Permutation-Equivariant 2D State Space Models: Theory and Canonical Architecture for Multivariate Time Series
2026-03-11T07:24:02Z•6cdacca7743a08cb23202472dc068a9b8541f267f4c97e1c5322f8ca41919725
AI alignmentBayesian neural networksConant-AshbyMidicothPPMVI-2D-SSMalignment verificationbalancing (regularization)causal inferencecompressiondenoisingflow matchinggeneralitygenerative modelshypergraph observersmodel architecturemultivariate time seriesnatural gradientpermutation-equivariantreversible sampling','Metropolis-Hastings','MMD','Thompson-SAMPLsimulation-based inferencesoundnessstate-space modelstractabilitytrilemma
What happened
Collection of 10 recent ML/statistics preprints covering theory and architectures across time-series modeling, AI alignment, compression, simulation-based inference, generative modelling, learning theory, and decision-making under partial compliance. Highlights: a canonical permutation‑equivariant 2D state‑space model (VI 2D SSM / VI 2D Mamba) that removes artificial variable ordering in multivariate time series; a formal trilemma showing no verification procedure can be simultaneously sound, general, and tractable for alignment certification; Midicoth, a micro‑diffusion denoiser to improve P(
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_stat_ml
- Record identifier
- 6cdacca7743a08cb23202472dc068a9b8541f267f4c97e1c5322f8ca41919725
- Enrichment time
- 2026-03-11T07:24:02Z
- 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.