Biologically Plausible Sequence Learning with Spiking Neural Networks
Authors: Zuozhu Liu, Thiparat Chotibut, Christopher Hillar, Shaowei Lin1316-1323
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we showcase the biologically plausible features and capabilities of the MPN.2 We first numerically verify the prediction of Theorem 0.1, reproducing STDP-like curves from our learning rule. The MPN is then shown to be self-consistent, i.e. accurate inference of its own generative model is achievable. We next demonstrate the ability of MPN to robustly memorize spatiotemporal patterns. Namely, we show that our model can be trained to memorize repeating sequences of binary vectors with one-hop transitions and that these memories are robust in the sense that they are stable attractors of the MPN dynamics in the deterministic limit. When the dynamics are stochastic, we show that stochasticity assists in multiple-memory switching, as opposed to the deterministic counterpart whose dynamics eventually fixes at only one memory. Finally, we end this section by highlighting two potentially useful sequence learning applications: memorizing a long sequence of binary pictures (shown in Fig. 4 and the video in SM) and learning a generative model of experimental neural spike-train data. The latter demonstrates that MPNs can effectively reproduce spike-timing statistics in neurobiology experiments. |
| Researcher Affiliation | Collaboration | 1Department of Statistics and Applied Probability, National University of Singapore, 2Department of Physics, Faculty of Science, Chulalongkorn University, Thailand, 3Engineering Systems and Design, Singapore University of Technology and Design, 4Awecom, Inc, 5Redwood Center for Theoretical Neuroscience, University of California, Berkeley |
| Pseudocode | Yes | See SM C for the training algorithm pseudocode. |
| Open Source Code | Yes | The codes are available at https://github.com/owen94/MPNets. |
| Open Datasets | Yes | The dataset is available on https://crcns.org/data-sets/vc/pvc-3 |
| Dataset Splits | No | The paper specifies a train/test split (70% training, 30% testing) for the neural spike-train data, but it does not explicitly mention a separate validation set or its proportion. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, memory, or other computational resources used for running the experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies or libraries with their version numbers that are necessary to replicate the experimental environment. |
| Experiment Setup | Yes | Here, we set the learning rate for the transition and holding updates to ηT = 0.05 and ηH = 0.001, respectively. Other parameters are wij = 1, bi = bj = 0. (...) The learning rate is set to η = 0.001. (...) The learning rate is set to η = 0.01. |