN 47906 mapolina@purdue.edu, choudh23@purdue.edu, kaushik@purdue.edu Abstract Continual learning ... | 6 | 1 | Considering a 5-layer AlexNet model with a typical parameter count around 5 million. The dataset CIFAR100 with 60,000 images (train + test) divided into 10 tasks suggests 6,000 images per task, trained for 200 epochs with a batch size of 64. This leads to significant computational overhead, but not excessive by modern ... | yes | Yes | CV | CODE-CL: Conceptor-Based Gradient Projection for Deep Continual Learning | 2024-11-21 0:00:00 | https://github.com/mapolinario94/CODE-CL | 1 | downloaded automatically when running script | 3.5 to 4.5 hours - Each epoch takes 15 ms and 100 epochs are there. For, 5 dataset it takes total 3.5 to 4.5hrs | https://colab.research.google.com/drive/1-kzSIjBoDKKhnP0x_UUcWJFSE3muxGCC?usp=sharing | Yes | -- Need to pass the arguments. Also dependency was installed accordingly. Everything is on google colab file. |
ISTD+ | RASM | [] | Regional Attention for Shadow Removal | 2024-11-21T00:00:00 | https://arxiv.org/abs/2411.14201v1 | [
"https://github.com/CalcuLuUus/RASM"
] | {'RMSE': '2.53'} | [
"RMSE",
"PSNR",
"SSIM",
"LPIPS"
] | Given the following paper and codebase:
Paper: Regional Attention for Shadow Removal
Codebase: https://github.com/CalcuLuUus/RASM
Improve the RASM model on the ISTD+ dataset. The result
should improve on the following metrics: {'RMSE': '2.53'}. You must use only the codebase provided.
| Regional Attention for Shadow Removal Hengxing Liu chrisliu.jz@gmail.com Tianjin University Tianjin, ChinaMingjia Li mingjiali@tju.edu.cn Tianjin University Tianjin, ChinaXiaojie Guo* xj.max.guo@gmail.com Tianjin University Tianjin, China Figure 1: (a) Performance comparison with previous SOTA methods. Our method achie... | 6 | 1 | The model proposed, RASM, has a lightweight architecture aiming for efficiency, suggesting a lower parameter count than bulkier models. Given that it adopts a U-shaped encoder-decoder architecture with a feature embedding dimension of 32 and focuses on regional attention rather than full-scale attention, I estimate app... | yes | Yes | CV | Regional Attention for Shadow Removal | 2024-11-21 0:00:00 | https://github.com/CalcuLuUus/RASM | 1 | https://drive.usercontent.google.com/download?id=1I0qw-65KBA6np8vIZzO6oeiOvcDBttAY&export=download&authuser=0 | 6min 23 sec * 1000 = 4.4 days | https://colab.research.google.com/drive/1OqVyOBRCgHGl5p0_lPBeW1xMLYVuZys7?usp=sharing | Yes | -- I have included all the path and commands in colab file. U can change the epoch to reduce the training time. |
Training and validation dataset of capsule vision 2024 challenge. | BiomedCLIP+PubmedBERT | [] | A Multimodal Approach For Endoscopic VCE Image Classification Using BiomedCLIP-PubMedBERT | 2024-10-25T00:00:00 | https://arxiv.org/abs/2410.19944v3 | [
"https://github.com/Satyajithchary/MedInfoLab_Capsule_Vision_2024_Challenge"
] | {'Total Accuracy': '97.75'} | [
"Total Accuracy"
] | Given the following paper and codebase:
Paper: A Multimodal Approach For Endoscopic VCE Image Classification Using BiomedCLIP-PubMedBERT
Codebase: https://github.com/Satyajithchary/MedInfoLab_Capsule_Vision_2024_Challenge
Improve the BiomedCLIP+PubmedBERT model on the Training and validation dataset of cap... | A MULTIMODAL APPROACH FOR ENDOSCOPIC VCE IMAGE CLASSIFICATION USING BiomedCLIP-PubMedBERT A PREPRINT Dr. Nagarajan Ganapathy∗ Department of Biomedical Engineering Indian Institute of Technology Hyderabad Sangareddy, Hyderabad, India gnagarajan@bme.iith.ac.in Podakanti Satyajith Chary Department of Biomedical Engineerin... | 6 | 1 | The BiomedCLIP model utilizes a Vision Transformer (ViT) and PubMedBERT, both known for their large parameter counts. Given the complexity of multimodal tasks (vision and language), a standard transformer model could have around 300 million parameters. Training with 37,607 frames and a batch size of 32 results in about... | yes | Yes | Multimodal | A Multimodal Approach For Endoscopic VCE Image Classification Using BiomedCLIP-PubMedBERT | 2024-10-25 0:00:00 | https://github.com/Satyajithchary/MedInfoLab_Capsule_Vision_2024_Challenge | 1 | https://github.com/misahub2023/Capsule-Vision-2024-Challenge. | 1hr * 3 epoch = 3 hour | https://colab.research.google.com/drive/19Y7kge6PwOugIf_jdkhXoxjUYkU3iqSG?usp=sharing | Yes | -- Dataset is downloaded using the script provided in github. Then need to change the path of the dataset link in the colab file.Downlaod the medinfolab-capsule-vision-2024-challenge.ipynb file from repo or just run this colab file i pasted to run the code. |
Electricity (192) | CycleNet | [] | CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns | 2024-09-27T00:00:00 | https://arxiv.org/abs/2409.18479v2 | [
"https://github.com/ACAT-SCUT/CycleNet"
] | {'MSE': '0.144', 'MAE': '0.237'} | [
"MSE",
"MAE"
] | Given the following paper and codebase:
Paper: CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns
Codebase: https://github.com/ACAT-SCUT/CycleNet
Improve the CycleNet model on the Electricity (192) dataset. The result
should improve on the following metrics: {'MSE': '0.144',... | CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns Shengsheng Lin1, Weiwei Lin1,2,∗, Xinyi Hu3, Wentai Wu4, Ruichao Mo1, Haocheng Zhong1 1School of Computer Science and Engineering, South China University of Technology, China 2Pengcheng Laboratory, China 3Department of Computer Science and E... | 6 | 1 | The CycleNet model has a minimal parameter count (around 472.9K for MLP and 123.7K for Linear variations), which suggests a relatively lightweight architecture, allowing efficient training on a single GPU. The datasets utilized are reasonably sized, with the largest being the Electricity dataset (26,304 timesteps with ... | yes | Yes | Time Series | CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns | 2024-09-27 0:00:00 | https://github.com/ACAT-SCUT/CycleNet | 1 | https://drive.usercontent.google.com/download?id=1bNbw1y8VYp-8pkRTqbjoW-TA-G8T0EQf&export=download&authuser=0 | 25s * 30 epochs = 12.5 min for each seq length. There are multiple seq length | https://drive.google.com/file/d/18IdZY2MOml8pmTVAoEcMoWWuU_1fI8aT/view?usp=sharing | Yes | -- Tested just for electricity. I have include the command on colab files. 192 seq was not available so I used 336 which was the nearest. It works just inspect run_main.sh |
PeMSD4 | PM-DMNet(R) | [] | Pattern-Matching Dynamic Memory Network for Dual-Mode Traffic Prediction | 2024-08-12T00:00:00 | https://arxiv.org/abs/2408.07100v1 | [
"https://github.com/wengwenchao123/PM-DMNet"
] | {'12 steps MAE': '18.37', '12 steps RMSE': '30.68', '12 steps MAPE': '12.01'} | [
"12 steps MAE",
"12 steps MAPE",
"12 steps RMSE"
] | Given the following paper and codebase:
Paper: Pattern-Matching Dynamic Memory Network for Dual-Mode Traffic Prediction
Codebase: https://github.com/wengwenchao123/PM-DMNet
Improve the PM-DMNet(R) model on the PeMSD4 dataset. The result
should improve on the following metrics: {'12 steps MAE': '18.37',... | 1 Pattern-Matching Dynamic Memory Network for Dual-Mode Traffic Prediction Wenchao Weng, Mei Wu, Hanyu Jiang, Wanzeng Kong, Senior Member, IEEE , Xiangjie Kong, Senior Member, IEEE , and Feng Xia, Senior Member, IEEE Abstract —In recent years, deep learning has increasingly gained attention in the field of traffic pred... | 6 | 1 | The PM-DMNet model employs a dynamic memory network with reduced computational complexity of O(N) compared to existing methods. Given the complexity of the architecture and typical dataset sizes in traffic prediction tasks, an estimated training time of 6 hours is reasonable assuming a moderate dataset size of around 1... | yes | Yes | Graph | Pattern-Matching Dynamic Memory Network for Dual-Mode Traffic Prediction | 2024-08-12 0:00:00 | https://github.com/wengwenchao123/PM-DMNet | 1 | https://drive.usercontent.google.com/download?id=1Q8boyeVNmZTz_HASN_57qd9wX1JZeGem&export=download&authuser=0 | 35 s avg * 500 epochs = 5 hour approx | https://colab.research.google.com/drive/1MGEsXeIEGO7AKBMZ6DBZoEt73bQaCTe2?usp=sharing | Yes | -- Fairly easy one. I have included the pip isntallation on collab file. This repo doesnot contain requirements.txt file. |
Kvasir-SEG | EffiSegNet-B5 | [] | EffiSegNet: Gastrointestinal Polyp Segmentation through a Pre-Trained EfficientNet-based Network with a Simplified Decoder | 2024-07-23T00:00:00 | https://arxiv.org/abs/2407.16298v1 | [
"https://github.com/ivezakis/effisegnet"
] | {'mean Dice': '0.9488', 'mIoU': '0.9065', 'F-measure': '0.9513', 'Precision': '0.9713', 'Recall': '0.9321'} | [
"mean Dice",
"Average MAE",
"S-Measure",
"max E-Measure",
"mIoU",
"FPS",
"F-measure",
"Precision",
"Recall"
] | Given the following paper and codebase:
Paper: EffiSegNet: Gastrointestinal Polyp Segmentation through a Pre-Trained EfficientNet-based Network with a Simplified Decoder
Codebase: https://github.com/ivezakis/effisegnet
Improve the EffiSegNet-B5 model on the Kvasir-SEG dataset. The result
should improve... | EffiSegNet: Gastrointestinal Polyp Segmentation through a Pre-Trained EfficientNet-based Network with a Simplified Decoder Ioannis A. Vezakis TECREANDO B.V . Amsterdam, The Netherlands 0000-0003-4976-4901Konstantinos Georgas Biomedical Engineering Laboratory School of Electrical and Computer Engineering National Techni... | 6 | 2 | The EffiSegNet model has multiple variants with EfficientNet as backbone, ranging from 4.0M to 63.8M parameters. Given the Kvasir-SEG dataset of 1000 images and training for 300 epochs with a batch size of 8, the number of parameters suggests a training time of around 6 hours using 2 GPUs. Each image requires resizing ... | yes | Yes | CV | EffiSegNet: Gastrointestinal Polyp Segmentation through a Pre-Trained EfficientNet-based Network with a Simplified Decoder | 2024-07-23 0:00:00 | https://github.com/ivezakis/effisegnet | 2 | Inside the repo | 45 sec * 300 epoch = 4 hour around | https://colab.research.google.com/drive/1YzKf-VnfFVZW67_SYj2295KmuwYAFgUB?usp=sharing | Yes | Fairly easy just create env and run |
clintox | BiLSTM | [] | Accelerating Drug Safety Assessment using Bidirectional-LSTM for SMILES Data | 2024-07-08T00:00:00 | https://arxiv.org/abs/2407.18919v1 | [
"https://github.com/kvrsid/toxic"
] | {'AUC': '0.97'} | [
"AUC"
] | Given the following paper and codebase:
Paper: Accelerating Drug Safety Assessment using Bidirectional-LSTM for SMILES Data
Codebase: https://github.com/kvrsid/toxic
Improve the BiLSTM model on the clintox dataset. The result
should improve on the following metrics: {'AUC': '0.97'}. You must use only t... | 393 Vol. 21, No. 1 , (2024) ISSN: 1005 -0930 Accelerating Drug Safety Assessment using Bidirectional -LSTM for SMILES Data K. Venkateswara Rao1, Dr. Kunjam Nageswara Rao2, Dr. G. Sita Ratnam3 1 Research Scholar, 2 Professor Department of Computer Science and Systems Engineering, Andhra University College of Engineering... | 6 | 1 | The proposed model employs a bidirectional LSTM architecture, which typically has a reasonable number of parameters compared to other complex models like transformers or very deep networks. Given the structure described, a rough estimate of 6 hours of training time on a standard single GPU is appropriate, taking into a... | yes | Yes | Bioinformatics | Accelerating Drug Safety Assessment using Bidirectional-LSTM for SMILES Data | 2024-07-08 0:00:00 | https://github.com/kvrsid/toxic | 1 | inside the repo as clintox.csv | Total 5 min on 100 epochs. | https://drive.google.com/file/d/1ut_cYbQzf3Pov5Xdu24TxA5WEEMucV-z/view?usp=sharing | Yes | -- I fixed 2 lines on code. I have commented on the colab file. |
ImageNet-10 | DPAC | [] | Deep Online Probability Aggregation Clustering | 2024-07-07T00:00:00 | https://arxiv.org/abs/2407.05246v2 | [
"https://github.com/aomandechenai/deep-probability-aggregation-clustering"
] | {'Accuracy': '0.97', 'NMI': '0.925', 'ARI': '0.935', 'Backbone': 'ResNet-34'} | [
"NMI",
"Accuracy",
"ARI",
"Backbone",
"Image Size"
] | Given the following paper and codebase:
Paper: Deep Online Probability Aggregation Clustering
Codebase: https://github.com/aomandechenai/deep-probability-aggregation-clustering
Improve the DPAC model on the ImageNet-10 dataset. The result
should improve on the following metrics: {'Accuracy': '0.97', 'N... | Deep Online Probability Aggregation Clustering Yuxuan Yan, Na Lu⋆, and Ruofan Yan Systems Engineering Institute, Xi’an Jiaotong University yan1611@stu.xjtu.edu.cn, lvna2009@mail.xjtu.edu.cn, yanruofan@stu.xjtu.edu.cn Abstract. Combining machine clustering with deep models has shown remarkable superiority in deep cluste... | 6 | 1 | The DPAC model uses a deep learning framework that incorporates modern neural network architectures suitable for deep clustering tasks, similar to models typical in the domain (like SimCLR). The datasets used (CIFAR-10, CIFAR-100, etc.) are of moderate size, and the paper mentions various datasets with up to 60,000 sam... | yes | Yes | CV | Deep Online Probability Aggregation Clustering | 2024-07-07 0:00:00 | https://github.com/aomandechenai/deep-probability-aggregation-clustering | 1 | downloads ciraf10 dataset from pre process step | 36 hour just for pre train. | https://drive.google.com/file/d/1-nXU0RbPPY9WObax53y0CrfOoQ-6cry4/view?usp=sharing | Yes | -- Need to change some line on pre_train,py. The changed code is there in colab in comment. Takes 36 hour just for 1 epoch as shown while training. May run with enough resources. |
CAT2000 | SUM | [] | SUM: Saliency Unification through Mamba for Visual Attention Modeling | 2024-06-25T00:00:00 | https://arxiv.org/abs/2406.17815v2 | [
"https://github.com/Arhosseini77/SUM"
] | {'KL': '0.27'} | [
"KL"
] | Given the following paper and codebase:
Paper: SUM: Saliency Unification through Mamba for Visual Attention Modeling
Codebase: https://github.com/Arhosseini77/SUM
Improve the SUM model on the CAT2000 dataset. The result
should improve on the following metrics: {'KL': '0.27'}. You must use only the code... | SUM: Saliency Unification through Mamba for Visual Attention Modeling Alireza Hosseini*,1Amirhossein Kazerouni*,2,3,4Saeed Akhavan1 Michael Brudno2,3,4Babak Taati2,3,4 1University of Tehran2University of Toronto3Vector Institute 4University Health Network {arhosseini77, s.akhavan }@ut.ac.ir, {amirhossein, brudno }@cs.t... | 6 | 1 | The SUM model utilizes a U-Net architecture integrated with Mamba, which is known for efficiency due to its linear complexity. While the total parameter count isn't specified, similar models in this domain typically range from 30M to 800M. Given the complexity, a reasonable estimate for training time is 6 hours based o... | yes | Yes | CV | SUM: Saliency Unification through Mamba for Visual Attention Modeling | 2024-06-25 0:00:00 | https://github.com/Arhosseini77/SUM | 1 | https://drive.usercontent.google.com/download?id=1Mdk97UB0phYDZv8zgjBayeC1I1_QcUmh&export=download&authuser=0 | 4 min * 30 epoch = 2 hr | https://colab.research.google.com/drive/1jdVKL-KYdo1CgCdOCzzKMFSmDBqrc8RX?usp=sharing | Yes | -- Dont run requirements.txt as it will produce dependency error. I have included the pip install command and also for running on CAT2000, need to make small changes which I have included on colab file. . Also need to set small command for matplotlib to run on collab. |
SumMe | CSTA | [] | CSTA: CNN-based Spatiotemporal Attention for Video Summarization | 2024-05-20T00:00:00 | https://arxiv.org/abs/2405.11905v2 | [
"https://github.com/thswodnjs3/CSTA"
] | {"Kendall's Tau": '0.246', "Spearman's Rho": '0.274'} | [
"F1-score (Canonical)",
"F1-score (Augmented)",
"Kendall's Tau",
"Spearman's Rho"
] | Given the following paper and codebase:
Paper: CSTA: CNN-based Spatiotemporal Attention for Video Summarization
Codebase: https://github.com/thswodnjs3/CSTA
Improve the CSTA model on the SumMe dataset. The result
should improve on the following metrics: {"Kendall's Tau": '0.246', "Spearman's Rho": '0.2... | CSTA: CNN-based Spatiotemporal Attention for Video Summarization Jaewon Son, Jaehun Park, Kwangsu Kim* Sungkyunkwan University {31z522x4,pk9403,kim.kwangsu }@skku.edu Abstract Video summarization aims to generate a concise repre- sentation of a video, capturing its essential content and key moments while reducing its o... | 6 | 1 | The model, CSTA, is based on a CNN architecture (GoogleNet) and integrates attention mechanisms. Given that model architectures like this usually have around 5-10 million parameters, I estimate approximately 6 hours of training time assuming a dataset with 50 videos (TVSum) and around 25 videos (SumMe), which is standa... | yes | Yes | CV | CSTA: CNN-based Spatiotemporal Attention for Video Summarization | 2024-05-20 0:00:00 | https://github.com/thswodnjs3/CSTA | 1 | https://github.com/e-apostolidis/PGL-SUM/tree/master/data | 5 MIN FOR SUMme dataset for 50 epochs | https://colab.research.google.com/drive/1zMK8TRHtdhQB7dkwkxA3ImblIstiU9ob?usp=sharing | Yes | -- Run perfectly on SUMme. But crashes on TVSum dataset saying out of GPU memory. |
HME100K | ICAL | [] | ICAL: Implicit Character-Aided Learning for Enhanced Handwritten Mathematical Expression Recognition | 2024-05-15T00:00:00 | https://arxiv.org/abs/2405.09032v4 | [
"https://github.com/qingzhenduyu/ical"
] | {'ExpRate': '69.06'} | [
"ExpRate"
] | Given the following paper and codebase:
Paper: ICAL: Implicit Character-Aided Learning for Enhanced Handwritten Mathematical Expression Recognition
Codebase: https://github.com/qingzhenduyu/ical
Improve the ICAL model on the HME100K dataset. The result
should improve on the following metrics: {'ExpRate... | ICAL: Implicit Character-Aided Learning for Enhanced Handwritten Mathematical Expression Recognition Jianhua Zhu1[0009 −0000−3982−2739], Liangcai Gao1( ), and Wenqi Zhao1 Wangxuan Institute of Computer Technology, Peking University, Beijing, China zhujianhuapku@pku.edu.cn gaoliangcai@pku.edu.cn wenqizhao@stu.pku.edu.cn... | 6 | 2 | The model uses a DenseNet encoder with multiple layers and a Transformer decoder, which suggests a moderate to high complexity. Given that DenseNet and Transformers are known to have significant memory and computational demands, along with the dataset sizes (8,836 training samples for CROHME with about 300,000 characte... | yes | Yes | CV | ICAL: Implicit Character-Aided Learning for Enhanced Handwritten Mathematical Expression Recognition | 2024-05-15 0:00:00 | https://github.com/qingzhenduyu/ical | 2 | https://disk.pku.edu.cn/anyshare/en-us/link/AAF10CCC4D539543F68847A9010C607139/EF71051AA2314E3AA921F528C70BF712/A2D37D1699B54529BA80157162294FA5?_tb=none | 1HR per epoch * 120 epoch = 120 hour | https://colab.research.google.com/drive/1ojkqF09KgeqtsgyPSDS0ya64VPddIgiz?usp=sharing | Yes | -- Cannot download the data directly into colab. Need to store in local and upload to the colab or use google drive to unzip the content to colab |
Kvasir-SEG | EMCAD | [] | EMCAD: Efficient Multi-scale Convolutional Attention Decoding for Medical Image Segmentation | 2024-05-11T00:00:00 | https://arxiv.org/abs/2405.06880v1 | [
"https://github.com/sldgroup/emcad"
] | {'mean Dice': '0.928'} | [
"mean Dice",
"Average MAE",
"S-Measure",
"max E-Measure",
"mIoU",
"FPS",
"F-measure",
"Precision",
"Recall"
] | Given the following paper and codebase:
Paper: EMCAD: Efficient Multi-scale Convolutional Attention Decoding for Medical Image Segmentation
Codebase: https://github.com/sldgroup/emcad
Improve the EMCAD model on the Kvasir-SEG dataset. The result
should improve on the following metrics: {'mean Dice': '0... | EMCAD: Efficient Multi-scale Convolutional Attention Decoding for Medical Image Segmentation Md Mostafijur Rahman, Mustafa Munir, and Radu Marculescu The University of Texas at Austin Austin, Texas, USA mostafijur.rahman, mmunir, radum@utexas.edu Abstract An efficient and effective decoding mechanism is crucial in medi... | 6 | 1 | The EMCAD model has approximately 1.91 million parameters and 0.381G FLOPs for a standard encoder, making it relatively lightweight compared to larger models like UNet and TransUNet, which have tens of millions of parameters and significantly higher FLOP counts. Given the medical image segmentation task, a typical trai... | yes | Yes | CV | EMCAD: Efficient Multi-scale Convolutional Attention Decoding for Medical Image Segmentation | 2024-05-11 0:00:00 | https://github.com/sldgroup/emcad | 1 | https://drive.google.com/drive/folders/1ACJEoTp-uqfFJ73qS3eUObQh52nGuzCd | 21 hour for SYNAPSE dataset. | https://colab.research.google.com/drive/1jYDic29ht3AjFGx5rXY_Fp5hcoxoRP6M?usp=sharing | Yes | -- No definiton on how to run on Kvsair - SEG datas. Need to create separate dataloader as input for Kavirseg data. But it ran on Synapse dataset provided as their default training set. |
ETTh1 (336) Multivariate | SOFTS | [] | SOFTS: Efficient Multivariate Time Series Forecasting with Series-Core Fusion | 2024-04-22T00:00:00 | https://arxiv.org/abs/2404.14197v3 | [
"https://github.com/secilia-cxy/softs"
] | {'MSE': '0.480', 'MAE': '0.452'} | [
"MSE",
"MAE"
] | Given the following paper and codebase:
Paper: SOFTS: Efficient Multivariate Time Series Forecasting with Series-Core Fusion
Codebase: https://github.com/secilia-cxy/softs
Improve the SOFTS model on the ETTh1 (336) Multivariate dataset. The result
should improve on the following metrics: {'MSE': '0.480... | SOFTS: Efficient Multivariate Time Series Forecasting with Series-Core Fusion Lu Han∗, Xu-Yang Chen∗, Han-Jia Ye†, De-Chuan Zhan School of Artificial Intelligence, Nanjing University, China National Key Laboratory for Novel Software Technology, Nanjing University, China {hanlu, chenxy, yehj, zhandc}@lamda.nju.edu.cn Ab... | 6 | 1 | The SOFTS architecture is based on an MLP and is designed to have linear complexity in terms of the number of channels, while the datasets reported have around 170 to 883 channels. A reasonable estimate for the number of parameters in this model, given the MLP structure, is in the order of 1-5 million parameters. Consi... | yes | Yes | Time Series | SOFTS: Efficient Multivariate Time Series Forecasting with Series-Core Fusion | 2024-04-22 0:00:00 | https://github.com/secilia-cxy/softs | 1 | https://drive.google.com/drive/folders/1ZOYpTUa82_jCcxIdTmyr0LXQfvaM9vIy | 30s for 336 seq | https://colab.research.google.com/drive/14p_kyKxFS9780yR-GJpq4foiumZUznJ4?usp=sharing | Yes | -- I have listed the requirement to install on collab cell. Just need to comment some line to run for only 336 seq. |
FER2013 | VGG based | [] | IdentiFace : A VGG Based Multimodal Facial Biometric System | 2024-01-02T00:00:00 | https://arxiv.org/abs/2401.01227v2 | [
"https://github.com/MahmoudRabea13/IdentiFace"
] | {'5-class test accuracy': '66.13%'} | [
"5-class test accuracy"
] | Given the following paper and codebase:
Paper: IdentiFace : A VGG Based Multimodal Facial Biometric System
Codebase: https://github.com/MahmoudRabea13/IdentiFace
Improve the VGG based model on the FER2013 dataset. The result
should improve on the following metrics: {'5-class test accuracy': '66.13%'}. ... | IdentiFace: A VGGNet -Based Multimodal Facial Biometric System Mahmoud Rabea, Hanya Ahmed, Sohaila Mahmoud, Nourhan Sayed Systems and Biomedical Department, Faculty of Engineering Cairo University Abstract - The development of facial biometric systems has contributed greatly to the development of the computer vision fi... | 6 | 1 | The model described is based on a simplified VGG-16 architecture with a lower number of layers and parameters compared to the original model. Given that this architecture has several layers and parameters, I estimate around 6 hours of training time based on the size of the datasets involved and the computational cost o... | yes | Yes | CV | IdentiFace : A VGG Based Multimodal Facial Biometric System | 2024-01-02 0:00:00 | https://github.com/MahmoudRabea13/IdentiFace | 1 | https://www.kaggle.com/datasets/msambare/fer2013 | 30s * 40 epoch = 20 min | https://drive.google.com/file/d/1NLLV2fLLpzBI3IQlCa6xac_SSr6q7ofN/view?usp=sharing | Yes | -- The training code is included in /Notebooks/Emotion/FER Dataset/Model.ipynb and inside the repo. I have linked the repo with proper fixes. Just run the colab file i have linked here. |
ETTh1 (336) Multivariate | AMD | [] | Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting | 2024-06-06T00:00:00 | https://arxiv.org/abs/2406.03751v1 | [
"https://github.com/troubadour000/amd"
] | {'MSE': '0.418', 'MAE': '0.427'} | [
"MSE",
"MAE"
] | Given the following paper and codebase:
Paper: Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting
Codebase: https://github.com/troubadour000/amd
Improve the AMD model on the ETTh1 (336) Multivariate dataset. The result
should improve on the following metrics: {'MSE': '0.418', 'MAE... | Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting Yifan Hu1,∗Peiyuan Liu3,∗Peng Zhu1Dawei Cheng1,BTao Dai2 1Tongji University2Shenzhen University 3Tsinghua Shenzhen International Graduate School {pengzhu, dcheng}@tongji.edu.cn {huyf0122, peiyuanliu.edu, daitao.edu}@gmail.com Abstract Transformer-... | 6 | 1 | The model proposed in the paper is an MLP-based Adaptive Multi-Scale Decomposition (AMD) framework, which likely has fewer parameters than Transformer-based models. The paper indicates a memory usage of 1349 MB, implying a moderate model size appropriate for single-GPU training. The training time of 17 ms/iteration sug... | yes | Yes | Time Series | Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting | 2024-06-06 0:00:00 | https://github.com/troubadour000/amd | 1 | Autoformer | 1 | Untitled12.ipynb | Yes | Working as expected with manually downloading and uploading data |
ZINC | NeuralWalker | [] | Learning Long Range Dependencies on Graphs via Random Walks | 2024-06-05T00:00:00 | https://arxiv.org/abs/2406.03386v2 | [
"https://github.com/borgwardtlab/neuralwalker"
] | {'MAE': '0.065 ± 0.001'} | [
"MAE"
] | Given the following paper and codebase:
Paper: Learning Long Range Dependencies on Graphs via Random Walks
Codebase: https://github.com/borgwardtlab/neuralwalker
Improve the NeuralWalker model on the ZINC dataset. The result
should improve on the following metrics: {'MAE': '0.065 ± 0.001'}. You must us... | Learning Long Range Dependencies on Graphs via Random Walks Dexiong Chen Till Hendrik Schulz Karsten Borgwardt Max Planck Institute of Biochemistry 82152 Martinsried, Germany {dchen, tschulz, borgwardt}@biochem.mpg.de Abstract Message-passing graph neural networks (GNNs) excel at capturing local relation- ships but str... | 6 | 1 | The proposed NeuralWalker architecture leverages random walks and message-passing mechanisms, which typically have a moderate level of complexity. It combines local and long-range dependencies, making it a compact yet powerful model. Given the extensive graphs and nodes it claims to handle (up to 1.6M nodes), one can e... | yes | Yes | Graph | Learning Long Range Dependencies on Graphs via Random Walks | 2024-06-05 0:00:00 | https://github.com/borgwardtlab/neuralwalker | 1 | In Code | 1 | Untitled13.ipynb | Yes | Works with installing micromamba first |
MalNet-Tiny | GatedGCN+ | [] | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00 | https://arxiv.org/abs/2502.09263v1 | [
"https://github.com/LUOyk1999/GNNPlus"
] | {'Accuracy': '94.600±0.570'} | [
"Accuracy",
"MCC"
] | Given the following paper and codebase:
Paper: Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence
Codebase: https://github.com/LUOyk1999/GNNPlus
Improve the GatedGCN+ model on the MalNet-Tiny dataset. The result
should improve on the following metrics: {... | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence Yuankai Luo1 2Lei Shi*1Xiao-Ming Wu*2 Abstract Message-passing Graph Neural Networks (GNNs) are often criticized for their limited expres- siveness, issues like over-smoothing and over- squashing, and challenges in captu... | 8 | 1 | The training process involves using 3 classic GNN architectures (GCN, GIN, GatedGCN) enhanced by the GNN+ framework. Given that each model has around 500K parameters and will be trained on 14 datasets with different sizes (ZINC has 12K graphs while ogbg-code2 has over 450K). The average time per epoch as reported is lo... | yes | Yes | Graph | Unlocking the Potential of Classic GNNs for Graph-level Tasks: Simple Architectures Meet Excellence | 2025-02-13T00:00:00.000Z | [https://github.com/LUOyk1999/GNNPlus] | 1 | http://malnet.cc.gatech.edu/graph-data/malnet-graphs-tiny.tar.gz,http://malnet.cc.gatech.edu/split-info/split_info_tiny.zip | 1 hour - (avg 24 sec * 150 epochs) | https://drive.google.com/file/d/1Y7jMNhNybbdgrUJa_MxcOrbwpJNkDPav/view?usp=sharing | Yes | null |
STL-10, 40 Labels | SemiOccam | [] | ViTSGMM: A Robust Semi-Supervised Image Recognition Network Using Sparse Labels | 2025-06-04T00:00:00 | https://arxiv.org/abs/2506.03582v1 | [
"https://github.com/Shu1L0n9/SemiOccam"
] | {'Accuracy': '95.43'} | [
"Accuracy"
] | Given the following paper and codebase:
Paper: ViTSGMM: A Robust Semi-Supervised Image Recognition Network Using Sparse Labels
Codebase: https://github.com/Shu1L0n9/SemiOccam
Improve the SemiOccam model on the STL-10, 40 Labels dataset. The result
should improve on the following metrics: {'Accuracy': '... | Rui et al. Harbin Engineering University VITSGMM: A R OBUST SEMI-SUPERVISED IMAGE RECOGNITION NETWORK USING SPARSE LABELS Rui Yann∗ Shu1L0n9@gmail.comXianglei Xing† xingxl@hrbeu.edu.cn General Artificial Intelligence Laboratory College of Intelligent Systems Science and Engineering Harbin Engineering University Harbin,... | 8 | 1 | The ViTSGMM model utilizes the Vision Transformer architecture (likely ViT-base or ViT-large), which has approximately 86 million parameters for ViT-base and over 300 million for ViT-large. Considering the CIFAR-10 dataset has 60,000 images and STL-10 has about 13,000 images, the added computational complexity from sem... | yes | Yes | CV | ViTSGMM: A Robust Semi-Supervised Image Recognition Network Using Sparse Labels | 2025-06-04T00:00:00.000Z | [https://github.com/Shu1L0n9/SemiOccam] | 1 | Code Downloads Dynamically after cahnging the dataset name | 3 Hours | Copy of experiment.ipynb | Yes | It starts and runs successfully |
CIFAR-10 | ResNet18 (FSGDM) | [] | On the Performance Analysis of Momentum Method: A Frequency Domain Perspective | 2024-11-29T00:00:00 | https://arxiv.org/abs/2411.19671v6 | [
"https://github.com/yinleung/FSGDM"
] | {'Percentage correct': '95.66'} | [
"Percentage correct",
"Top-1 Accuracy",
"Accuracy",
"Parameters",
"Top 1 Accuracy",
"F1",
"Cross Entropy Loss"
] | Given the following paper and codebase:
Paper: On the Performance Analysis of Momentum Method: A Frequency Domain Perspective
Codebase: https://github.com/yinleung/FSGDM
Improve the ResNet18 (FSGDM) model on the CIFAR-10 dataset. The result
should improve on the following metrics: {'Percentage correct'... | Published as a conference paper at ICLR 2025 ON THE PERFORMANCE ANALYSIS OF MOMENTUM METHOD : A F REQUENCY DOMAIN PERSPECTIVE Xianliang Li∗1,2, Jun Luo∗1,2, Zhiwei Zheng∗3, Hanxiao Wang2,4, Li Luo5, Lingkun Wen2,6, Linlong Wu7, Sheng Xu†1 1Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences 2Univers... | 8 | 1 | The paper involves training two models: ResNet50 on CIFAR-100 and VGG16 on CIFAR-10. ResNet50 has approximately 25.6 million parameters, while VGG16 has around 138 million parameters. Both datasets (CIFAR-10 and CIFAR-100) are relatively small (60,000 images total for CIFAR-10 and 100,000 images for CIFAR-100) and are ... | yes | Yes | CV | On the Performance Analysis of Momentum Method: A Frequency Domain Perspective | 2024-11-29T00:00:00.000Z | [https://github.com/yinleung/FSGDM] | 1 | dataset or example for train found at: [https://github.com/yinleung/FSGDM/tree/main/examples/CIFAR100] | 10 | https://colab.research.google.com/drive/1rYHru1icUH3Yj4kvEvVuriMhdqM--kCS?usp=sharing | YES, Successfully Run! | But need to chnage in little bit code and optimize it. And for training on a example need too much time. |
CIFAR-10 | ResNet18 (FSGDM) | [] | On the Performance Analysis of Momentum Method: A Frequency Domain Perspective | 2024-11-29T00:00:00 | https://arxiv.org/abs/2411.19671v6 | [
"https://github.com/yinleung/FSGDM"
] | {'Percentage correct': '95.66'} | [
"Percentage correct",
"Top-1 Accuracy",
"Accuracy",
"Parameters",
"Top 1 Accuracy",
"F1",
"Cross Entropy Loss"
] | Given the following paper and codebase:
Paper: On the Performance Analysis of Momentum Method: A Frequency Domain Perspective
Codebase: https://github.com/yinleung/FSGDM
Improve the ResNet18 (FSGDM) model on the CIFAR-10 dataset. The result
should improve on the following metrics: {'Percentage correct'... | Published as a conference paper at ICLR 2025 ON THE PERFORMANCE ANALYSIS OF MOMENTUM METHOD : A F REQUENCY DOMAIN PERSPECTIVE Xianliang Li∗1,2, Jun Luo∗1,2, Zhiwei Zheng∗3, Hanxiao Wang2,4, Li Luo5, Lingkun Wen2,6, Linlong Wu7, Sheng Xu†1 1Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences 2Univers... | 8 | 1 | The paper involves training two models: ResNet50 on CIFAR-100 and VGG16 on CIFAR-10. ResNet50 has approximately 25.6 million parameters, while VGG16 has around 138 million parameters. Both datasets (CIFAR-10 and CIFAR-100) are relatively small (60,000 images total for CIFAR-10 and 100,000 images for CIFAR-100) and are ... | yes | Yes | CV | On the Performance Analysis of Momentum Method: A Frequency Domain Perspective | 2024-11-29 0:00:00 | https://github.com/yinleung/FSGDM | 1 | inside the repo examples CIFAR1OO Folder | 300 epochs * 2.5 min = 12.5 hOURS | https://drive.google.com/file/d/1grWsTDyc3MOwfbwob2EbMbL7GPmjsKfI/view?usp=sharing | Yes | -- Run by going inside the examples/cifar1oo/main.py |
ogbl-ddi | GCN (node embedding) | [] | Can GNNs Learn Link Heuristics? A Concise Review and Evaluation of Link Prediction Methods | 2024-11-22T00:00:00 | https://arxiv.org/abs/2411.14711v1 | [
"https://github.com/astroming/GNNHE"
] | {'Test Hits@20': '0.9549 ± 0.0073', 'Validation Hits@20': '0.9098 ± 0.0294', 'Number of params': '5125250', 'Ext. data': 'No'} | [
"Ext. data",
"Test Hits@20",
"Validation Hits@20",
"Number of params"
] | Given the following paper and codebase:
Paper: Can GNNs Learn Link Heuristics? A Concise Review and Evaluation of Link Prediction Methods
Codebase: https://github.com/astroming/GNNHE
Improve the GCN (node embedding) model on the ogbl-ddi dataset. The result
should improve on the following metrics: {'Te... | Can GNNs Learn Link Heuristics? A Concise Review and Evaluation of Link Prediction Methods Shuming Liang*1, Yu Ding2, Zhidong Li1, Bin Liang1, Siqi Zhang3, Yang Wang1, Fang Chen1 1University of Technology Sydney first name.last name@uts.edu.au 2University of Wollongong dyu@uow.edu.au 3Zhejiang University siqizhang@zju.... | 8 | 1 | The experiments utilize OGB datasets which are well-known benchmarks in link prediction tasks. Based on similar models in the literature, training one of these GNNs on datasets like ogbl-collab or ogbl-ddi typically ranges from 4 to 8 hours on a single GPU when standard hyperparameters are used. Given the dataset sizes... | yes | Yes | Graph | Can GNNs Learn Link Heuristics? A Concise Review and Evaluation of Link Prediction Methods | 2024-11-22 0:00:00 | https://github.com/astroming/GNNHE | 1 | Inside the /GNNHE/ogbl-ddi_95.49_10runs/dataset folder of repo | 30 sec * 2000 = 16.7hours | https://colab.research.google.com/drive/1LD0gm45pSoZMyKFWrm23d4s0_DQpgtx8?usp=sharing | Yes | -- Since the requirements were vauge .. I used grok to fix the dependency issue and the installment process is recorded in collab notebook. |
TXL-PBC: a freely accessible labeled peripheral blood cell dataset | yolov5n | [] | TXL-PBC: a freely accessible labeled peripheral blood cell dataset | 2024-07-18T00:00:00 | https://arxiv.org/abs/2407.13214v1 | [
"https://github.com/lugan113/TXL-PBC_Dataset"
] | {'mAP50': '0.958'} | [
"mAP50"
] | Given the following paper and codebase:
Paper: TXL-PBC: a freely accessible labeled peripheral blood cell dataset
Codebase: https://github.com/lugan113/TXL-PBC_Dataset
Improve the yolov5n model on the TXL-PBC: a freely accessible labeled peripheral blood cell dataset dataset. The result
should improve ... | TXL-PBC: A FREELY ACCESSIBLE LABELED PERIPHERAL BLOOD CELL DATASET Lu Gan Northern Arizona University Flagstaff, AZ, USA lg2465@nau.eduXi Li Independent Researcher Chengdu, China reilixi723@gmail.com ABSTRACT In a recent study, we found that publicly BCCD and BCD datasets have significant issues such as labeling errors... | 8 | 1 | The TXL-PBC dataset has 1,008 training samples, with a batch size of 16 and an image resolution of 320x320 pixels. Training with YOLOv8n for 100 epochs means that there are a total of 1,008/16 = 63 iterations per epoch, resulting in approximately 6,300 total iterations. Given the complexity of YOLOv8n and the semi-auto... | yes | Yes | CV | TXL-PBC: a freely accessible labeled peripheral blood cell dataset | 2024-07-18 0:00:00 | https://github.com/lugan113/TXL-PBC_Dataset | 1 | isnside the repo on TXL-PBC folder | 17s * 100 epoch = 29 minutes aprox | https://drive.google.com/file/d/1NdhlcOZdyojbL8kctTFOo8eFA03PMWdl/view?usp=sharing | Yes | -- I have fixed the train.py file with correct arguments and file path. I have commented the fixes on collab file. |
ZJU-RGB-P | CSFNet-2 | [] | CSFNet: A Cosine Similarity Fusion Network for Real-Time RGB-X Semantic Segmentation of Driving Scenes | 2024-07-01T00:00:00 | https://arxiv.org/abs/2407.01328v1 | [
"https://github.com/Danial-Qashqai/CSFNet"
] | {'mIoU': '91.40', 'Frame (fps)': '75 (3090)'} | [
"mIoU",
"Frame (fps)"
] | Given the following paper and codebase:
Paper: CSFNet: A Cosine Similarity Fusion Network for Real-Time RGB-X Semantic Segmentation of Driving Scenes
Codebase: https://github.com/Danial-Qashqai/CSFNet
Improve the CSFNet-2 model on the ZJU-RGB-P dataset. The result
should improve on the following metric... | CSFNet: A Cosine Similarity Fusion Network for Real-Time RGB -X Semantic Segmentation of Driving Scenes Danial Qashqaia,*, Emad Mousaviana, Shahriar B. Shokouhia, Sattar Mirzakuchakia aDepartment of Electrical Engineering, Iran University of Science and Technology, Tehran, Iran Abstract Semantic segmentation, as a cruc... | 8 | 1 | The CSFNet model described in the paper is utilizing a dual and single-branch architecture with low complexity, specifically designed for faster inference. Given that it is trained on Cityscapes, MFNet, and ZJU datasets with a moderate number of parameters (around 11.31M to 19.37M), and considering the training setting... | yes | Yes | CV | CSFNet: A Cosine Similarity Fusion Network for Real-Time RGB-X Semantic Segmentation of Driving Scenes | 2024-07-01 0:00:00 | https://github.com/Danial-Qashqai/CSFNet | 1 | https://drive.google.com/file/d/1TugQ16fcxbmPBJD0EPMHHmjdK9IE4SAO/view | 34s * 600 epoch = 5.67 hour | https://drive.google.com/file/d/12nPSCuyG-9-eA3bAaqDBWAzJGoddQyDn/view?usp=sharing | Yes | -- Just need to change the path on the argument while calling training script. All the data in proper structure is in colab file. Also the backbone has been downloaded and add resp. |
MIMIC-III | FLD | [] | Functional Latent Dynamics for Irregularly Sampled Time Series Forecasting | 2024-05-06T00:00:00 | https://arxiv.org/abs/2405.03582v2 | [
"https://github.com/kloetergensc/functional-latent_dynamics"
] | {'MSE': '0.444 ± 0.027'} | [
"MSE",
"NegLL"
] | Given the following paper and codebase:
Paper: Functional Latent Dynamics for Irregularly Sampled Time Series Forecasting
Codebase: https://github.com/kloetergensc/functional-latent_dynamics
Improve the FLD model on the MIMIC-III dataset. The result
should improve on the following metrics: {'MSE': '0.4... | Functional Latent Dynamics for Irregularly Sampled Time Series Forecasting Christian Kl¨ otergens( )12, Vijaya Krishna Yalavarthi1, Maximilian Stubbemann12, and Lars Schmidt-Thieme12 1ISMLL, University of Hildesheim, Germany {kloetergens, yalavarthi, stubbemann, schmidt-thieme }@ismll.de 2VWFS Data Analytics Research C... | 8 | 1 | The Functional Latent Dynamics (FLD) model employs a multi-head attention mechanism and a feedforward neural network for its architecture, which entails a moderate level of complexity. Considering the datasets used (which have varying sample sizes and sparsity), realistic estimates suggest that with diverse datasets li... | yes | Yes | Time Series | Functional Latent Dynamics for Irregularly Sampled Time Series Forecasting | 2024-05-06 0:00:00 | https://github.com/kloetergensc/functional-latent_dynamics | 1 | https://physionet.org/content/mimiciii/1.4/, Goodwin dataset inside the repo. | 4 min for 100 epoch on goodwin dataset | https://colab.research.google.com/drive/1c3AQIu4CXDrXGjt_Ft_W2B3OMPepaQ97?usp=sharing | Yes | -- MIMIC-III dataser REQUIRES some training course on the ir website to be completed to acess. But the model runs on goodwin dataset. |
Stanford Cars | ProMetaR | [] | Prompt Learning via Meta-Regularization | 2024-04-01T00:00:00 | https://arxiv.org/abs/2404.00851v1 | [
"https://github.com/mlvlab/prometar"
] | {'Harmonic mean': '76.72'} | [
"Harmonic mean"
] | Given the following paper and codebase:
Paper: Prompt Learning via Meta-Regularization
Codebase: https://github.com/mlvlab/prometar
Improve the ProMetaR model on the Stanford Cars dataset. The result
should improve on the following metrics: {'Harmonic mean': '76.72'}. You must use only the codebase pro... | Prompt Learning via Meta-Regularization Jinyoung Park, Juyeon Ko, Hyunwoo J. Kim* Department of Computer Science and Engineering, Korea University {lpmn678, juyon98, hyunwoojkim }@korea.ac.kr Abstract Pre-trained vision-language models have shown impres- sive success on various computer vision tasks with their zero-sho... | 8 | 1 | The proposed ProMetaR framework builds upon existing vision-language models (VLMs) like CLIP, which are pre-trained on millions of image-text pairs. Given the extensive experiments mentioned, it's reasonable to assume a substantial dataset for fine-tuning, likely similar to CLIP's 400 million pairs. Fine-tuning such mo... | yes | Yes | CV | Prompt Learning via Meta-Regularization | 2024-04-01 0:00:00 | https://github.com/mlvlab/prometar | 1 | !git clone https://github.com/jhpohovey/StanfordCars.git
!mv StanfordCars/stanford_cars ./stanford_cars! | 2hr 10min for 10 epochs according to logs | https://drive.google.com/file/d/1gthiYFsffpGbuJcRv9QhtjyG8CRBi-rt/view?usp=sharing | Yes | -- Official website is down but found a git repo for dataset. |
CIFAR-10-LT (ρ=50) | SURE(ResNet-32) | [] | SURE: SUrvey REcipes for building reliable and robust deep networks | 2024-03-01T00:00:00 | https://arxiv.org/abs/2403.00543v1 | [
"https://github.com/YutingLi0606/SURE"
] | {'Error Rate': '9.78'} | [
"Error Rate"
] | Given the following paper and codebase:
Paper: SURE: SUrvey REcipes for building reliable and robust deep networks
Codebase: https://github.com/YutingLi0606/SURE
Improve the SURE(ResNet-32) model on the CIFAR-10-LT (ρ=50) dataset. The result
should improve on the following metrics: {'Error Rate': '9.78... | SURE: SUrvey REcipes for building reliable and robust deep networks Yuting Li1,2, Yingyi Chen3, Xuanlong Yu4,5, Dexiong Chen†6, and Xi Shen†1 1Intellindust, China 2China Three Gorges University, China 3ESAT-STADIUS, KU Leuven, Belgium 4SATIE, Paris-Saclay University, France 5U2IS, ENSTA Paris, Institut Polytechnique de... | 8 | 1 | The paper mentions using ResNet architectures and training over 200 epochs with a batch size of 128 on datasets like CIFAR-10 and CIFAR-100. Given that CIFAR-10 has 60,000 images (with a standard resolution of 32x32) and CIFAR-100 has 60,000 images as well, training these relatively smaller datasets with modern archite... | yes | Yes | CV | SURE: SUrvey REcipes for building reliable and robust deep networks | 2024-03-01 0:00:00 | https://github.com/YutingLi0606/SURE | 1 | downloaded through script | 1.5 min * 200 epoch = 5 hours | https://drive.google.com/file/d/1vEVFbmY0jFyW0WQj34SxupBXznFySsPL/view?usp=sharing | Yes | -- Just change the requirements.yml file as i noted and need to change the folder name an jsuut run the script. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.