Gregory Valiant

Stanford computer scientist who, with his frequent co-author Paul Valiant, proved two related but distinct results in the statistics of the unseen: a 2011 STOC paper showing a matching O(n/log n) sample complexity for estimating entropy and support size, and a 2015 paper ("Instance Optimal Learning") that independently claims an n·log n-scale extrapolation range for species-count prediction — using an error metric a concurrent rival paper's own comparison calls exponentially weaker in the hardest regime. He matters to this vault because an existing note had compressed both papers, and both of their "optimal" claims, into a single "concurrent" sentence; disentangling them — and correctly crediting the earlier, non-concurrent 2011 result — was the point of the capture that created this page.

References

• claim-valiant-2015-nlogn-range-matches-osw-but-error-metric-exponentially-weaker
• claim-valiant-2011-stoc-paper-is-real-nonconcurrent-nlogn-antecedent
• claim-unseen-mass-is-predictable-only-to-n-log-n (revised in place, 2026-08-25)
• Captures: 10-inbox/raw/2026-08-25-verify-the-nlog-n-unseen-prediction-horizon-and.md