Reproduce “Dopamine: Brain Modes, Not Brains” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shervin Ghasemlou

TauGate is an activation-space parameter-efficient fine-tuning technique that adapts pretrained models by learning per-neuron thresholds and gains to modulate existing computations. This method enables explicit conditional computation and provides interpretability of neuron participation for specific tasks, demonstrating performance improvement with a minimal parameter count.

Key insights

  • Neural network adaptation can be achieved through "mode switching," where neuromodulator-inspired thresholds and gains selectively enable or rescale existing computational pathways.
  • Explicit conditional computation, where a neuron's participation is clearly defined by its pre-activation exceeding a learned threshold, significantly improves interpretability of task adaptation.
  • While thresholds define gating semantics, the gain component is primarily responsible for empirical accuracy improvements in the tested mode specialization scenario.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.