Research map / Probabilistic and scientific modeling / Molecules and proteins
Generating Correct Answers for Progressive Matrices Intelligence Tests
Niv Pekar, Yaniv Benny, Lior Wolf
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 |
|---|---|---|---|
| Replace latent reparameterization with concatenated random noise (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 60.1 | Hurt |
| Replace Dynamic Selective KLD with standard KLD (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 53.3 | Hurt |
| Apply KLD to a fixed half of the latent dimensions (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 55.8 | Hurt |
| Remove the VAE and train the generator with a discriminator (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 55.6 | Hurt |
| Remove auxiliary rule-type prediction (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 57.2 | Hurt |
| Replace contrastive relation loss with target-only MSE (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 52.7 | Hurt |
| Compare generated and choice image features instead of relation-wise encodings (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 48.4 | Hurt |
| Compare generated and choice images pixel-wise (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 48.1 | Hurt |
| Remove selective freezing during perceptual-loss backpropagation (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 44.6 | Hurt |
| Remove randomness during test-time generation (ablation) | Full answer generator · PGM neutral | accuracy (MRNet evaluator): 65.4 -> 65.5 | No clear effect |
Closest papers by meaning
- Towards Generative Abstract Reasoning: Completing Raven’s Progressive Matrix via Rule Abstraction and Selection (ICLR 2024 · similarity 0.87)
- GenVP: Generating Visual Puzzles with Contrastive Hierarchical VAEs (ICLR 2025 · similarity 0.84)
- Attention as a Hypernetwork (ICLR 2025 · similarity 0.81)
- Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions (ICLR 2025 · similarity 0.81)
- XTC: Head-Aware Sampling by Excluding Top Choices (NeurIPS 2026 · similarity 0.80)
- Divide-and-Conquer Predictive Coding: a structured Bayesian inference algorithm (NeurIPS 2024 · similarity 0.80)
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.
