A fast algorithm to simulate nonlinear resistive networks
Authors: Benjamin Scellier
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We use our exact block coordinate descent algorithm to train deep resistive networks (DRNs) with equilibrium propagation (EP) (Scellier & Bengio, 2017) on the MNIST classification task. We train DRNs of one, two and three hidden layers (each comprising 1024 units), denoted DRN-1H, DRN-2H and DRN-3H, respectively. We also train another two DRNs with a single hidden layer each, compring 32784 units and 100 units, denoted DRN-XL and DRN-XS, respectively. [...] Table 1 shows the results. |
| Researcher Affiliation | Industry | Benjamin Scellier 1 1Rain AI, San Francisco, CA, USA. Correspondence to: Benjamin Scellier <benjamin@rain.ai>. |
| Pseudocode | No | The paper describes the algorithms and update rules in text (e.g., in Section 2.2 and 3.2) but does not provide them in a structured pseudocode or algorithm block. |
| Open Source Code | Yes | The code to reproduce the results is available at https://github.com/rain-neuromorphics/ energy-based-learning |
| Open Datasets | Yes | We use our exact block coordinate descent algorithm to train deep resistive networks (DRNs) with equilibrium propagation (EP) (Scellier & Bengio, 2017) on the MNIST classification task. |
| Dataset Splits | No | The MNIST dataset of handwritten digits (Le Cun et al., 1998) consists of 60,000 training examples and 10,000 test examples. While train and test set sizes are given, there is no explicit mention of a validation split or its size/methodology. |
| Hardware Specification | Yes | The simulations were carried out on a single Nvidia A100 GPU. |
| Software Dependencies | Yes | The code for the simulations uses Py Torch 1.13.1 and Torch Vision 0.14.1. |
| Experiment Setup | Yes | Table 2. Hyper-parameters used for initializing and training the five DRN models (DRN-XS, DRN-XL, DRN-1H, DRN-2H and DRN-3H) to reproduce the results in Table 1. LR stands for learning rate . |