Fly-on-the-Wall Trading: Coinbase Engineer Simulates Fly Brain for Crypto Day Trades
An engineer at Coinbase has released an open-source project called Stonkfly that uses a simulated neural network from a male fruit fly's brain to perform cryptocurrency day trading. The model, based on 'MaleCNS v1.0,' simulates 166,700 neurons and approximately 25.6 million connections.
The system generates a 320x180 pixel image from Coinbase's public price data, displaying historical price movements, trading pairs, and current bid and ask prices. The decision to buy or sell uses the number of times specific nerve cells 'fire,' which emit electrical signals.
A predetermined rule maps the difference in firing frequency between left and right groups of nerve cells, and the presence or absence of firing in other nerve cells, to 'buy,' 'sell,' or 'do nothing.' The system also incorporates a mechanism to feed back trading results into the neural circuitry, stimulating dopamine cells when the total asset value increases or decreases.
Nftechie states that the fact that visual stimuli are transmitted to neural circuits, causing dopamine cells to fire and changing the strength of connections, is different from learning profitable trading methods. As of now, Stonkfly has not demonstrated improved trading ability or performance with real funds.