Contrastive Learning of Structured World Models


Summary

This recent paper by Thomas Kipf, Elise van der Pol and Max Welling highlights different cognitive architectures 'how to see' in the Reinforcement Learning setting. Their proposed C-SWM model is composed of a CNN-based objectextractor, an MLP-based object encoder, a GNN-based relational transition model, and an object-factorized contrastive loss.

References

  2. Contrastive Learning of Structured World Models (2020)

    Kipf, Thomas and van der Pol, Elise and Welling, Max

  4. Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (2019)

    Crawford, Eric and Pineau, Joelle

  6. Multi-Object Representation Learning with Iterative Variational Inference (2019)

    Greff, Klaus and Kaufman, Raphaël Lopez and Kabra, Rishabh and Watters, Nick and Burgess, Chris and Zoran, Daniel and Matthey, Loic and Botvinick, Matthew and Lerchner, Alexander

  8. DeepUSPS: Deep Robust Unsupervised Saliency Prediction via Self-Supervision (2019)

    Nguyen, Tam and Dax, Maximilian and Mummadi, Chaithanya Kumar and Ngo, Nhung and Nguyen, Thi Hoai Phuong and Lou, Zhongyu and Brox, Thomas

  10. World Models (2018)

    Ha, David and Schmidhuber, Jürgen