Organizing recurrent network dynamics by task-computation to enable continual learning
Authors: Lea Duncker, Laura Driscoll, Krishna V. Shenoy, Maneesh Sahani, David Sussillo
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Employing a set of tasks used in neuroscience, we demonstrate that our approach successfully eliminates catastrophic interference and offers a substantial improvement over previous continual learning algorithms. |
| Researcher Affiliation | Collaboration | Lea Duncker Gatsby Unit, UCL London, UK duncker@gatsby.ucl.ac.uk Laura N. Driscoll Stanford University Stanford, CA lndrisco@stanford.edu Krishna V. Shenoy Stanford University Stanford, CA shenoy@stanford.edu Maneesh Sahani Gatsby Unit, UCL London, UK maneesh@gatsby.ucl.ac.uk David Sussillo Google Brain, Google Inc. Mountain View, CA sussillo@google.com |
| Pseudocode | No | The paper describes the proposed algorithm using mathematical equations (3), (4), and (5), but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or a link to open-source code for the described methodology. |
| Open Datasets | Yes | We demonstrate our continual learning approach on a set of tasks previously used for studying multi-task representations in RNNs [9]. |
| Dataset Splits | No | The paper refers to 'test trials' and 'test error' but does not explicitly provide details about training, validation, and test dataset splits, such as percentages or sample counts. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper does not specify version numbers for any software dependencies or libraries used for the implementation. |
| Experiment Setup | Yes | Networks with rectified-linear activation functions were trained on these tasks to minimize the squared error between readouts and target outputs under added L2-norm regularization of network weights and activity. |