/Feedforward_Adaptation

Online Model Adaptation with Feedforward Compensation

Primary LanguagePythonMIT LicenseMIT

Online Model Adaptation with Feedforward Compensation

This repository contains the code for Feedforward and Online Model Adaptation, as demonstrated in the following papers:

Abulikemu Abuduweili, and Changliu Liu, "Online Model Adaptation with Feedforward Compensation," CoRL, 2023.

Abstract

To cope with distribution shifts or non-stationarity in system dynamics, online adaptation algorithms have been introduced to update offline-learned prediction models in real-time. Existing online adaptation methods focus on optimizing the prediction model by utilizing feedback from the latest prediction error. Unfortunately, this feedback-based approach is susceptible to forgetting past information. This work proposes an online adaptation method with feedforward compensation, which uses critical data samples from a memory buffer, instead of the latest samples, to optimize the prediction model. We prove that the proposed approach achieves a smaller error bound compared to previously utilized methods in slow time-varying systems. Furthermore, our feedforward adaptation technique is capable of estimating an uncertainty bound for predictions.

About Code

Install Requirments

pip install numpy pandas scikit-learn torch

Training the Model on etth1/ill/exchange:

python train.py --data etth1

Adapting the trained model with Feedforward Adaptation:

python adap.py --data etth1 --adapt sgd --buffer_size 1000

Citation

If you find the code helpful in your research or work, please cite the following papers.

@inproceedings{
abuduweili2023online,
title={Online Model Adaptation with Feedforward Compensation},
author={ABULIKEMU ABUDUWEILI and Changliu Liu},
booktitle={7th Annual Conference on Robot Learning},
year={2023},
}