Natural gradient enables fast sampling in spiking neural networks

Authors: Paul Masset, Jacob Zavatone-Veth, J. Patrick Connor, Venkatesh Murthy, Cengiz Pehlevan

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Figure 2 and in Supplemental Figures E.1-E.3, we show that the geometry-aware sampler is more robust to increasing the correlation of the parameters and the dimensionality, allowing inference at behavioral timescales. In Figures 3 and 4, and in Supplemental Figure E.4, we show that naïve choices of Γ lead to vanishing spike rates at strong correlations ρ and large parameter-space dimensionalities np. This results in dramatic underestimation of the mean and variance of the target distribution, which is resolved by choosing the geometry appropriately, again allowing inference at behavioral time-scales.
Researcher Affiliation Academia Paul Masset1,2 , Jacob A. Zavatone-Veth1,3 , J. Patrick Connor4, Venkatesh N. Murthy1,2 , Cengiz Pehlevan1,4 1Center for Brain Science, 2Department of Molecular and Cellular Biology, 3Department of Physics, 4John A. Paulson School of Engineering and Applied Sciences, Harvard University Cambridge, MA 02138
Pseudocode No The paper describes its methods using mathematical equations and prose but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any statement about making source code publicly available or provide any links to a code repository.
Open Datasets No The paper describes generating synthetic data for simulations (e.g., 'equicorrelated multivariate distributions', 'Gaussian distribution N(µ, Σ)') but does not provide concrete access information or citations for a publicly available or open dataset.
Dataset Splits No The paper discusses 'simulations' and 'realizations' for statistical analysis, but it does not specify explicit training, validation, or test dataset splits.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library names with version numbers.
Experiment Setup No The paper mentions some parameters related to the model (e.g., 'τm = 20 ms', 'np = 20-dimensional', 'nn = 200 neurons', 'correlation ρ = 0.75') and simulation conditions, but it lacks a comprehensive description of the experimental setup, including specific hyperparameter values, optimizer settings, or detailed training configurations.