arXiv subjects

Yulin Peng

Publications and source records attributed to Yulin Peng.

6 recordsLinked to original sources

arXiv2026

A New Perspective on Clustering: A Mixed-norm Model and its Solution by Progressive Integer Programming

Extending the classical $K$-means and $K$-medians models, this paper introduces an $\ell_{p,q}$ mixed-norm clustering model where the centroid updates and cluster assignments are under the $\ell_p$ and $\ell_q$ norms, respectively. The model is formulated as a mixed-integer program (MIP) with Heaviside composite constraints that describe the nearest-center assignments. The framework recovers $K$-means and $K$-medians when $p=q=2$ and $p=q=1$, respectively, and yields new models when $p\ne q$. To address the computational challenges, we develop a progressive integer programming (PIP) method that adaptively fixes confident assignments and solves restricted mixed-integer subproblems. For $q=1$, we develop a convex inner approximation of the difference-of-convex constraints in the restricted subproblems, for which a global solution can be computed. Importantly, we establish the connection between the local minimizer and the strong center-local minimizer of the mixed-norm clustering problem and the global optimal solution of the restricted subproblems under certain assumptions. This connection provides a practical certificate of a local minimizer of the nonconvex mixed-norm clustering model. We further develop techniques for constructing adaptive fixing sets and working sets for $q=1$. Extensive numerical experiments demonstrate the superior performance of the mixed-norm clustering model and the efficiency of PIP for solving the MIP model, which may be intractable otherwise. In particular, the $\ell_{2,1}$ mixed-norm clustering model is effective under coordinate-sparse, mean-balanced contamination, whereas the $\ell_{1,2}$ model is preferred under dense coordinatewise Cauchy contamination. The numerical results also show that PIP can escape poor alternating solutions and obtain substantially better feasible clustering, while preserving strong warm starts when no improvement is found.

math.OC↗

arXiv2026

Coxeter Condorcet domains

Condorcet domains are subsets of permutations that ensure pairwise majority voting yields acyclic outcomes, and they form an active area of research at the intersection of social choice theory and combinatorics. In this paper, we extend the theory of Condorcet domains to the broader setting of arbitrary finite Coxeter groups. The core contribution of our approach is the introduction of Condorcet root posets, defined on the chosen root systems. Notably, we establish a natural bijection between closed Condorcet domains and Condorcet root posets, which facilitates the study of Condorcet domains. Using this correspondence, we extend the median graph representation of closed Condorcet domains to arbitrary finite Coxeter groups, demonstrating that these domains can be characterized by the skeletons of their associated Condorcet root posets. These results are novel even in type $A$. Furthermore, these posets give a unified language that efficiently captures a wide range of desirable properties of Condorcet domains, such as being maximal, connected, peak-pit, and of tiling type. Using this framework, we strengthen and generalize several classical results: we establish that a maximal Condorcet domain is connected if and only if it is peak-pit; we prove that the tiling-type property is equivalent to the combination of being maximal and connected, and having maximal width; and we show that strictly positive voting profiles on connected Condorcet domains yield outcomes with only simple ties.

math.CO↗

arXiv2026

SAGE: Multi-Agent Self-Evolution for LLM Reasoning

Reinforcement learning with verifiable rewards improves reasoning in large language models (LLMs), but many methods still rely on large human-labeled datasets. While self-play reduces this dependency, it often lacks explicit planning and strong quality control, limiting stability in long-horizon multi-step reasoning. We present SAGE (Self-evolving Agents for Generalized reasoning Evolution), a closed-loop framework where four agents: Challenger, Planner, Solver, and Critic, co-evolve from a shared LLM backbone using only a small seed set. The Challenger continuously generates increasingly difficult tasks; the Planner converts each task into a structured multi-step plan; and the Solver follows the plan to produce an answer, whose correctness is determined by external verifiers. The Critic scores and filters both generated questions and plans to prevent curriculum drift and maintain training signal quality, enabling stable self-training. Across mathematics and code-generation benchmarks, SAGE delivers consistent gains across model scales, improving the Qwen-2.5-7B model by 8.9% on LiveCodeBench and 10.7% on OlympiadBench.

cs.AI↗

arXiv2026

SEMAG: Self-Evolutionary Multi-Agent Code Generation

Large Language Models (LLMs) have made significant progress in handling complex programming tasks. However, current methods rely on manual model selection and fixed workflows, which limit their ability to adapt to changing task complexities. To address this, we propose SEMAG, a Self-Evolutionary Multi-Agent code Generation framework that mimics human coding practices. It decomposes programming tasks into stages, including planning, coding, debugging, and discussion, while adapting workflows to task difficulty. Its self-evolutionary agents can access the latest models in real time and automatically upgrade the backbone model. SEMAG sets new state-of-the-art Pass@1 accuracy across benchmarks. Using identical backbone models, SEMAG outperforms prior methods by 3.3% on CodeContests. When augmented with self-evolutionary model selection that automatically identifies optimal backbones, SEMAG reaches 52.6%, showcasing both framework effectiveness and adaptability to evolving LLM capabilities.

cs.SE↗

arXiv2025

Quasi-difference-convexity: Modernization of Quasi-differentiable Optimization

Quasi-differentiable functions were introduced by Pshenichnyi in a 1969 monograph written in Russian and translated in an English version in 1971. This class of nonsmooth functions was studied extensively in two decades since but has not received much attention in today's wide optimization literature. This regrettable omission is in spite of the fact that many functions in modern day applications of optimization can be shown to be quasi-differentiable. In essence, a quasi-differentiable function is one whose directional derivative at an arbitrary reference vector, as a function of the direction, is the difference of two positively homogenous, convex functions. Thus, to bring quasi-differentiable functions closer to the class of difference-of-convex functions that has received fast growing attention in recent years in connection with many applied subjects, we propose to rename quasi-differentiable functions as quasi-difference-convex (quasi-dc) functions. Besides modernizing and advancing this class of nonconvex and nondifferentiable functions, our research aims to put together a unified treatment of iterative convex-programming based descent algorithms for solving a broad class of composite quasi-dc programs and to establish their subsequential convergence, sequential convergence, and rates of convergence; the latter two topics are in line with the modern focus of such analysis for convex programs and some extensions and are departures from the sole emphasis of subsequential convergence in the traditional studies of quasi-differentiable optimization. Through this research, we have gained significant new insights and understanding, advanced the fundamentals, and broadened the applications of this neglected yet pervasive class of nonconvex and nondifferentiable functions and their optimization.

math.OC↗

arXiv2024

The monoid representation of upho posets and total positivity

We show that all totally positive formal power series with integer coefficients and constant term $1$ are precisely the rank-generating functions of Schur-positive upho posets, thereby resolving the main conjecture proposed by Gao, Guo, Seetharaman, and Seidel. To achieve this, we construct a bijection between finitary colored upho posets and atomic, left-cancellative, invertible-free monoids, which restricts to a correspondence between $\mathbb{N}$-graded colored upho posets and left-cancellative homogeneous monoids. Furthermore, we introduce semi-upho posets and develop a convolution operation on colored upho posets with colored semi-upho posets within this monoid-theoretic framework.

math.CO↗