Randomness Extraction from Somewhat Dependent Sources
summary
Defines three dependence models and gives sharp positive and negative results: T-coordinated sources admit deterministic extraction — any two‑source extractor for independent sources with min‑entropy K yields a 2ε extractor for T‑coordinated sources with min‑entropy K+T+log(1/ε), and this K+T+log loss is optimal. The other models (bounded cross‑influence and bounded mutual information) resist extraction but admit condensation: cross‑influence condensers incur a 2^T error blowup and +T deficiency, and those parameters are essentially tight. Also shows sampling inner‑product needs ≥0.49N bits.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:05Natural sources of randomness
- 3:23Seeded extractors
- 5:50Background
- 7:05Models
- 9:35TCoordinated Sources
- 12:13TCross Influence
- 15:37Mutual Information
- 16:54Relation Between Models
- 18:54Extractors and Condensers
- 20:39Condensers
- 23:13Theorem
related talks
21mIncompressiblity and Next-Block Pseudoentropy
47mIncompressiblity and Next-Block Pseudoentropy
31mCryptographic Hardness of Correlated Sampling
47mHardness of LWE on General Entropic Distributions
48mOutput-Compressing Randomized Encodings and Applications
41mReducibility and Statistical-Computational Gaps from Secret Leakage
01h 09mNoisy Connections: A Survey of Interactive Coding and Its Borders with Other Topics
40mOn the Complexity of Two-Party Differential Privacy
