kcompher / amLDS

'Aligned mixture of latent dynamical systems' (amLDS) for stimulus decoding probabilistic manifold alignment across animals. P. Herrero-Vidal et al. NeurIPS 2021 code.

Across-animal odor decoding by probabilistic manifold alignment (NeurIPS 2021)

This repository is the official implementation of aligned mixture of latent dynamical systems (amLDS) published at NeurIPS 2021.

amLDS is a probabilistic method to align neural responses and efficiently decode stimuli across animals. It learns independent mappings of different recordings into a shared latent manifold, where stimulus-evoked dynamics are similar (identical) across animals but distint across stimuli allowing for accurate stimulus decoding.

A full description of the algorithm can be found in the preprint.

Requirements

• numpy == 1.16.2
• matplotlib == 3.0.3
• seaborn == 0.9.0
• sklearn == 0.20.3
• scipy == 1.2.1
• python == 3.7.3+

Usage

To get started, run the example notebook 'amLDS_example'. This notebook contains an example on the use of amLDS on synthetic data. It shows how to perform parameter learning, inference and stimulus decoding; as well as latent dimensionality estimation.

To explore other properties and capabilities of amLDS check the 'amLDS_mixturesConcentration' notebook or run the performance script as python3 'amLDS_Performance_DataDemands_ModelComparison.py'.

Copyrights and license

This code has been released under the GNU AGPLv3 license. For the usage of modification of any of this repository content cite:

Pedro Herrero-Vidal, Dmitry Rinberg, Cristina Savin, "Across-animal odor decoding by probabilistic manifold alignment". bioRxiv, 2021