![[2512.19223] Phase-space entropy at acquisition reflects downstream learnability](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2512.19223] Phase-space entropy at acquisition reflects downstream learnability
Summary
The paper explores how phase-space entropy at the acquisition stage can predict the learnability of downstream models, offering a new metric for evaluating data acquisition strategies across various domains.
Why It Matters
Understanding how data acquisition affects model performance is crucial for improving machine learning systems. This study introduces a novel metric that can guide the selection of sampling strategies, potentially enhancing efficiency and effectiveness in diverse applications such as MRI and image classification.
Key Takeaways
- Introduces a new metric, ΔS_B, for quantifying information preservation during data acquisition.
- Demonstrates that phase-space entropy can predict downstream learnability without prior training.
- Applies findings to various fields, including masked image classification and accelerated MRI.
- Suggests minimizing ΔS_B for optimal sampling strategies, enhancing model performance.
- Highlights the importance of understanding data structure before model training.
Computer Science > Machine Learning arXiv:2512.19223 (cs) [Submitted on 22 Dec 2025 (v1), last revised 20 Feb 2026 (this version, v2)] Title:Phase-space entropy at acquisition reflects downstream learnability Authors:Xiu-Cheng Wang, Jun-Jie Zhanga, Nan Cheng, Long-Gang Pang, Taijiao Du, Deyu Meng View a PDF of the paper titled Phase-space entropy at acquisition reflects downstream learnability, by Xiu-Cheng Wang and 5 other authors View PDF HTML (experimental) Abstract:Modern learning systems work with data that vary widely across domains, but they all ultimately depend on how much structure is already present in the measurements before any model is trained. This raises a basic question: is there a general, modality-agnostic way to quantify how acquisition itself preserves or destroys the information that downstream learners could use? Here we propose an acquisition-level scalar $\Delta S_{\mathcal B}$ based on instrument-resolved phase space. Unlike pixelwise distortion or purely spectral errors that often saturate under aggressive undersampling, $\Delta S_{\mathcal B}$ directly quantifies how acquisition mixes or removes joint space--frequency structure at the instrument scale. We show theoretically that \(\Delta S_{\mathcal B}\) correctly identifies the phase-space coherence of periodic sampling as the physical source of aliasing, recovering classical sampling-theorem consequences. Empirically, across masked image classification, accelerated MRI, and massive MIMO (inclu...
