This repository contains an implementation of the model introduced in The Traveling Observer Model: Multi-task Learning Through Spatial Variable Embeddings, published with a spotlight presentation at ICLR 2021. Spotlight presentation: https://iclr.cc/virtual/2021/poster/3007.
The model is defined at traveling_observer_model.py.
A script is also provided, demonstrating how the implementation can be applied:
python train_concentric_hyperspheres.py
The script is applied to the Concentric Hyperspheres problem, where TOM has shown the most striking advantage over alternative methods.
This code was developed with Python version 3.7.12. See requirements.txt for the full python environment, which can be installed with
pip install -r requirements.txt