Neuronal Spiking Ensembles: Dynamics of Representational Geometry

Last updated on Jan 24, 2024

Title

Neural Spiking Ensembles: Dynamics of Representational Geometry

Leaders

Richard Song: @richardwsong

Collaborators

Ken Rahman: @RahmanKF22 Sam Abbaspoor: @SAbbaspoor Kari Hoffman: @perpl_lab

Brainhack Global 2023 Event

BrainHack Vanderbilt

Project Description

What are the meaningful changes in the brain with experience, that allows for adaptive behavior? When we look at the coordinated activity across spiking networks of neuronal ensembles, we see a delicate balance of stability and flexibility, as needed for a system that can both learn and remember. In this project, we present a population of simultaneously-recorded neurons from the non-human primate during learning of a complex sequence memory task, and in sleep afterwards. These data are exceptionally rich for exploration, but also to address three fundamental questions: 1. can we decode behavioral states from the ensemble dynamics, 2. what is the core representational geometry of the ensembles (what factors are best preserved/differentiated in low-dimensional spaces, and how does the geometry constrain the computations and dynamics of the network, and finally, 3. Does the ensemble activity drift with time and experience, and if so, how?

https://github.com/hoffman-lab/BrainHacks24-NeuralManifolds

Goals for Brainhack Global

Our goals for you include:

  1. Being able to understand the format of neuronal spiking and the corresponding behavioral data.
  2. Use manifold learning techniques (e.g. PCA, tSNE, UMAP, CEBRA) to decode relevant behaviors in complex neural spiking data in low dimensional space.
  3. Modify manifold learning hyperparameters to achieve best behavioral decoding ability.
  4. Explore changes in single-cell level or behaviors to elucidate mechanisms of representational drift.

Good first issues

  1. Experiment on neural manifold creation with using different hyperparameters (e.g. n_neighbors or min_dist in UMAP). Different parameters can greatly change the shape of the manifold and thus can affect the ability to decode different behaviors. Afterwards, consider creating UI to visualize neural manifolds across different hyperparameters. Here is a great example.
  2. Parallelize dimensionality reduction across a set of hyperparameters
  3. Experiment with which behavioral parameters are most separated in low dimensional space (block type, trial number, time, head position, angular velocity, etc.)
  4. Explore characteristics that may be leading to representational drift. How is firing rate of the cells changing across trials and over time? What about accuracy, time to perform the task, or reward received?
  5. Structural Index for Neural Manifolds

Communication channels

manifold_tuning channel on the discord: https://discord.gg/jbQWFhKn

Skills

Onboarding documentation

No response

What will participants learn?

Participants will gain experience analyzing high dimensional neural spiking data at a population-level using manifold learning in Python. They will work at the cutting-edge of systems neuroscience research, working with novel and groundbreaking data collected on non-human primates.

Data to use

No response

Number of collaborators

4

Credit to collaborators

Project contributors are credited on the Readme and major contributors may be considered for coauthorship.

Image

image image (Sebastian et al., Nature Neuroscience, 2023)

Type

coding_methods, pipeline_development, visualization

Development status

1_basic_structure

Topic

data_visualisation, machine_learning, neural_decoding, PCA, single_neuron_models

Tools

Jupyter, other

Programming language

Python

Modalities

other

Git skills

2_branches_PRs

Anything else?

No response

Things to do after the project is submitted and ready to review.


Date
Jan 1, 0001 12:00 AM