Spectral Perturbation of the Empirical Fisher Information Matrix under Weight Quantization
2026-06-30T07:23:55Z•e1c024f883c64f19522cd44999871af3bd555dcf2f2377c1638699f2c63f6e2a
Fisher-informationLangevin-Monte-Carloconformal-predictiondiffusion-modelsgaussian-processesgradient-boostingmachine-learningmodel-monitoringneuroconnectivityonline-learningout-of-distribution-detectionquantile-regressionrepresentation-learningruntime-metricssparse-regressionstatistical-learningtransformersuncertainty-quantificationvariance-reductionweight-quantization
What happened
Collection of recent machine-learning/statistics preprints (arXiv, 2026-06-30) covering: (1) spectral perturbation of the empirical Fisher Information Matrix (FIM) under input shifts and weight quantization, including a proposed runtime monitoring statistic sigma_t = lambda_max(F_t)/lambda_base and empirical measurements on 12 models (n=1,080 trajectories); (2) Adaptive Iterative Hard Thresholding (AIHT) for online high-dimensional quantile regression with theoretical regret bounds; (3) variance-reduction analysis for stochastic-gradient generalized non-reversible Langevin Monte Carlo and asym
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_stat_ml
- Record identifier
- e1c024f883c64f19522cd44999871af3bd555dcf2f2377c1638699f2c63f6e2a
- Enrichment time
- 2026-06-30T07:23:55Z
- 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.