vaxenfly
A fruit fly, the animal whose brain vaxenfly reads

Read every agent state through a real fly brain

CA 0xxx @Valencefly

Four pictures

  1. A fruit fly in close-up: big red eyes, two wings, six legs

    This is a fruit fly.

  2. A render of the real fly brain from the connectome, with its two mushroom bodies glowing orange

    Its brain, mapped neuron by neuron.

  3. A render of one mushroom body split into its sixteen compartments, good ones orange, bad ones blue, each labelled

    Deep inside is the mushroom body: 16 rooms, each saying good or bad.

  4. A bar chart with one row per compartment: bars to the right for good, dithered bars to the left for bad

    vaxenfly pushes your agent's state through those rooms and reads 16 good/bad numbers.

No dependencies. About 0.65 ms per read.

npx skills add https://github.com/vaxenfly/vaxenfly

How it reads

Four stages.

  1. 01. Project
  2. 02. Expand
  3. 03. Sparsify
  4. 04. Read out

Signed by the fly

Each compartment's sign is its own PAM to PPL1 ratio.

  • Derived signs

    Nobody chose them.

  • No update rule

    Same input, same output, forever.

  • Zero dependencies

    No network.

  • Proven red

    18 gate arms, each broken once on purpose.

The compartments

Sixteen compartments, each with its sign.

See every compartment
  • CA0
  • a'1-
  • a'2-
  • a'3-
  • a1+
  • a2-
  • a3-
  • b'1+
  • b'2+
  • b1+
  • b2+
  • g1-
  • g2-
  • g3+
  • g4+
  • g5+

+ appetitive, - aversive, 0 no teaching signal.

Give your agent a gut

Install it, read a state, compare the scalar across states.

  1. Install it
  2. Read a state
  3. Compare the scalar
Install vaxenfly

MIT licensed. Nothing to train.