Dr. Elena Rivers

DriftDB

2026-05-01T00:00:00+00:00

DriftDB is a curated dataset of two-photon calcium imaging from mouse visual cortex, recorded from the same neurons across six weeks during a stable discrimination task. It is the dataset behind our 2026 work on representational drift.

What’s inside</h2>

~4,000 neurons tracked across 30+ sessions per animal (6 animals).</li>
Aligned ROIs, deconvolved activity, and full behavioral logs.</li>
A NWB-formatted release plus loader utilities for NeuroDyn</a>.</li> </ul>
We release DriftDB under CC-BY so the community can test theories of drift and stability without re-collecting expensive longitudinal data.

LocalProp

2025-11-01T00:00:00+00:00

LocalProp collects clean, benchmarked implementations of learning rules that avoid the biologically implausible weight transport of backpropagation — feedback alignment, target propagation, predictive coding, and our own local error signals from the NeurIPS 2025 paper.
Goals</h2>

One consistent training loop so rules can be compared fairly.</li>
Benchmarks on vision and sequence tasks with logged compute budgets.</li>
A teaching resource: each rule has an annotated, minimal notebook.</li> </ul>
If you are exploring how the brain might approximate gradient descent, this is a good place to start experimenting.

NeuroDyn

2024-06-01T00:00:00+00:00

NeuroDyn fits state-space and switching linear dynamical systems to recordings from thousands of simultaneously imaged or recorded neurons. It is built on JAX, so models scale to GPU/TPU and differentiate end-to-end.
Highlights</h2>

Pluggable observation models (Gaussian, Poisson, negative-binomial).</li>
Variational and Laplace inference with a unified API.</li>
Tools for cross-validation, model comparison, and trajectory visualization.</li>
Tutorials reproducing results from several published datasets.</li> </ul>
Why it exists</h2>
Most labs re-implement the same inference machinery for every project. NeuroDyn packages it once, tests it thoroughly, and gets out of your way so you can focus on the science.
import neurodyn as nd model = nd.SLDS(n_states=4, n_latents=8, observations="poisson") posterior = model.fit(spikes, n_iters=500) nd.plot.trajectories(posterior)</code></pre>