PulseAugur / Brief
EN
LIVE 10:48:37

Brief

last 24h
[7/7] 224 sources

Multi-source AI news clustered, deduplicated, and scored 0–100 across authority, cluster strength, headline signal, and time decay.

  1. TOOL · arXiv cs.CV English(EN) ·

    Low-Rank Tensor Completion Based on Fractional Regularization with Ky Fan p-k Norm

    Researchers have introduced a new method for low-rank tensor completion (LRTC) that utilizes a novel nonconvex surrogate called the tensor nuclear norm to tensor Ky Fan p-k norm (TNPK). This approach aims to accurately approximate the tensor tubal rank and offers properties like scale invariance and parameter flexibility. The paper details a LRTC model and proves that low-rank tensors are local minimizers under specific conditions. An efficient algorithm, the alternating direction method of multipliers (ADMM), has been developed for this model, and experimental results show superior performance compared to existing methods. AI

    Low-Rank Tensor Completion Based on Fractional Regularization with Ky Fan p-k Norm
  2. TOOL · arXiv stat.ML English(EN) ·

    On the Relationship Between CoCoA and ADMM for Distributed Empirical Risk Minimization

    A new research paper explores the relationship between two families of distributed optimization algorithms, CoCoA and ADMM. By unifying them through a primal-dual perspective, the study reveals that certain ADMM variants can perform comparably to or better than CoCoA for ridge-regularized empirical risk minimization problems. The unified view also provides a new primal-dual gap stopping criterion for consensus ADMM and a consistent convergence analysis for ADMM-type methods. AI

    On the Relationship Between CoCoA and ADMM for Distributed Empirical Risk Minimization

    IMPACT Provides a unified theoretical framework for distributed optimization, potentially improving efficiency in training large-scale machine learning models.

  3. TOOL · arXiv cs.CV English(EN) ·

    Physics-Aware Linearized ADMM and Its Unrolling

    Researchers have developed a new algorithm called Physics-Aware Linearized ADMM (PA-LADMM) for solving inverse problems in signal processing that involve complex partial differential equations (PDEs). This method simplifies PDE subproblems for more efficient computation, requiring only PDE solver and gradient evaluations per iteration. The algorithm is theoretically guaranteed to converge under specific conditions and has been enhanced with deep unfolding techniques for parameter training. Experiments in optical fiber communication and image restoration validated the effectiveness of PA-LADMM. AI

    Physics-Aware Linearized ADMM and Its Unrolling
  4. TOOL · Hugging Face Daily Papers English(EN) ·

    Convex Low-resource Accent-Robust Language Detection in Speech Recognition

    Researchers have developed a new convex optimization framework called Convex Language Detection (CLD) to improve language identification in speech recognition systems, particularly for low-resource accents and dialects. This method uses efficient ADMM in JAX to achieve global optimality and theoretical guarantees against dialectal variations. CLD demonstrates high accuracy (97-98%) even with limited training data, significantly reducing cross-lingual decoding failures and compute costs compared to traditional approaches. AI

    Convex Low-resource Accent-Robust Language Detection in Speech Recognition

    IMPACT Improves speech recognition equity and efficiency for diverse global accents and dialects.

  5. RESEARCH · arXiv stat.ML English(EN) · · [5 sources]

    Active multiple matrix completion with adaptive confidence sets

    Researchers have developed a novel algorithm for low-rank tensor completion, extending matrix completion techniques using an alternating direction method of multipliers (ADMM) optimization framework. This new method reformulates the problem into subproblems solved iteratively, incorporating over-relaxation and adaptive penalty parameters to enhance convergence and performance. Separately, a new multi-task active learning algorithm called MAlocate has been proposed for simultaneously solving multiple matrix completion problems, adapting to unknown matrix ranks and demonstrating minimax-optimality. AI

    Active multiple matrix completion with adaptive confidence sets

    IMPACT These papers introduce new algorithmic approaches for tensor and matrix completion, potentially improving data imputation and analysis in various machine learning applications.

  6. RESEARCH · arXiv cs.LG English(EN) · · [3 sources]

    Learning Over-Relaxation Policies for ADMM with Convergence Guarantees

    Researchers have developed a method to learn online updates for the relaxation parameter in the Alternating Direction Method of Multipliers (ADMM). This approach aims to improve the performance of ADMM, a technique used in structured convex optimization, by adapting parameters for specific problem classes. The learned policies have demonstrated improvements in both iteration count and execution time on benchmark quadratic programs compared to standard methods. AI

    Learning Over-Relaxation Policies for ADMM with Convergence Guarantees

    IMPACT Potential for faster, more efficient optimization in AI model training and deployment.

  7. RESEARCH · arXiv stat.ML English(EN) · · [4 sources]

    Sparse Network Inference under Imperfect Detection and its Application to Ecological Networks

    Researchers have introduced a novel framework for structured sparse nonnegative low-rank factorization to improve the inference of latent structures in bipartite networks, particularly those used in ecological research. This method addresses limitations in existing models by incorporating detection probability estimation and imposing nonconvex $\ell_{1/2}$ regularization to promote sparsity and better relative scaling. An ADMM-based algorithm was developed to solve the resulting nonconvex and nonsmooth optimization problem, with experiments showing enhanced recovery of latent factors and network structures on both synthetic and real ecological datasets. AI

    Sparse Network Inference under Imperfect Detection and its Application to Ecological Networks