- Run following command to set up codebase and Python environment.
git clone https://github.com/AnikiFan/MoETTA.git
cd MoETTA
# In case you haven't install uv, run following command if you are using Linux
# curl -LsSf https://astral.sh/uv/install.sh | sh
uv sync
uv run ray start --head- Create
.envfile underMoETTAdirectory.
RAY_ADDRESS=<YOUR RAY SERVER ADDRESS> # Get it by running `uv run ray start --head`
WANDB_API_KEY=<YOUR WANDB API KEY>
WANDB_BASE_URL=<YOUR WANDB SERVER URL> # If you are not using wandb-local, then fill it with `https://api.wandb.ai`- Tailor the environment configuration to yours.
The base configuration is located at config/config.py, where the configuration related to path needed to be changed according to your environment.
# Run an experiment locally, i.e., without ray
uv run main.py base --env.local
# Run an experiment with wandb offline
uv run main.py base --env.wandb_mode offline
# Run a hyper-parameter tuning/sweep by designating search space configuration
uv run main.py base --tune.search_space /home/fx25/workspace/MoETTA/config/search_space/seed.yaml
uv run main.py base --algo.algorithm eata
uv run main.py base --algo.algorithm moetta
uv run main.py base --algo.algorithm moetta --data.corruption potpourri+
Base configuration is located at config/config.py.
Derived configuration can be stored in config/subconfigs/ and config/subconfigs/potpourri.py serves as an example.
To add a configuration, only two things need to be done:
- Add a configuration file into
config/subconfigs/ - Import the added file into
config/__init__.py
@inproceedings{DBLP:conf/aaai/FanJCHCJZTW26,
author = {Xiao Fan and
Jingyan Jiang and
Zhaoru Chen and
Fanding Huang and
Xiao Chen and
Qinting Jiang and
Bowen Zhang and
Xing Tang and
Zhi Wang},
editor = {Sven Koenig and
Chad Jenkins and
Matthew E. Taylor},
title = {MoETTA: Test-Time Adaptation Under Mixed Distribution Shifts with
MoE-LayerNorm},
booktitle = {Fortieth {AAAI} Conference on Artificial Intelligence, Thirty-Eighth
Conference on Innovative Applications of Artificial Intelligence,
Sixteenth Symposium on Educational Advances in Artificial Intelligence,
{AAAI} 2026, Singapore, January 20-27, 2026},
pages = {21011--21019},
publisher = {{AAAI} Press},
year = {2026},
url = {https://doi.org/10.1609/aaai.v40i25.39243},
doi = {10.1609/AAAI.V40I25.39243},
timestamp = {Fri, 27 Mar 2026 17:13:39 +0100},
biburl = {https://dblp.org/rec/conf/aaai/FanJCHCJZTW26.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}