EVOTS Evolutionary Architecture Search for Time Series Forecasting
July 2, 2026
EVOTS uses a modular genome representation and a repair mechanism to evolve task-adaptive Transformer architectures for multivariate time-series forecasting. The framework automates the discovery of attention and projection components without relying on hand-crafted design rules.
HOW THIS AFFECTS YOU
●
builderYou can use task-specific architectures to improve forecasting accuracy on specialized datasets.
●
researcherYou can move beyond fixed Transformer architectures using automated evolutionary search.