Graph Conditional Variational Models: Too Complex for Multiagent Trajectories?

Research output: Journal contributionsConference article in journalResearchpeer-review

Authors

Recent advances in modeling multiagent trajectories combine graph architectures such as graph neural networks (GNNs) with conditional variational models (CVMs) such as variational RNNs (VRNNs). Originally, CVMs have been proposed to facilitate learning with multi-modal and structured data and thus seem to perfectly match the requirements of multi-modal multiagent trajectories with their structured output spaces. Empirical results of VRNNs on trajectory data support this assumption. In this paper, we revisit experiments and proposed architectures with additional rigour, ablation runs and baselines. In contrast to common belief, we show that prior results with CVMs on trajectory data might be misleading. Given a neural network with a graph architecture and/or structured output function, variational autoencoding does not seem to contribute statistically significantly to empirical performance. Instead, we show that well-known emission functions do contribute, while coming with less complexity, engineering and computation time.
Original languageEnglish
JournalProceedings of Machine Learning Research
Volume137
Pages (from-to)136-147
Number of pages12
ISSN2640-3498
Publication statusPublished - 2020
Event34rd Conference on Neural Information Processing Systems - NeurIPS 2020: Neural Information Processing Systems Online Conference 2020 - digital
Duration: 06.12.202012.12.2020
Conference number: 34
https://neurips.cc/virtual/2020/public/index.html
https://proceedings.mlr.press/v137/

Bibliographical note

Publisher Copyright:
© Proceedings of Machine Learning Research 2020.

Recently viewed

Publications

  1. Neural Combinatorial Optimization on Heterogeneous Graphs
  2. Selection and Recognition of Statistically Defined Signals in Learning Systems
  3. Fostering Circularity: Building a Local Community and Implementing Circular Processes
  4. Binary Random Nets I
  5. Detection and mapping of water pollution variation in the Nile Delta using multivariate clustering and GIS techniques
  6. Modeling Effective and Ineffective Knowledge Communication and Learning Discourses in CSCL with Hidden Markov Models
  7. Fixed-term Contracts and Wages Revisited Using Linked Employer-Employee Data from Germany
  8. Towards improved dispatching rules for complex shop floor scenarios - A genetic programming approach
  9. FFTSMC with Optimal Reference Trajectory Generated by MPC in Robust Robotino Motion Planning with Saturating Inputs
  10. On the Nonlinearity Compensation in Permanent Magnet Machine Using a Controller Based on a Controlled Invariant Subspace
  11. 'SPREAD THE APP, NOT THE VIRUS’ – AN EXTENSIVE SEM-APPROACH TO UNDERSTAND PANDEMIC TRACING APP USAGE IN GERMANY
  12. Simultaneous Constrained Adaptive Item Selection for Group-Based Testing
  13. Analyzing different types of moderated method effects in confirmatory factor models for structurally different methods
  14. Dynamic Lot Size Optimization with Reinforcement Learning
  15. Design and characterization of an EOG signal acquisition system based on the programming of saccadic movement routines
  16. Methodologies for Noise and Gross Error Detection using Univariate Signal-Based Approaches in Industrial Application
  17. Modelling tasks—The relation between linguistic skills, intra-mathematical skills, and context-related prior knowledge
  18. Finding Similar Movements in Positional Data Streams
  19. Multidimensional Cross-Recurrence Quantification Analysis (MdCRQA)–A Method for Quantifying Correlation between Multivariate Time-Series
  20. Lagged Multidimensional Recurrence Quantification Analysis for Determining Leader–Follower Relationships Within Multidimensional Time Series
  21. Mapping interest rate projections using neural networks under cointegration
  22. Switching from a Managing to a Monitoring Function on the Board
  23. Towards a Dynamic Interpretation of Subjective and Objective Values
  24. Performance analysis for loss systems with many subscribers and concurrent services
  25. Using mixture distribution models to test the construct validity of the Physical Self-Description Questionnaire
  26. Effectiveness of a guided multicomponent internet and mobile gratitude training program - A pragmatic randomized controlled trial
  27. Study on the effects of tool design and process parameters on the robustness of deep drawing
  28. A Proposal for Integrating Theories of Complexity for Better Understanding Global Systemic Risks
  29. Model-based logistic controlling of converging material flows
  30. A Cross-Classified CFA-MTMM Model for Structurally Different and Nonindependent Interchangeable Methods
  31. Anomaly detection in formed sheet metals using convolutional autoencoders