2021/07/24 by Ondrej Biza, Ondřej Bíža, Biza, Ondrej +4
Computer Science · #Artificial intelligence #Autoencoder #Computer science #Computer vision #Contrastive analysis #Deep learning #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Linguistics #Machine Learning (cs.LG) #Machine learning #Natural language processing #Sample (material) #Sampling (signal processing) #Set (abstract data type) #Topic Modeling #cs.LG
paper · pdf · doi:10.48550/arxiv.2107.11676
published in arXiv (Cornell University) (Cornell University) · This work appeared at the ICML 2021 Workshop: Self-Supervised Learning for Reasoning and Perception
arxiv created 2021/07/24 · openalex publication_date 2021/07/24 · arxiv updated 2021/07/27 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
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.