The Representational Limit of Scalar Interactions: An Interventional Decomposition
2026-06-19T07:23:56Z•a63d1f70aec04af98a46a470385079d155d58926f63d9f95bab974c9764e1045
GPU-reliabilityLLM-evaluationLayerNormactive-learningbayesian-mixturescausal-inferencecomputational-identifiabilityconformal-predictioncontextual-banditsfeature-interactionsfederated-learninginterpretabilitymissing-not-at-randommodel-auditingmodel-diagnosticsoff-policy-evaluationpredict-then-optimizeprivacyreinforcement-learningset-membership-estimationsolver-free-trainingstructural-reliabilitysurvival-analysistransformers
What happened
Collection of new STAT/ML preprints (arXiv 2026-06-19) covering methods for interpretability, optimization-free training, federated Bayesian inference, LLM evaluation auditing, bandits with bounded noise, conformal-certified active failure-probability estimation, off-policy evaluation under MNAR rewards, theory of training/generalization, survival prediction for GPU failures, computational identifiability, and a LayerNorm-only diagnostic for dead directions in transformers. Highlights: Stochastic Hi‑Fi provides an interventional U/R/S decomposition for feature interactions with finite-sample/
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_stat_ml
- Record identifier
- a63d1f70aec04af98a46a470385079d155d58926f63d9f95bab974c9764e1045
- Enrichment time
- 2026-06-19T07:23:56Z
- 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.