Multi-Objective Neural Bandits

This repository contains the official implementation of our IJCAI 2025 paper:
Multi-Objective Neural Bandits with Random Scalarization

Overview

We provide code for simulating and solving multi-objective contextual bandit problems using neural-based methods. The key components include:

• environments.py
  Contains simulators for multi-objective contextual bandits.
  To apply the framework to a real-world dataset, implement custom versions of the _sample_context and _eval_expected_reward methods in a subclass of the base class moContextMABSimulator.

• agents.py
  Implements multi-objective neural bandit algorithms, including MONeural-UCB and MONeural-TS.
  For a quick start, refer to the usage example in examples.ipynb.

• utils.py
  Provides utility functions to support multi-objective optimization tasks.

Citation

If you find this work useful in your research, please consider citing:

@inproceedings{cheng2025multi,
  title={Multi-Objective Neural Bandits with Random Scalarization},
  author={Cheng, Ji and Xue, Bo and Lu, Chengyu and Cui, Ziqiang and Zhang, Qingfu},
  booktitle={Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, {IJCAI-25}},
  year={2025},
}