All things neural distribution alignment | neuralign

Dimensionality reduction is a powerful tool to analyze neural activity at the population level. However, a number of considerations - including the non-uniqueness of low-dimensional representations, instabilities in recording, and biological factors like plasticity - can cause the latent representations of two recordings, which should be similar enough to be directly comparable, to differ by a transformation. The objective of neural distribution alignment is to recover and invert this transformation, so that we can robustly compare neural activity across time, space, and even subjects. We provide here open-source access to two algorithms designed to tackle this problem: Distribution Alignment Decoding, which uses a brute-force approach to find the optimal transformation; and Hierarchical Wasserstein Alignment, which exploits the tendency of neural distributions to be composed of clusters to find the solution more elegantly. Find below implementations in MATLAB and the Python scientific

All things neural distribution alignment | neuralign neuralign All things neural distribution alignment Dimensionality reduction is a powerful tool to analyze neural activity at the population level. However, a number of considerations - including the non-uniqueness of low-dimensional representations, instabilities in recording, and biological factors like plasticity - can cause the latent representations of two recordings, which should be similar enough to be directly comparable, to differ by a transformation. The objective of neural distribution alignment is to recover and invert this transf

related reading

• Dimensionality reduction in neural data analysisxcorr.net
• 2205.08413.pdfarxiv.org
• Toward accessible, real-time brain decoding: Introducing ENIGMA | Alljoined Blogalljoined.com
• Estimating the dimensionality of the manifold underlying multi-electrode neural recordingsncbi.nlm.nih.gov
• Towards a “universal translator” for neural dynamics at single-cell, single-spike resolutionarxiv.org
• Naturally Occurring Equivariance in Neural Networksdistill.pub
• Neural Data Transformerarxiv.org
• POYO-1poyo-brain.github.io
• Neural population geometry and optimal coding of tasks with shared latent structurenature.com
• [2305.18274] Reconstructing the Mind's Eye: fMRI-to-Image with Contrastive Learning and Diffusion Priorsarxiv.org
• Arousal as a universal embedding for spatiotemporal brain dynamicsnature.com
• Alljoined1 -- A dataset for EEG-to-Image decodingarxiv.org