Using the Transformer architecture for time-series forecasting and Kalman filter for the model modification.
Transformer architecture is so useful when it comes to seq2seq models that we can easily using Transformer for time-series forecasting. However, models are always not generalizing enough especially for stocks data.
Kalman filter can be applied to linear-model modification. The Unscented Kalman Filter (UKF) can be used for nonlinear estimation. Therefore, I believe we can apply UKF to nonlinear-model such as transformer or other deep learning models.
Anyone who interesting in this project or Quant, please feel free to contact me. I'm looking forward to working with you.