Biological credit assignment through dynamic inversion of feedforward networks
Authors: Bill Podlaski, Christian K. Machens
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We tested dynamic inversion (DI) and non-dynamic inversion (NDI) against backpropagation (BP), feedback alignment (FA), and pseudobackprop (PBP) on four modest supervised and unsupervised learning tasks linear regression, nonlinear regression, MNIST classification, and MNIST autoencoding. |
| Researcher Affiliation | Academia | William F. Podlaski Champalimaud Research Champalimaud Centre for the Unknown 1400-038 Lisbon, Portugal Christian K. Machens Champalimaud Research Champalimaud Centre for the Unknown 1400-038 Lisbon, Portugal Correspondence: {william.podlaski, christian.machens}@research.fchampalimaud.org |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks (e.g., a section labeled 'Algorithm' or 'Pseudocode'). |
| Open Source Code | No | The paper does not provide any concrete access information (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper. |
| Open Datasets | Yes | We next tested dynamic inversion on the MNIST handwritten digit dataset, where we use the standard training and test datasets (Le Cun et al., 1998)... |
| Dataset Splits | No | The paper mentions using 'standard training and test datasets' for MNIST, but does not explicitly provide details for a separate validation split or specific percentages/counts for all three splits (train/validation/test). |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | Learning rate =10 2 for all algorithms. and DI was simulated numerically using 1000 Euler steps with dt = 0.5. and mini-batch training (100 examples per batch). and learning rate = 10 3 for all algorithms. and learning rate = 10 6 for all algorithms. |