A new arXiv paper develops projection-free algorithms for stochastic optimization problems whose objectives are nested compositions of smooth functions over a closed convex decision set. The work targets the multi-level compositional setting, where gradients must be estimated through several layers of functions. It aims to avoid costly projection steps while still handling constraints.
A new arXiv paper presents a deterministic algorithm for online inverse linear optimization that attains O(d) regret, a bound that holds uniformly across the time horizon. Each round requires O(d^2) computation time. The result matches a recent bound by Dewasurendra and addresses an open question raised by Gollapudi and colleagues.
A new arXiv preprint introduces an algorithm for approximate quantum state preparation that is built on Lie algebra structures. The method combines a quantum neural network approach with the SRBB (standard recursive block-by-block?) decomposition, though the abstract excerpt is truncated. The work targets a problem considered fundamental across many areas of quantum computing.
A new arXiv paper presents Meta-LinEXP3, an algorithm that brings meta-learning to adversarial linear contextual bandits, a setting where earlier results largely covered only stochastic or non-contextual variants. The method uses an online-within-online framework to transfer knowledge across a sequence of bandit tasks.
An arXiv preprint introduces CAST, a data structure for building approximate Cholesky factorizations of graph-structured matrices. It targets settings where many linear systems share a single Laplacian or SDDM coefficient matrix, which arises in tasks like diffusion estimation, ranking, and semi-supervised learning. The method aims to reduce the cost of repeated solves in large-scale graph processing pipelines.
A newly posted arXiv paper introduces a kernel-based modular discriminant analysis framework targeting the small-sample-size problem in machine learning, where labeled data are scarce due to cost, accessibility, or ethical constraints. The work addresses limitations of existing methods that struggle to perform reliably when training examples are limited.
A new arXiv preprint presents a DIRECT-type algorithm aimed at extending deterministic global optimization to medium-scale black-box problems. Classic DIRECT methods are mostly effective in low dimensions and tend to converge slowly, so the authors focus on making the framework practical at larger scales. The method retains the balanced exploration-exploitation behavior that made DIRECT widely used.
A new paper proves a hardness limit for consistent submodular maximization, where an algorithm keeps at most k elements while the ground set grows over time. It shows that beating the 2-√2 approximation ratio would force either exponentially many queries or recourse that is linear in the number of arrivals. The result sets a boundary on how much solution quality and stability can be improved at once under a monotone submodular objective.
A new arXiv paper introduces Probabilistic Focal Search, a variant of Focal Search that aims to cut the effort needed to find solutions within a bounded factor of optimal. The method advances lower bounds to guide which frontier nodes are explored, replacing the deterministic selection policy used in standard Focal Search. The work targets heuristic search settings where near-optimal solutions are acceptable in exchange for faster computation.