Spectral Learning of Shared Dynamics Between Generalized-Linear Processes

Authors: Lucine L Oganesian, Omid G. Sani, Maryam Shanechi

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In simulations, we demonstrate our algorithm s ability to dissociate and model the dynamics within two time-series sources while being agnostic to their respective observation distributions. In neural data, we consider two specific applications of our algorithm for modeling discrete population spiking activity with respect to a secondary time-series. In both synthetic and real data, GLDMs learned with our algorithm more accurately decoded one time-series from the other using lower-dimensional latent states, as compared to models identified using existing GLDM learning algorithms.
Researcher Affiliation Academia Lucine L. Oganesian Ming Hsieh Department of Electrical and Computer Engineering University of Southern California Los Angeles, CA loganesi@usc.edu
Pseudocode No The paper describes the algorithm steps in detail using mathematical formulations and descriptive text, but it does not include a dedicated "Pseudocode" or "Algorithm" block or figure.
Open Source Code Yes Corresponding author. Project code: https://github.com/ShanechiLab/PGLDM
Open Datasets Yes Next we demonstrate our method on two public non-human primate (NHP) datasets of discrete population spiking activity recorded from different brain regions and during different contexts [16 18].
Dataset Splits Yes We evaluated decoding performance of learned models using five-fold cross validation across six recording sessions (see appendix A.7.2 for cross-validation details).
Hardware Specification Yes All running time analyses were performed on a 2020 Macbook Pro using 2 GHz Quad-Core Intel Core i5 CPU with 16GB of 3733 MHz RAM.
Software Dependencies No The paper mentions using "Python s CVXPY package" and "scikit-learn; statsmodels" but does not specify exact version numbers for these software components or Python itself. For example, "We used Python s CVXPY package to solve the semidefinite programming problem defined in equation (30) [39, 40]."
Experiment Setup Yes For our synthetic data in section 4.1, we simulated generalized-linear observations from random models as per equation (6). For the simulations used to generate the results in Table 1, we fixed the number of shared and private latent states as n1 = 2, n2 = 6, and n3 = 4. We randomly selected the observation dimensions with uniform probability from the following ranges: 10 nr 15 and either 5 nz 10, when the secondary observation was Gaussian, or 10 nz 15, when Poisson.