Fisher-Geometric Diffusion in Stochastic Gradient Descent: Optimal Rates, Oracle Complexity, and Information-Theoretic Limits

2026-03-04T20:00:20Z41b479e34343a7244b289af2f9a280e058e9825edeb8da224d5e7d5759dd993f
Finsler-geometryFisher-informationGodambe-matrixGromov-WassersteinPOMDPs','belief-space-metrics','clustering','elbow-method','UMAPSHAPSPD-matricescategorical-datacausal-inferencecombinatorial-algorithmsconformal-predictiondensity-clusteringdifferential-geometrydiffusion-approximationfunctional-ANOVAgeneralized-Bayesgraphslow-degree-methodmachine-learningmartingale-posteriorrobust-subspace-recoverysparse-PCAstatisticsstochastic-gradient-descentuncertainty-quantification

What happened

Batch of newly announced arXiv papers (stat.ML) covering theoretical and methodological advances in high-dimensional statistics, optimization, and machine learning. Highlights include: a Fisher-/Godambe-structured diffusion approximation for SGD yielding sharp oracle complexity and minimax rates; a conformal prediction framework for graph-valued outputs using Z‑Gromov–Wasserstein distances; new geometric (Finsler/dual) structures on the SPD matrix cone via James’ bicone; a counterexample showing failure of the low-degree method for robust subspace recovery alongside a polynomial-time anti-conc

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_stat_ml
Record identifier
41b479e34343a7244b289af2f9a280e058e9825edeb8da224d5e7d5759dd993f
Enrichment time
2026-03-04T20:00:20Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.