🔬
EM-NAV: Geometry & Transferability of Spatial Representations
License: GPL v3 · GitHub Repo
→
Computational neuroscience research study investigating how population sparsity, spiking thresholds, and network recurrence shape the geometric layout, linear decodability, and zero-shot transfer stability of emergent spatial manifolds learned from egocentric sensory data.
Layperson Translation:
We are testing if forcing artificial intelligence to obey the same biological constraints as the human brain (like using spikes of electricity and very little energy) forces the AI to build a "mental map" of its world, just like we do.
Research Stack:
Python 3.10+ · snnTorch · PyTorch · Blender 3D · MiniGrid · PPO