Sticker

Sticker is a fundamental neural network architecture with shallow mixture-of-experts (MoE) consensus based on biological principles.

The name comes from the phrase 'scalping on tickers', related to the first performed financial market mid-price scalping test showcasing the model.

As a continual learning model, Sticker outperforms all other neural network models in accuracy and latency, with orders of magnitude higher efficiency than traditional neural networks in the test. It achieves the best throughput among the tested baselines, including classical and statistical models. Sticker represents a new state-of-the-art in continual learning while establishing a new Pareto frontier in performance versus efficiency.

Test details, configuration and comparison are available in mid_price_movement/