TiDE: the ‘embarrassingly’ simple MLP that beats Transformers
A deep exploration of TiDE, its implementation using Darts and a real life use case comparison with DeepAR and TFT (a Transformer architecture)
As industries continue to evolve, the importance of an accurate forecasting becomes a non-negotiable asset whether you work in e-commerce, healthcare, retail or even in agriculture. The importance of being able to foresee what comes next and plan accordingly to overcome future challenges is what can make you ahead of competition and thrive in an economy where margins are tight and the customers are more demanding than ever.
Transformer architectures have been the hot topic in AI for the past few years, specially due to their success in Natural Language Processing (NLP) being one of the most successful use cases the chatGPT that took the attention of everyone regardless if you were an AI enthusiastic or not. But NLP is not the only subject where Transformers have been shown to outperform the state-of-the-art solutions, in Computer Vision as well with Stable Diffusion and its variants.
But can Transformers outperform state-of-the-art models in time series? Although many efforts have been made to develop Transformers for time series forecasting, it seems that for long term horizons, simple linear models can outperform several Transformer based approaches.
