Related Experiment Video
Updated: Aug 19, 2025
08:28
Automated Multimodal Stimulation and Simultaneous Neuronal Recording from Multiple Small Organisms
Published on: March 3, 2023
1.1K
A large-scale neural network training framework for generalized estimation of single-trial population dynamics
Mohammad Reza Keshtkaran1, Andrew R Sedler1,2, Raeed H Chowdhury3,4
1Wallace H. Coulter Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, GA, USA.
Nature Methods
|November 28, 2022
Summary
AutoLFADS automates hyperparameter tuning for deep neural population dynamics models. This framework achieves state-of-the-art performance across diverse brain datasets without task-specific information.
Area of Science:
- Computational Neuroscience
- Machine Learning for Neuroscience
Background:
- Deep neural population dynamics models achieve high performance but require extensive dataset-specific hyperparameter tuning.
- Automating this tuning process is crucial for broader application of these models.
Purpose of the Study:
- To introduce AutoLFADS, an automated framework for tuning deep neural population dynamics models.
- To demonstrate the framework's ability to achieve state-of-the-art performance without using behavioral or task information.
Main Methods:
- Developed AutoLFADS, a model-tuning framework utilizing autoencoding models.
- Applied the framework to diverse rhesus macaque neural datasets from motor cortex, somatosensory cortex, and dorsomedial frontal cortex.
Main Results:
- AutoLFADS automatically generated high-performing autoencoding models.
- The framework demonstrated broad applicability across different brain areas and tasks, including reaching and cognitive timing.
Conclusions:
- AutoLFADS effectively automates the tuning of deep neural population dynamics models.
- The framework offers a generalizable solution for achieving high performance on neural data without task-specific inputs.
