Research map / Probabilistic and scientific modeling / Molecules and proteins
A Geometric Lens on Physics-Aligned Data Compression
Aleix Segui Ugalde, Wesley Armour
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 |
|---|---|---|---|
| increase physics weight beta in the training loss (ablation) | hyperprior and factorised VAE codecs · PDEBench turbulent Navier-Stokes, vorticity observable, fixed rate | vorticity error vs velocity MSE | Mixed |
| increase physics weight beta (ablation) | hyperprior codec · PDEBench CFD, vorticity | Align_k(W,G) for k=8,16,32 | Improved |
| physics-aware training on EM coarse-structure observable (higher alignment) | codec · EM minnie65, coarse structures | MSE degradation: MSE change 0.014 ± 0.024 at 0.64 bps vs 0.445 ± 0.104 for contour features at 0.11 bps | Improved |
| physics-aware training on EM contour-feature observable (low alignment) | codec · EM minnie65, contour features | tradeoff between MSE and physics error: Align16 0.048 ± 0.092 at 0.11 bps | Hurt |
| physics-aware loss with beta>0 (ablation) | hyperprior model · PDEBench CFD, vorticity, equal PSNR | rate (BD-rate) | Improved |
| compare second-order surrogate models to measured rate and distortion under latent noise | local tangent-space model · PDEBench CFD, moderate latent noise | R2 of quadratic approximations | Improved |
Closest papers by meaning
- SAFE-SVD: Sensitivity-Aware Fidelity-Enforcing SVD for Physics Foundation Models (NeurIPS 2026 · similarity 0.89)
- MeLLoC: Lossless Compression with High-order Mechanism Learning (NeurIPS 2024 · similarity 0.85)
- When Less is More: The LLM Scaling Paradox in Context Compression (NeurIPS 2026 · similarity 0.84)
- Optimal Neural Compressors for the Rate-Distortion-Perception Tradeoff (NeurIPS 2025 · similarity 0.84)
- Fix the Structural Bottleneck: Context Compression via Explicit Information Transmission (NeurIPS 2026 · similarity 0.84)
- OmniArch: Building Foundation Model for Scientific Computing (ICML 2025 · similarity 0.84)
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.
