Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Continuous Attractor Neural Networks Toolkit: Standardizing Computational Neuroscience Research on Attractor Dynamics

Sichao He¹, AIERSI TUERHONG², Shangjun Sheh¹, Si Wu¹; ¹Peking University, ²Chongqing University

Presenter: Sichao He

Continuous attractor neural networks (CANNs) are a canonical computational model for how the brain encodes continuous variables such as spatial position, head direction, and moving direction. They successfully explain key phenomena in hippocampal place cells, entorhinal grid cells, and head direction systems. Despite their importance, CANN research suffers from fragmentation: researchers implement models from scratch, use incompatible codebases, and face significant reproducibility barriers. We present a unified Python toolkit for CANN research built on BrainPy and JAX, providing standardized implementations, analysis pipelines, and reproducible experimental workflows. The toolkit includes: (1) a model library covering mathematically tractable 1D/2D CANNs, spike-frequency adaptation (SFA) augmented models, and grid cell networks; (2) task generators for smooth tracking, population coding, and open/closed-loop navigation; (3) an analyzer suite with energy landscape visualization, tuning curve analysis, and topological data analysis (TDA) for characterizing attractor geometry; and (4) the Attractor Structure Analyzer (ASA) pipeline---a GUI/TUI tool for applying TDA-based methods to experimental neural recordings to detect attractor signatures. The toolkit reproduces key published results including SFA-driven anticipative tracking dynamics and theta-sweep phenomena in head direction and place cell systems, providing trusted baselines for new modeling studies. High-performance computation is achieved via JAX JIT compilation with GPU/TPU support and optional Rust-accelerated backends for performance-critical routines. By standardizing the CANN research workflow, analogous to how Hugging Face Transformers unified NLP research, this toolkit lowers barriers for computational neuroscientists, AI researchers, and students, and enables direct comparison of alternative mechanisms under shared experimental setups. We demonstrate the toolkit's capabilities through simulations of canonical attractor phenomena and analysis of spatial navigation dynamics.

Topic Area: Methods, Tools, Theory & Neural Coding