The Impact of Negative Sampling on Contrastive Structured World Models
arXiv:2107.11676
Abstract
World models trained by contrastive learning are a compelling alternative to autoencoder-based world models, which learn by reconstructing pixel states. In this paper, we describe three cases where small changes in how we sample negative states in the contrastive loss lead to drastic changes in model performance. In previously studied Atari datasets, we show that leveraging time step correlations can double the performance of the Contrastive Structured World Model. We also collect a full version of the datasets to study contrastive learning under a more diverse set of experiences.
This work appeared at the ICML 2021 Workshop: Self-Supervised Learning for Reasoning and Perception
References in corpus (7)
- Variational Graph Auto-Encoders
- MONet: Unsupervised Scene Decomposition and Representation
- Learning Robust Representations via Multi-View Information Bottleneck
- On the Binding Problem in Artificial Neural Networks
- A Perspective on Objects and Systematic Generalization in Model-Based RL
- Slot Contrastive Networks: A Contrastive Approach for Representing Objects
- Learning Object-Centric Video Models by Contrasting Sets