USC paper proposes a Zeta Law for predicting when biomedical data is enough for discovery
PTenigma · x · 2026-11-17
A new paper by Paul Thompson (USC) proposes a unified framework for cross-modal discoverability based on the spectral structure of data, signals, and their alignment, tackling the question of how much data is enough for reliable scientific discovery.
- As biomedical datasets reach millions of individuals and AI models billions of parameters, deciding when more data meaningfully improves effect detection still relies on trial-and-error benchmarking.
- Existing scaling laws and sample complexity theory fail to explain the highly variable gains across domains, especially in biomedical settings aligning high-dimensional signals across heterogeneous modalities.
- The work derives a "double-zeta law" for AUC as a function of sample size N and scan length T (involving the Hurst exponent and spectral horizon), featuring the Riemann zeta function.
- The author also proposes applying DMD modes as frozen encoders to the large-N ENIGMA cross-disorder fMRI datasets using Schaefer-400 parcels plus subcortical/cerebellum regions.
More from Research
- NVIDIA open-sources Nemotron-Labs-Sigma: continuous diffusion LLMs in 14 checkpoints at 3B/8B — ArashVahdat · 2026-11-17
- Layer-hot decoding experiment on M4 fails: negative-result spec decoding writeup — azterizm · 2026-11-17
- ByteDance-backed team open-sources Axion1.0: two-stage SFT+RL agent training beats GPT-6 on HLE — 机器之心 · 2026-11-17
- Scientific American editor builds open-source Deep Time Explorer with Claude — anselm · 2026-11-17
- E11 Bio hiring ML scientist to automate proofreading of petabyte-scale brain connectomics data — anderssandberg · 2026-11-17
- 144,000 probing evals show Act@100 and LN2@60 beat last-layer probing in transformers — CatAstro_Piyush · 2026-11-17