Research map / Probabilistic and scientific modeling / Molecules and proteins
Generalized Belief Transport
Junqi Wang, PEI WANG, Patrick Shafto
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 |
|---|---|---|---|
| learner estimates eta empirically from current data (update) instead of having access to the true hypothesis (exact) (ablation) | GBT learner · 10x10 and 5x3 matrices, convergence rounds | episodes reaching 0.999 threshold by round | Improved |
| increase number of data outcomes (dimension of D) with fixed number of hypotheses | GBT learner · m x 10 rectangular matrices, m in 5..100 | convergence speed | Improved |
| scale square problem size m from 10 to 100 | GBT learner · m x m matrices | convergence | No clear effect |
| vary eps_eta from 0 to infinity with eps_P=1, eps_theta=infinity (ablation) | GBT learner · 5x5 problem, 400,000 episodes | limit distribution over hypotheses | Mixed |
| set eps_theta=0 so the column-sum constraint no longer affects the transport plan (ablation) | GBT learner · theory (Proposition 14) | posterior convergence | Hurt |
| replace sudden Bayesian-to-discriminative jump with gradual interpolation (Orange path) (ablation) | GBT learner · 4x4 matrix, 10 samples | conclusiveness (mean): 0.9406 -> 0.9964 | Improved |
| replace sudden jump with parabola interpolation (Red path) (ablation) | GBT learner · 4x4 matrix, 10 samples | conclusiveness (mean): 0.9406 -> 0.9834 | Improved |
| replace sudden jump with gradual interpolation (Orange path) (ablation) | GBT learner · 4x4 matrix, 10 samples | posterior entropy (mean): 0.1261 -> 0.0079 | Improved |
| replace sudden jump with parabola interpolation (Red path) (ablation) | GBT learner · 4x4 matrix, 10 samples | posterior entropy (mean): 0.1261 -> 0.0388 | Improved |
| use general GBT (eps in cube interior) instead of Bayesian or SCBI learner in periodically changing world | GBT learner · 3x3 matrix, period 18, 400 steps | detection of periodicity in posterior | Improved |
| vary radius and shape of the true hypothesis path | GBT learner eps=(1,10,10) · period 20, 300 steps | posterior to hypothesis path area ratio: 0.1620 -> 0.1616 | No clear effect |
Closest papers by meaning
- Probability-Polarized Optimal Transport for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.84)
- Partial Transportability for Domain Generalization (NeurIPS 2024 · similarity 0.84)
- Model-Based Transfer Learning for Contextual Reinforcement Learning (NeurIPS 2024 · similarity 0.83)
- Multi-Agent Learning under Uncertainty: Recurrence vs. Concentration (NeurIPS 2025 · similarity 0.83)
- Preferential Normalizing Flows (NeurIPS 2024 · similarity 0.83)
- SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training (ICML 2025 · similarity 0.83)
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.
