Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Convolutional Channel-Wise Competitive Learning for the Forward-Forward Algorithm
Authors: Andreas Papachristodoulou, Christos Kyrkou, Stelios Timotheou, Theocharis Theocharides
AAAI 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our method outperforms recent FF-based models on image classification tasks, achieving testing errors of 0.58%, 7.69%, 21.89%, and 48.77% on MNIST, Fashion MNIST, CIFAR-10 and CIFAR-100 respectively. Our approach is benchmarked against FF-based and non Backpropagation methods reported in the literature on image classification tasks. |
| Researcher Affiliation | Academia | KIOS Research and Innovation Center of Excellence, Department of Electrical and Computer Engineering, University of Cyprus EMAIL |
| Pseudocode | Yes | Algorithm 1: Interleaved Layer Training (ILT) Strategy |
| Open Source Code | Yes | Our source code and supplementary material are available at https://github.com/andreaspapac/CwComp. |
| Open Datasets | Yes | The performance of the various models was evaluated on three benchmark datasets commonly used in literature, namely MNIST (Le Cun et al. 1998), Fashion MNIST (Xiao, Rasul, and Vollgraf 2017), and CIFAR-10 (Krizhevsky and Hinton 2009). Our method outperforms all FF-based models on basic image classification tasks, demonstrating quicker convergence rates and significantly better performance, achieving testing errors of 0.58%, 7.69%, 21.89%, and 48.77% on MNIST (Le Cun et al. 1998), Fashion-MNIST (Xiao, Rasul, and Vollgraf 2017), CIFAR10, and CIFAR-100 (Krizhevsky and Hinton 2009) respectively. |
| Dataset Splits | No | The paper does not explicitly provide details about training/validation/test dataset splits (e.g., percentages, sample counts for each split) nor does it specify how a validation set was used for the standard benchmark datasets. |
| Hardware Specification | Yes | Our models were trained on an NVIDIA-RTX 4080 GPU and the i9-13900K Intel CPU for 20 epochs on MNIST, 50 epochs on Fashion-MNIST, and CIFAR-10. |
| Software Dependencies | No | The paper mentions 'torchinfo (Harris 2016)' as an open-source toolbox for calculating parameters and operations, but it does not specify a version number for torchinfo or any other software dependencies crucial for reproducibility. |
| Experiment Setup | Yes | In the experiments, the learning rate used for both the convolutional and dense layers was set to 0.01, and the batch size to 128. For comprehensive details on the model configurations, refer to the supplementary material. |