Neuroscience Machine learning14 September 2026

Fast-weight continual learning in a fly

Peter Wang and Nico Christie

Code on GitHub

A fly that leaves a drop of food and wanders in the dark can walk straight back to it. To do that it keeps a running total of every step it takes. We read the fly’s complete wiring diagram to ask where that total is kept and how it steers the fly home.

  1. A fly may remember the way home in its synapses

    The running total of the fly’s steps could be stored in the strength of its connections rather than in cells that keep firing, and the wiring points to two cell types nobody has recorded from.

  2. Two unstudied cell types turn “where I went” into “go back”

    The stored total points away from the food. Only two cell types are wired to flip it around, so they are the likely route home.

  3. The fly’s sense of its own motion comes from unnamed cells

    The signals telling the compass how fast the fly turns and walks trace back to cell types nobody has named, fed by nerves coming up from the body.

  4. The fly’s compass is wired with a brake

    The cells that turn the compass also push back on themselves. Taken at face value that would stop the compass turning, so something must cancel it.