Training stochastic stabilized supralinear networks by dynamics-neutral growth
Authors: Wayne Soo, Mate Lengyel
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We first show how SSNs can be trained to perform typical machine learning tasks by training an SSN on MNIST classification. We then demonstrate the effectiveness of our method by training an SSN on the challenging task of performing amortized Markov chain Monte Carlo-based inference under a Gaussian scale mixture generative model of natural image patches with a rich and diverse set of basis functions something that was not possible with previous methods. |
| Researcher Affiliation | Academia | Wayne W.M. Soo Department of Engineering University of Cambridge wmws2@cam.ac.uk Máté Lengyel Department of Engineering University of Cambridge Department of Cognitive Science Central European University m.lengyel@eng.cam.ac.uk |
| Pseudocode | No | The paper describes the training method in Section 2.3 and Figure 1 (Method sketch), but it does not include a section or figure explicitly labeled 'Pseudocode' or 'Algorithm'. |
| Open Source Code | Yes | The link to our code can be found in the supplementary material. |
| Open Datasets | Yes | We first train SSNs on MNIST digit classification [33]. CIFAR-10 and MNIST are publicly available datasets. |
| Dataset Splits | No | The paper states 'We first train SSNs on MNIST digit classification' and 'The encoded data are labeled according to their reduced dimensionality', and refers to 'MNIST data test accuracy' and 'CIFAR-10 images', but it does not specify exact train/validation/test split percentages or sample counts in the main text. The ethics review claims 'Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes]' but this claim itself isn't specific enough in the main paper without supplementary material access. |
| Hardware Specification | No | The main paper does not explicitly describe the hardware used to run its experiments. The ethics review states that 'Computational resources and related information can be found in the supplementary material,' but this information is not provided in the main text. |
| Software Dependencies | No | The paper mentions using 'backpropagation through time with Adam [44]' but does not provide specific software dependencies (e.g., library or solver names with version numbers like Python 3.8, PyTorch 1.9) in the main text. |
| Experiment Setup | No | The paper describes network architecture details (e.g., '80 excitatory and 20 inhibitory neurons') and components of the loss function (e.g., 'constant coefficients λµ, λσ2 and λΣ'), but it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs) or detailed training configurations in the main text. The ethics review states such details are in the supplementary material. |