Accurate Identification of Communication Between Multiple Interacting Neural Populations

Authors: Belle Liu, Jacob Sacks, Matthew D. Golub

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To evaluate MR-LFADS, we developed 37 synthetic multi-region datasets that capture real-world challenges in communication modeling across a range of neuroscience-relevant scenarios. On these datasets, MR-LFADS consistently outperforms existing models in recovering the pathways and content of communication. Through targeted ablations of key design features, we demonstrate that these features indeed improve the identification of communication. We then applied MR-LFADS to multi-region electrophysiological recordings in mice performing a decision-making task (Chen et al., 2024).
Researcher Affiliation Academia 1Graduate Program in Neuroscience, University of Washington; 2Paul G. Allen School of Computer Science & Engineering, University of Washington. Correspondence to: Matthew Golub <EMAIL>.
Pseudocode No The paper describes the MR-LFADS model and related methods using mathematical equations and prose. It does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statement about releasing the source code for the MR-LFADS methodology, nor does it provide a link to a code repository.
Open Datasets Yes We then applied MR-LFADS to multi-region electrophysiological recordings in mice performing a decision-making task (Chen et al., 2024).
Dataset Splits Yes In a subset of trials that were held out during model fitting, photoinhibition was applied to the anterior lateral motor cortex. MR-LFADS was trained only on unperturbed ( control ) trials, with photoinhibition trials held out for validation. ... During training, we held out 10% of the neurons in each real or synthetic brain region for validation. ... Synthetic datasets for networks from Experiments 1-3 have 1024, 1024 and 820 total trials respectively, of which 85% is used for training, and 15% for validation.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments. It only mentions using 'large-scale electrophysiology' for data collection but not for computation.
Software Dependencies No The paper mentions 'PyTorch', 'Optuna', 'Ray Tune', and 'TPE algorithm' but does not specify any version numbers for these software components.
Experiment Setup Yes Table S1. Key hyperparameters for MR-LFADS models. ... learning rate [10 5, 0.004] ... Total time steps used for inferring inferred inputs ... Total time steps used for inferring the initial condition ... Total number of epochs ... KL penalty coefficient for u; performs search for this hyperparameter ... Epoch at which βu starts increasing from 0 ... Number of epochs for βu to reach the maximum value ... KL penalty coefficient for m; performs search for this hyperparameter ... L2 penalty coefficient ... Number of neurons ... Generator size ... Factor size ... Inferred input dimension ... Inferred message dimension.