SASA replaces single-vector decoders in SAEs with learned subspaces plus block sparsity and nuclear-norm regularization, proving that a single group becomes the global minimizer once block size meets intrinsic dimension and yielding polynomial rather than exponential sample complexity.
Exact matrix completion via convex optimization
10 Pith papers cite this work, alongside 5,144 external citations. Polarity classification is still indexing.
10
Pith papers citing it
5,144
external citations · OpenAlex
representative citing papers
A T-estimation-based procedure for adaptive density estimation and optimal control in offline contextual MDPs without stationarity, providing oracle risk bounds under two loss functions and finite-sample cost guarantees.
Wasserstein least squares extends Euclidean least squares to distribution-valued responses via convex analysis, yielding n^{-1/2} rates under template deformation and faster barycenter rates than prior work.
A structure-preserving low-rank factorization of 2RDMs achieves linear rank scaling with system size and ~99% compression while retaining chemical accuracy for correlated states.
IMR models an incomplete matrix as intercepts plus Lasso-penalized covariates plus low-rank latent factors with ridge penalties on known kernels, estimated by modular ALS with non-asymptotic error bounds.
Proposes PcovRnnp method enabling simultaneous dimension reduction and regularized coefficient estimation via nuclear norm penalty in high-dimensional settings.
Iteris, an agentic research system, produced evidence and drafts for two open computational math problems that were verified after human correction.
Group RC-DMC extends RC-DMC by adding Set-Transformer group aggregation, low-rank regularization via nuclear-norm proximal steps, and a low-rank decoder to improve group-level RMSE on MovieLens and Goodbooks while staying competitive on precision, recall, and F1.
The paper frames Cayley-table completion as the discrete algebraic analog to matrix completion and poses the open problem of proving exact recovery bounds under flatness priors that favor associativity.
Extends Fano bounds to sufficiency of low conditional entropy and defines a quantum entanglement task for infinite-dimensional systems with bounds via maximal singlet fraction of finite-dimensional approximations.
citing papers explorer
-
Subspace-Aware Sparse Autoencoders for Effective Mechanistic Interpretability
SASA replaces single-vector decoders in SAEs with learned subspaces plus block sparsity and nuclear-norm regularization, proving that a single group becomes the global minimizer once block size meets intrinsic dimension and yielding polynomial rather than exponential sample complexity.
-
Adaptive Estimation and Optimal Control in Offline Contextual MDPs without Stationarity
A T-estimation-based procedure for adaptive density estimation and optimal control in offline contextual MDPs without stationarity, providing oracle risk bounds under two loss functions and finite-sample cost guarantees.
-
Wasserstein Least Squares: A Canonical Regression Method for Probability Distributions
Wasserstein least squares extends Euclidean least squares to distribution-valued responses via convex analysis, yielding n^{-1/2} rates under template deformation and faster barycenter rates than prior work.
-
Low-rank compression of two-electron reduced density matrices
A structure-preserving low-rank factorization of 2RDMs achieves linear rank scaling with system size and ~99% compression while retaining chemical accuracy for correlated states.
-
Incomplete Matrix Regression
IMR models an incomplete matrix as intercepts plus Lasso-penalized covariates plus low-rank latent factors with ridge penalties on known kernels, estimated by modular ALS with non-asymptotic error bounds.
-
Principal Covariate Regression with Nuclear Norm Penalty
Proposes PcovRnnp method enabling simultaneous dimension reduction and regularized coefficient estimation via nuclear norm penalty in high-dimensional settings.
-
Iteris: Agentic Research Loops for Computational Mathematics
Iteris, an agentic research system, produced evidence and drafts for two open computational math problems that were verified after human correction.
-
Rank-Constrained Deep Matrix Completion for Group Recommendation
Group RC-DMC extends RC-DMC by adding Set-Transformer group aggregation, low-rank regularization via nuclear-norm proximal steps, and a low-rank decoder to improve group-level RMSE on MovieLens and Goodbooks while staying competitive on precision, recall, and F1.
-
Open Problem: Separating Geometric and Algorithmic Compression via Cayley-Table Completion
The paper frames Cayley-table completion as the discrete algebraic analog to matrix completion and poses the open problem of proving exact recovery bounds under flatness priors that favor associativity.
-
On the coherent extension of some Fano-type learning bounds
Extends Fano bounds to sufficiency of low conditional entropy and defines a quantum entanglement task for infinite-dimensional systems with bounds via maximal singlet fraction of finite-dimensional approximations.