Selected work

Representative Papers

Monte Carlo Graph Coloring

Apr 04, 2025Communications in Computer and Information Science

This work addresses the NP-hard large-scale graph coloring problem by systematically adapting Monte Carlo search (MCS) paradigms—specifically Nested Monte Carlo Search (NMCS) and Nested Rollout Policy Adaptation (NRPA)—to this classical combinatorial optimization task for the first time. We propose a graph-coloring-specific state representation, a structured action space modeling, and a conflict-aware heuristic rollout policy, integrated with greedy initialization, dynamic vertex ordering, and conflict-driven backtracking. Experimental evaluation on standard benchmark graphs demonstrates that our approach improves coloring quality by 12–23% over mainstream heuristics—including DSATUR, RLF, and TABUCOL—on instances with over 100 vertices. Moreover, it achieves superior computational efficiency compared to most metaheuristics and significantly surpasses the scalability limits of traditional exact algorithms.

18 citationsRead paper

Minimum Stable Cut and Treewidth

Apr 27, 2021International Colloquium on Automata, Languages and Programming

This paper studies the Minimum Stable Cut problem on graphs—i.e., finding a locally optimal cut of minimum weight, where no single-vertex reassignment improves the cut weight. We investigate its computational complexity and algorithm design under the dual parameters treewidth (tw) and maximum degree (Δ). We present the first FPT exact algorithm parameterized by tw and Δ, running in $2^{O(Delta cdot tw)}(n + log W)^{O(1)}$ time. We establish ETH-tight lower bounds: unless ETH fails, no $(nW)^{o(pw)}$ or $2^{o(Delta cdot pw)}$ algorithm exists. For unweighted graphs, we give an optimal $Delta^{O(tw)} n^{O(1)}$ algorithm. Moreover, we devise the first FPT approximation scheme, circumventing NP-hardness. Our techniques integrate dynamic programming over tree decompositions, pathwidth-based analysis, pseudo-polynomial encoding, and fine-grained ETH reductions.

7 citationsRead paper

Parameterized Spanning Tree Congestion

Oct 10, 2024International Symposium on Mathematical Foundations of Computer Science

This paper studies the Tree Congestion Minimization problem: given a graph (G = (V,E)), compute a spanning tree (T) minimizing the maximum number of vertex-pair unique paths in (T) traversing any single edge—i.e., the edge congestion. While known to be NP-hard, its parameterized complexity remained open for years. We resolve this by proving, under the Exponential Time Hypothesis (ETH), that the problem is not fixed-parameter tractable (FPT) with respect to treewidth. Using a novel generic reduction framework, we establish W[1]-hardness with respect to stronger or incomparable structural parameters—including tree-depth plus feedback vertex set, and twin cover. Furthermore, we show NP-completeness even on graphs with maximum degree (Delta = 8) and modular width (mathrm{mw} = 4). These results comprehensively settle multiple long-standing open questions and significantly advance the theoretical boundaries of structural parameterized algorithms.

2 citationsRead paper

First Order Logic on Pathwidth Revisited Again

Oct 18, 2022International Colloquium on Automata, Languages and Programming

Courcelle’s Theorem enables MSO/FO model checking on graphs of bounded treewidth, but its runtime involves a tower of exponentials whose height depends on the number of quantifier alternations in the formula—a lower bound that is tight for treewidth. This work focuses on the more restricted class of graphs with bounded pathwidth. We establish the first result showing that FO model checking on such graphs can avoid tower-type dependence on the formula: leveraging path decompositions and a novel logical metatheorem, we extend the Gajarský–Hliněný framework to design a linear-time algorithm whose dependence on the formula is elementary (i.e., bounded by a fixed iterate of exponentiation), not non-elementary. This breakthrough reveals a fundamental complexity-theoretic distinction between pathwidth and treewidth in the context of first-order logic, and constitutes the first elementary-dependence FO model-checking algorithm for a natural graph class beyond those of bounded tree depth.

2 citationsRead paper

Limit Points of Reflow with Minibatch Optimal Transport

Aug 07, 2026

This work investigates the asymptotic behavior of alternating Reflow iterations with mini-batch optimal transport (OT) under a fixed batch size. By introducing a novel notion of weakly corrected couplings, the authors prove that the limiting coupling satisfies N-cyclical monotonicity and, under suitable support conditions, coincides with the optimal transport map between the endpoint distributions. The analysis integrates the Reflow framework, mini-batch OT, N-cyclical monotonicity, and gradient-field constraints to demonstrate that the limit exhibits favorable structural properties—such as correctability and linearity. This study establishes, for the first time, a rigorous theoretical connection between the Reflow limit and classical optimal transport theory, thereby providing a formal convergence guarantee for the Reflow process.

0 citationsRead paper

Recent publications

Latest Papers

Limit Points of Reflow with Minibatch Optimal Transport

Aug 07, 2026

This work investigates the asymptotic behavior of alternating Reflow iterations with mini-batch optimal transport (OT) under a fixed batch size. By introducing a novel notion of weakly corrected couplings, the authors prove that the limiting coupling satisfies N-cyclical monotonicity and, under suitable support conditions, coincides with the optimal transport map between the endpoint distributions. The analysis integrates the Reflow framework, mini-batch OT, N-cyclical monotonicity, and gradient-field constraints to demonstrate that the limit exhibits favorable structural properties—such as correctability and linearity. This study establishes, for the first time, a rigorous theoretical connection between the Reflow limit and classical optimal transport theory, thereby providing a formal convergence guarantee for the Reflow process.

0 citationsRead paper

Forcing and duality-corrected contracts for volatility control

Jul 29, 2026

This study addresses the challenge in continuous-time principal–agent problems where existing methods fail to construct optimal contracts under joint control of drift and volatility, due to the breakdown of key structural assumptions. To overcome this limitation, the paper introduces a general incentive contract framework parameterized by a function ψ, which simultaneously satisfies revelation and principal-losslessness properties. Within this framework, two classes of contracts are constructed: the first employs backward stochastic differential equations (BSDEs) to implement an enforcement mechanism that ensures controllability of the agent’s actions; the second leverages second-order BSDEs (2BSDEs) to correct the duality gap and recover optimality without relying on the original restrictive assumptions. This work thus transcends the dependence of prior theory on specific structural conditions and establishes a novel pathway for optimal contract design in general settings.

0 citationsRead paper

Kemeny Rank Aggregation is NP-Hard for Three Voters

Jul 28, 2026

This study resolves an open problem posed by Dwork et al. in 2001 concerning the computational complexity of Kemeny rank aggregation with only three voters. By constructing a polynomial-time reduction from the MAX CUT problem, the authors provide the first rigorous proof that Kemeny rank aggregation remains NP-complete even when the number of input rankings is exactly three. This result sharpens the known complexity boundary for the problem and establishes its intrinsic intractability at the smallest nontrivial number of voters, thereby completing the characterization of its computational hardness across all relevant input sizes.

0 citationsRead paper

Reinforcement Learning for Code Optimization

Jul 28, 2026

Traditional reinforcement learning struggles in code optimization due to execution-time noise, sparse rewards, and policy instability, often failing to balance correctness and efficiency. This work proposes a three-stage learnable execution-time framework: first, constructing a high-quality test suite and a calibrated sandbox to refine the evaluation environment; second, designing a composite reward function that jointly optimizes correctness and speedup ratio; and third, enhancing the GRPO algorithm to handle sparse, noisy temporal signals, augmented with an offline performance simulator to predict efficient configurations. Evaluated on Qwen 2.5 7B and CWM 32B, the approach achieves top-50% pass@1 rates of 31.3% and 50.4%, respectively. On CWM 32B, it yields a 125% relative improvement under the top-30% metric, attains an 83% speed win rate, and reaches half the complexity-optimization capability of human experts.

0 citationsRead paper

Separating Representation from Reconstruction Enables Scalable Text Encoders

Jul 04, 2026

This work addresses the limitation of conventional BERT, where representation learning is tightly coupled with token reconstruction, thereby constraining representation usability under frozen probing and hindering model scalability. To overcome this, the authors propose CrossBERT, the first approach to decouple these two objectives through a dual-branch architecture and complementary masking strategies. This design enables high masking ratios (≥50%) while allowing gradient computation over all tokens, substantially improving training efficiency and sample utilization. Evaluated on the MTEB (English, v2) and frozen GLUE benchmarks, CrossBERT demonstrates superior performance, achieving 1.5–2× higher throughput, 2× better sample efficiency, and monotonic scalability with respect to model size.

0 citationsRead paper