Adaptive whitening with fast gain modulation and slow synaptic plasticity

Authors: Lyndon Duong, Eero Simoncelli, Dmitri Chklovskii, David Lipshutz

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our model on synthetic and natural datasets and find that the synapses learn optimal configurations over long timescales that enable adaptive whitening on short timescales using gain modulation. and 5 Numerical experiments We test Alg. 1 on stimuli s1, s2, . . . drawn from slowly fluctuating latent contexts c1, c2, . . .
Researcher Affiliation Academia 1 Center for Computational Neuroscience, Flatiron Institute 2 Center for Neural Science, New York University 3 Neuroscience Institute, NYU Langone Medical School and {lyndon.duong, eero.simoncelli}@nyu.edu {dchklovskii, dlipshutz}@flatironinstitute.org
Pseudocode Yes Algorithm 1: Multi-timescale adaptive whitening via synaptic plasticity and gain modulation
Open Source Code Yes Python code accompanying this study can be found at https://github.com/lyndond/multi_timescale_whitening.
Open Datasets Yes We test our algorithm on 56 highresolution natural images [34] and [34] JH van Hateren and A van der Schaaf. Independent component filters of natural images compared with simple cells in primary visual cortex. Proceedings: Biological Sciences, 265 (1394):359 366, 1998.
Dataset Splits No We train our algorithm in the offline setting where we have direct access to the context-dependent covariance matrices (Appx. C, Alg. 2, α = 1, J = 50, ηg =5E-1, ηw =5E-2) with K = N = 25 and random W0 O(25) on a training set of 50 of the images, presented uniformly at random 1E3 total times. and We test the circuit with fixed synaptic weights WT and modulated (adaptive) gains g on stimuli from the held-out images. There's no explicit mention of a validation set or how the dataset was split for training, validation, and testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions 'Python code' in footnote 2 but does not list any specific software dependencies with version numbers.
Experiment Setup Yes We train our algorithm in the offline setting where we have direct access to the context-dependent covariance matrices (Appx. C, Alg. 2, α = 1, J = 50, ηg =5E-1, ηw =5E-2) with K = N = 25 and random W0 O(25) on a training set of 50 of the images, presented uniformly at random 1E3 total times.