Parameter-free conformal martingales detect distribution shifts at unknown times
_onionesque · x · 2026-10-22
A new arXiv paper by Tuo Liu, Drew Prinster and Francesco Orabona removes the hyperparameter burden from conditional conformal test martingales for online distribution-shift detection.
- Problem: prior penalized betting functions rely on smoothing, clipping and other hyperparameters and adapt slowly when a shift follows a long stationary period.
- Key idea: reframe penalized betting as portfolio selection in a three-asset market, using a restarting universal portfolio with heavy-tailed wealth allocation over time — no hyperparameters, provably fast adaptation to shifts at unknown times.
- Results: anytime false-alarm control plus optimal finite-sample bounds on detection delay; an FFT-based implementation cuts complexity from O(T³) to O(T²ln T).
- Experiments: on synthetic and real data, detection delay depends far less on shift time than prior baselines.
More from Research
- Michael Levin polls the public: how do physical, living, and minds overlap? — drmichaellevin · 2026-10-22
- ICLR reviewer: 2 of 3 assigned papers are complete AI slop, worst batch ever seen — AkariAsai · 2026-10-22
- Training-free 20B model Viveki generates real-time video heatmaps under 150ms — not-ekalabya · 2026-10-22
- JD paper proposes ROAD MOPD: cross-teacher guarding to fix gradient conflicts in multi-teacher distillation — teortaxesTex · 2026-10-22
- A complete math roadmap for machine learning: linear algebra, calculus, probability — TivadarDanka · 2026-10-22
- Microsoft's ScholarEvolve lets agents improve by reading papers, lifting Qwen's AppWorld score from 69% to 81.4% — bendee983 · 2026-10-22