Research map / Probabilistic and scientific modeling / Molecules and proteins
Inferring the Future by Imagining the Past
Kartik Chandra, Tony Chen, Tzu-Mao Li, Jonathan Ragan-Kelley, Joshua B. Tenenbaum
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 |
|---|---|---|---|
| Use backward-sampled importance sampling instead of rejection sampling (true posterior = Ours@1k) | Posterior inference with 10 samples · Quantitative benchmark | Benchmark suite average TV: 0.063 -> 0.0257 | Improved |
| Use the proposed sampler instead of rejection sampling (true posterior = Rejection@10k) | Posterior inference with 10 samples · Quantitative benchmark | Keys (observed holding no key) TV: 0.672 -> 0.410 | Improved |
| Allow uniform prior over start state instead of two fixed entryways | Proposed sampler · Quantitative benchmark | Grid TV vs Ours@1k: 0.159 -> 0.0538 | Improved |
| Proposed sampler versus rejection sampling on multi-stage keys task (pink key) | Proposed sampler · Quantitative benchmark | Keys TV vs Rejection@10k: 0.456 -> 0.382 | Improved |
| Proposed sampler versus rejection sampling on green key | Proposed sampler · Quantitative benchmark | Keys TV vs Rejection@10k: 0.463 -> 0.312 | Improved |
| Increase rejection sampling to 1,000 samples in blocks domain | Rejection sampling · Quantitative benchmark | Valid samples | No clear effect |
| Use rejection sampling with 10,000 samples as reference on Blocks | Rejection sampling · Quantitative benchmark | Convergence | Hurt |
| Sample paths backward in time and only through the observed state | Proposed sampler · Quantitative benchmark | Qualitative posterior vs humans (N=200) | Improved |
| Compare likelihoods of proposed sampler to rejection sampling at 25,000 samples each, with and without bidirectional path tracing (ablation) | Proposed sampler · 4x4 gridworld | Likelihood per cell, 4x4 gridworld | No clear effect |
| Apply sampler to continuous cart-pole with learned reverse dynamics | Proposed sampler · Cart-pole | Velocity posterior | Improved |
Closest papers by meaning
- MARPLE: A Benchmark for Long-Horizon Inference (NeurIPS 2024 · similarity 0.86)
- Imagination Helps Visual Reasoning, But Not Yet in Latent Space (ICML 2026 · similarity 0.83)
- Episodic Memories Generation and Evaluation Benchmark for Large Language Models (ICLR 2025 · similarity 0.83)
- FIction: 4D Future Interaction Prediction from Video (CVPR 2025 · similarity 0.83)
- Evolving Minds: Logic-Informed Inference from Temporal Action Patterns (ICML 2025 · similarity 0.83)
- Overcoming Multi-step Complexity in Multimodal Theory-of-Mind Reasoning: A Scalable Bayesian Planner (ICML 2025 · similarity 0.82)
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.
