CMU statisticians unveil COSI, a framework for robust inference under weaker assumptions
_onionesque · x · 2026-11-20
arXiv paper "COSI: A Framework for Robustly Valid Inference" by Arun K. Kuchibhotla, Sivaraman Balakrishnan and Larry Wasserman (CMU).
- COSI (COnfidence from Scale Invariance) builds on HulC (Hull-based Confidence), using median unbiasedness to construct confidence sets from batch estimates
- Core idea: split data into batches, compute an estimate per batch, and retain candidate parameter values whose batch estimates pass a scale-invariance test
- Asymptotically valid under weaker assumptions than classical Wald/bootstrap, with small-sample coverage closer to nominal levels
- Applications include robust inference for means, regression coefficients in misspecified linear models, tensor norms, and Fréchet means
More from Research
- RUC & Fudan release CoworkOS: agent runtime for long-horizon work with human edits, boosting benchmarks from 33.19 to 42.95 — jiqizhixin · 2026-11-20
- PostTrainBench Author: 10-Hour GPU Cap Explains Benchmark Hill-Climbing — maksym_andr · 2026-11-20
- Team claims Codex cracked 35 open Erdős problems in 2 months, some Lean-verified — IgorCarron · 2026-11-20
- ForgeSystem paper proposes AI agents forging demand-specific software — Div_pradeep · 2026-11-20
- Accord: open-source dense metric mapping from a single camera and IMU — BE2R · 2026-11-20
- AltRA-Test: ~100 samples suffice to decide if LLM judges can replace humans — Technion · 2026-11-20