mlsanityData to 7 Oct 2026

The Impact of Negative Sampling on Contrastive Structured World Models

arXiv 2107.11676 ↗First posted 24 Jul 2021cs.LGcited by 2

On the map: World models → Latent dynamics models and model-based RL

Abstract (arXiv)

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.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Latent dynamics models and model-based RL, one of 11 directions in World models. Builds on 25 of its 33 references that are in the map. First posted on arXiv 24 Jul 2021.

Builds on

25 of 33 references.

References · 25

  1. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 11,979
  2. 2018Representation Learning with Contrastive Predictive Codingcited by 7,662
  3. 2013Efficient Estimation of Word Representations in Vector Spacecited by 4,346
  4. 2016Asynchronous Methods for Deep Reinforcement Learningcited by 2,218
  5. 2016node2vec: Scalable Feature Learning for Networkscited by 1,913
  6. 2018Unsupervised Feature Learning via Non-Parametric Instance-level Discriminationcited by 1,838
  7. 2016Variational Graph Auto-Encoderscited by 1,373
  8. 2018Learning deep representations by mutual information estimation and maximizationcited by 1,265
  9. 2020CURL: Contrastive Unsupervised Representations for Reinforcement Learningcited by 745
  10. 2017Time-Contrastive Networks: Self-Supervised Learning from Videocited by 453
  11. 2019MONet: Unsupervised Scene Decomposition and Representationcited by 388
  12. 2019Multi-Object Representation Learning with Iterative Variational Inferencecited by 359
13 more
  1. 2014Discriminative Unsupervised Feature Learning with Exemplar Convolutional Neural Networkscited by 299
  2. 2020On the Binding Problem in Artificial Neural Networkscited by 230
  3. 2019GENESIS: Generative Scene Inference and Sampling with Object-Centric Latent Representationscited by 227
  4. 2019Contrastive Learning of Structured World Modelscited by 206
  5. 2020Learning Robust Representations via Multi-View Information Bottleneckcited by 189
  6. 2019Entity Abstraction in Visual Model-Based Reinforcement Learningcited by 110
  7. 2017Visual Interaction Networkscited by 94
  8. 2020Plannable Approximations to MDP Homomorphisms: Equivariance under Actionscited by 61
  9. 2018Reasoning About Physical Interactions with Object-Oriented Prediction and Planningcited by 51
  10. 2019Structured Object-Aware Physics Prediction for Video Modeling and Planningcited by 42
  11. 2020Learning Object-Centric Video Models by Contrasting Setscited by 19
  12. 2019A Perspective on Objects and Systematic Generalization in Model-Based RLcited by 14
  13. 2020Slot Contrastive Networks: A Contrastive Approach for Representing Objectscited by 9

Built on by

Cited by year

Citing papers per year: 2022: 2.

2 papers.

  1. 2022Toward Compositional Generalization in Object-Oriented World Modelingcited by 22
  2. 2022Binding Actions to Objects in World Modelscited by 9

Similar, not cited

Close in wording, no citation either way

  1. 2026Contrastive World Models
  2. 2023ReCoRe: Regularized Contrastive Representation Learning of World Model
  3. 2025Learning Transformer-based World Models with Contrastive Predictive Coding
  4. 2025Zero-shot World Models via Search in Memory
  5. 2022Contrastive Unsupervised Learning of World Model with Invariant Causal Features
  6. 2025Can World Models Benefit VLMs for World Dynamics?
  7. 2025Simple, Good, Fast: Self-Supervised World Models Free of Baggage
  8. 2025EDELINE: Enhancing Memory in Diffusion-based World Models via Linear-Time Sequence Modeling