Research map / Probabilistic and scientific modeling / Molecules and proteins
Analyzing the Variety Loss in the Context of Probabilistic Trajectory Prediction
Luca Anthony Thiede, Pratik Prabhanjan Brahma
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| square and renormalise the learned PDF to undo MoN dilation | MoN-trained toy generative model · Toy mixture of Gaussians | Jensen-Shannon divergence (lower is better): 0.1282 -> 0.0191 | Improved |
| apply compensation transformation to the marginalized PDF | LSTM MoN generator · NGSIM time step 1 | marginalized log likelihood: -5.48 -> -5.28 | Improved |
| apply compensation transformation to the marginalized PDF | LSTM MoN generator · NGSIM time step 2 | marginalized log likelihood: -5.46 -> -5.13 | Improved |
| apply compensation transformation to the marginalized PDF | LSTM MoN generator · NGSIM time step 3 | marginalized log likelihood: -5.50 -> -5.12 | Improved |
| apply compensation transformation to the marginalized PDF | Social-GAN · Zara1 time step 1 | marginalized log likelihood: -5.57 -> -5.24 | Improved |
| apply compensation transformation to the marginalized PDF | Social-GAN · Zara1 time step 2 | marginalized log likelihood: -5.87 -> -5.77 | Improved |
| apply compensation at later time steps t>=3 and harder pedestrian data | Social-GAN · Zara1 time step 3+ | marginalized log likelihood | Hurt |
| start MoN training with low N and slowly increase to 128 | MoN-trained toy generative model · Toy mixture of Gaussians | learned PDF quality | Improved |
| increase N in the MoN loss | MoN loss, 1-D PDF family · Toy PDF family exponent search | exponent k minimising MoN: k approaches 0.5 as N grows | Improved |
| use 10-dimensional PDF instead of 1-D | MoN loss · Toy PDF family exponent search | exponent k minimising MoN: k close to 0.5 even for small N | Mixed |
Closest papers by meaning
- Generative Active Learning for Long-tail Trajectory Prediction via Controllable Diffusion Model (ICCV 2025 · similarity 0.87)
- Diverse Trajectory Forecasting with Determinantal Point Processes (ICLR 2020 · similarity 0.87)
- Likelihood-Based Diverse Sampling for Trajectory Forecasting (ICCV 2021 · similarity 0.87)
- On Exposing the Challenging Long Tail in Future Prediction of Traffic Actors (ICCV 2021 · similarity 0.86)
- LaKD: Length-agnostic Knowledge Distillation for Trajectory Prediction with Any Length Observations (NeurIPS 2024 · similarity 0.85)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
