arXiv subjects

Nan Jiang

Publications and source records attributed to Nan Jiang.

At least 55 records · Page 3Linked to original sources

arXiv2026

DIVERSED: Relaxed Speculative Decoding via Dynamic Ensemble Verification

Speculative decoding is an effective technique for accelerating large language model inference by drafting multiple tokens in parallel. In practice, its speedup is often bottlenecked by a rigid verification step that strictly enforces the accepted token distribution to exactly match the target model. This constraint leads to the rejection of many plausible tokens, lowering the acceptance rate and limiting overall time speedup. To overcome this limitation, we propose Dynamic Verification Relaxed Speculative Decoding (DIVERSED), a relaxed verification framework that improves time efficiency while preserving generation quality. DIVERSED learns an ensemble-based verifier that blends the draft and target model distributions with a task-dependent and context-dependent weight. We provide theoretical justification for our approach and demonstrate empirically that DIVERSED achieves substantially higher inference efficiency compared to standard speculative decoding methods. Code is available at: https://github.com/comeusr/diversed.

cs.CL↗

arXiv2026

Beyond Semantic Manipulation: Token-Space Attacks on Reward Models

Reward models (RMs) are widely used as optimization targets in reinforcement learning from human feedback (RLHF), yet they remain vulnerable to reward hacking. Existing attacks mainly operate within the semantic space, constructing human-readable adversarial outputs that exploit RM biases. In this work, we introduce a fundamentally different paradigm: Token Mapping Perturbation Attack (TOMPA), a framework that performs adversarial optimization directly in token space. By bypassing the standard decode-re-tokenize interface between the policy and the reward model, TOMPA enables the attack policy to optimize over raw token sequences rather than coherent natural language. Using only black-box scalar feedback, TOMPA automatically discovers non-linguistic token patterns that elicit extremely high rewards across multiple state-of-the-art RMs. Specifically, when targeting Skywork-Reward-V2-Llama-3.1-8B, TOMPA nearly doubles the reward of GPT-5 reference answers and outperforms them on 98.0% of prompts. Despite these high scores, the generated outputs degenerate into nonsensical text, revealing that RMs can be systematically exploited beyond the semantic regime and exposing a critical vulnerability in current RLHF pipelines.

cs.LG↗

arXiv2026

Chvátal-Gomory Rounding of Eigenvector Inequalities for QCQPs

We introduce and analyze a class of valid inequalities for nonconvex quadratically constrained optimization problems (QCQPs) which we call Eigen-CG inequalities. These inequalities are obtained by applying a Chvátal-Gomory (CG) rounding to the well-known eigenvector inequalities for QCQPs, and transferring binary-valid inequalities to the continuous setting via a result of Burer and Letchford (2009). We define three nested subfamilies and prove that they are strictly contained in one another. However, we show that the convex conic closure of two of these subfamilies is equal and, in fact, coincides with the Boros-Hammer inequalities -- a powerful family of inequalities that include, in particular, the triangle and McCormick inequalities. Using this CG perspective, we also prove that dense Eigen-CG inequalities are ineffective when used with the standard SDP+McCormick relaxation. This provides a complementary perspective on what is observed in practice: that sparse inequalities are impactful. Finally, based on these insights, we develop a computational strategy to find sparse Eigen-CG cuts and verify their effectiveness in nonconvex QCQP instances. Our results confirm that density quickly degrades effectiveness, but that including sparse inequalities beyond triangle inequalities can provide significant improvements in dual bounds.

math.OC↗

arXiv2026

Approximating Pareto Frontiers in Stochastic Multi-Objective Optimization via Hashing and Randomization

Stochastic Multi-Objective Optimization (SMOO) is critical for decision-making trading off multiple potentially conflicting objectives in uncertain environments. SMOO aims at identifying the Pareto frontier, which contains all mutually non-dominating decisions. The problem is highly intractable due to the embedded probabilistic inference, such as computing the marginal, posterior probabilities, or expectations. Existing methods, such as scalarization, sample average approximation, and evolutionary algorithms, either offer arbitrarily loose approximations or may incur prohibitive computational costs. We propose XOR-SMOO, a novel algorithm that with probability $1-δ$, obtains $γ$-approximate Pareto frontiers ($γ>1$) for SMOO by querying an SAT oracle poly-log times in $γ$ and $δ$. A $γ$-approximate Pareto frontier is only below the true frontier by a fixed, multiplicative factor $γ$. Thus, XOR-SMOO solves highly intractable SMOO problems (\#P-hard) with only queries to SAT oracles while obtaining tight, constant factor approximation guarantees. Experiments on real-world road network strengthening and supply chain design problems demonstrate that XOR-SMOO outperforms several baselines in identifying Pareto frontiers that have higher objective values, better coverage of the optimal solutions, and the solutions found are more evenly distributed. Overall, XOR-SMOO significantly enhanced the practicality and reliability of SMOO solvers.

cs.LG↗

arXiv2026

Tightening CVaR Approximations via Scenario-Wise Scaling for Chance-Constrained Programming

Chance-constrained programs (CCPs) provide a powerful modeling framework for decision-making under uncertainty, but their nonconvex feasible regions make them computationally challenging. A widely used convex inner approximation replaces chance constraints with Conditional Value-at-Risk (CVaR) constraints; however, the resulting solutions can be overly conservative and suboptimal. We propose a scenario-wise scaling approach that strengthens CVaR approximations for CCPs with finitely supported uncertainty. The method introduces scaling factors that reweight individual scenarios within the CVaR constraint, yielding a family of potentially tighter inner approximations. We establish sufficient conditions under which, for a suitable choice of scaling factors, the scaled CVaR approximation attains the same optimal value as the original CCP and admits a (near-)optimal solution of the CCP. We show that these conditions are tight and further relax them in the convex setting. We also show that optimizing over scenario-wise scaling factors is NP-hard. To address this computational challenge, we develop efficient heuristic and sequential convex approximation algorithms that iteratively update the scaling factors and generate improved feasible solutions. Numerical experiments demonstrate that the proposed methods consistently improve upon standard CVaR and state-of-the-art convex approximations, often reducing conservativeness while maintaining tractability.

math.OC↗

arXiv2026

Solving Chance Constrained Programs via a Penalty based Difference of Convex Approach

We develop two penalty based difference of convex (DC) algorithms for solving chance constrained programs. First, leveraging a rank-based DC decomposition of the chance constraint, we propose a proximal penalty based DC algorithm in the primal space that does not require a feasible initialization. Second, to improve numerical stability in the general nonlinear settings, we derive an equivalent lifted formulation with complementary constraints and show that, after minimizing primal variables, the penalized lifted problem admits a tractable DC structure in the dual space over a simple polyhedron. We then develop a penalty based DC algorithm in the lifted space with a finite termination guarantee. We establish exact penalty and stationarity guarantees under mild constraint qualifications and identify the relationship of the local minimizers between the two formulations. Numerical experiments demonstrate the efficiency and effectiveness of our proposed methods compared with state-of-the-art benchmarks.

math.OC↗

arXiv2026

Ultra-Fast Language Generation via Discrete Diffusion Divergence Instruct

Fast and high-quality language generation is the holy grail that people pursue in the age of AI. In this work, we introduce Discrete Diffusion Divergence Instruct (DiDi-Instruct), a training-based method that initializes from a pre-trained diffusion large language model (dLLM) and distills a few-step student for fast generation. The model distilled with DiDi-Instruct matches or surpasses its dLLM teacher and the GPT-2 baseline while providing up to 64$\times$ acceleration. The theoretical foundation of DiDi-Instruct is a novel framework based on integral KL-divergence minimization, which leads to a practical training algorithm. We further introduce grouped reward normalization, intermediate-state matching, and the reward-guided ancestral sampler to improve training stability, model coverage, and inference quality. On the OpenWebText benchmark, DiDi-Instruct achieves perplexity ranging from 62.2 (8 NFEs) to 18.4 (128 NFEs), outperforming prior accelerated dLLMs and the GPT-2 baseline. These gains incur a negligible entropy loss (around $1$%) and reduce additional training wall-clock time by more than $20\times$ compared to competing dLLM distillation methods. We further validate the robustness and effectiveness of DiDi-Instruct through extensive ablation studies, model scaling, downstream task evaluations, and unconditional protein sequence generation. In conclusion, DiDi-Instruct enables efficient and effective distillation for language generation in the blink of an eye.

cs.CL↗

arXiv2026

Impact of the out-of-plane conductivity on spin transport evaluation in a van der Waals material

Layered materials are promising candidates for spintronic applications due to their unique electronic structures and spin transport properties. However, the strong anisotropic conductivity inherent in these materials complicates the quantitative evaluation of spin Hall conductivity and spin diffusion length. In this work, we present a comprehensive study of spin transport in a transition metal dichalcogenide PtTe$_2$ by combining a three-dimensional finite element model with nonlocal spin valve structures. We developed a theoretical model that treats an anisotropic spin diffusion in the same way as the conventional isotropic model, enabling the extraction of spin diffusion lengths along both the in-plane and out-of-plane directions. Our analysis revealed that the conventional isotropic assumption tends to overestimate some values, particularly for the out-of-plane spin diffusion length and spin Hall conductivity. These findings provide new insight into anisotropic spin diffusion and spin-charge conversions in layered materials and emphasize the importance of accounting for anisotropic conductivity in the design of spintronic devices.

cond-mat.mes-hall↗

arXiv2026

WAFFLE: Finetuning Multi-Modal Models for Automated Front-End Development

Web development involves turning UI designs into functional webpages, which can be difficult for both beginners and experienced developers due to the complexity of HTML's hierarchical structures and styles. While Large Language Models (LLMs) have shown promise in generating source code, two major challenges persist in UI-to-HTML code generation: (1) effectively representing HTML's hierarchical structure for LLMs, and (2) bridging the gap between the visual nature of UI designs and the text-based format of HTML code. To tackle these challenges, we introduce Waffle, a new fine-tuning strategy that uses a structure-aware attention mechanism to improve LLMs' understanding of HTML's structure and a contrastive fine-tuning approach to align LLMs' understanding of UI images and HTML code. Models fine-tuned with Waffle show up to 9.00 pp (percentage point) higher HTML match, 0.0982 higher CW-SSIM, 32.99 higher CLIP, and 27.12 pp higher LLEM on our new benchmark WebSight-Test and an existing benchmark Design2Code, outperforming current fine-tuning methods.

cs.SE↗

arXiv2026

EGG-SR: Embedding Symbolic Equivalence into Symbolic Regression via Equality Graph

Symbolic regression seeks to uncover physical laws from experimental data by searching for closed-form expressions, which is an important task in AI-driven scientific discovery. Yet the exponential growth of the search space of expression renders the task computationally challenging. A promising yet underexplored direction for reducing the search space and accelerating training lies in *symbolic equivalence*: many expressions, although syntactically different, define the same function -- for example, $\log(x_1^2x_2^3)$, $\log(x_1^2)+\log(x_2^3)$, and $2\log(x_1)+3\log(x_2)$. Existing algorithms treat such variants as distinct outputs, leading to redundant exploration and slow learning. We introduce EGG-SR, a unified framework that integrates symbolic equivalence into a class of modern symbolic regression methods, including Monte Carlo Tree Search (MCTS), Deep Reinforcement Learning (DRL), and Large Language Models (LLMs). EGG-SR compactly represents equivalent expressions through the proposed EGG module (via equality graphs), accelerating learning by: (1) pruning redundant subtree exploration in EGG-MCTS, (2) aggregating rewards across equivalent generated sequences in EGG-DRL, and (3) enriching feedback prompts in EGG-LLM. Theoretically, we show the benefit of embedding EGG into learning: it tightens the regret bound of MCTS and reduces the variance of the DRL gradient estimator. Empirically, EGG-SR consistently enhances a class of symbolic regression models across several benchmarks, discovering more accurate expressions within the same time limit. Project page is at: https://nan-jiang-group.github.io/egg-sr.

cs.SC↗

arXiv2026

CUA-Skill: Develop Skills for Computer Using Agent

Computer-Using Agents (CUAs) aim to autonomously operate computer systems to complete real-world tasks. However, existing agentic systems remain difficult to scale and lag behind human performance. A key limitation is the absence of reusable and structured skill abstractions that capture how humans interact with graphical user interfaces and how to leverage these skills. We introduce CUA-Skill, a computer-using agentic skill base that encodes human computer-use knowledge as skills coupled with parameterized execution and composition graphs. CUA-Skill is a large-scale library of carefully engineered skills spanning common Windows applications, serving as a practical infrastructure and tool substrate for scalable, reliable agent development. Built upon this skill base, we construct CUA-Skill Agent, an end-to-end computer-using agent that supports dynamic skill retrieval, argument instantiation, and memory-aware failure recovery. Our results demonstrate that CUA-Skill substantially improves execution success rates and robustness on challenging end-to-end agent benchmarks, establishing a strong foundation for future computer-using agent development. On WindowsAgentArena, CUA-Skill Agent achieves state-of-the-art 57.5% (best of three) successful rate while being significantly more efficient than prior and concurrent approaches. The project page is available at https://microsoft.github.io/cua_skill/.

cs.AI↗

arXiv2026

Towards Sample Efficient Entanglement Classification for 3 and 4 Qubit Systems: A Tailored CNN-BiLSTM Approach

Accurate classification of multipartite entanglement in high-dimensional quantum systems is crucial for advancing quantum communication and information processing. However, conventional methods are resource-intensive, and even many machine-learning-based approaches necessitate large training datasets, creating a significant experimental bottleneck for data acquisition. To address this challenge, we propose a hybrid neural network architecture integrating Convolutional and Bidirectional Long Short-Term Memory networks (CNN-BiLSTM). This design leverages CNNs for local feature extraction and BiLSTMs for sequential dependency modeling, enabling robust feature learning from minimal training data. We investigate two fusion paradigms: Architecture 1 (flattening-based) and Architecture 2 (dimensionality-transforming). When trained on only 100 samples, Architecture 2 maintains classification accuracies exceeding 90% for both 3-qubit and 4-qubit systems, demonstrating rapid loss convergence within tens of epochs. Under full-data conditions (400 000 samples), both architectures achieve accuracies above 99.97%. Comparative benchmarks reveal that our CNN-BiLSTM models, especially Architecture 2, consistently outperform standalone CNNs, BiLSTMs, and MLPs in low-data regimes, albeit with increased training time. These results demonstrates that the tailored CNN-BiLSTM fusion significantly alleviates experimental data acquisition burden, offering a practical pathway toward scalable entanglement verification in complex quantum systems.

quant-ph↗

arXiv2026

A Unifying View of Coverage in Linear Off-Policy Evaluation

Off-policy evaluation (OPE) is a fundamental task in reinforcement learning (RL). In the classic setting of linear OPE, finite-sample guarantees often take the form $$ \textrm{Evaluation error} \le \textrm{poly}(C^π, d, 1/n,\log(1/δ)), $$ where $d$ is the dimension of the features and $C^π$ is a coverage parameter that characterizes the degree to which the visited features lie in the span of the data distribution. While such guarantees are well-understood for several popular algorithms under stronger assumptions (e.g. Bellman completeness), the understanding is lacking and fragmented in the minimal setting where only the target value function is linearly realizable in the features. Despite recent interest in tight characterizations of the statistical rate in this setting, the right notion of coverage remains unclear, and candidate definitions from prior analyses have undesirable properties and are starkly disconnected from more standard definitions in the literature. We provide a novel finite-sample analysis of a canonical algorithm for this setting, LSTDQ. Inspired by an instrumental-variable view, we develop error bounds that depend on a novel coverage parameter, the feature-dynamics coverage, which can be interpreted as linear coverage in an induced dynamical system for feature evolution. With further assumptions -- such as Bellman-completeness -- our definition successfully recovers the coverage parameters specialized to those settings, finally yielding a unified understanding for coverage in linear OPE.

cs.LG↗

arXiv2026

Streamlining evidence based clinical recommendations with large language models

Clinical evidence underpins informed healthcare decisions, yet integrating it into real-time practice remains challenging due to intensive workloads, complex procedures, and time constraints. This study presents Quicker, an LLM-powered system that automates evidence synthesis and generates clinical recommendations following standard guideline development workflows. Quicker delivers an end-to-end pipeline from clinical questions to recommendations and supports customized decision-making through integrated tools and interactive interfaces. To evaluate how closely Quicker can reproduce guideline development processes, we constructed Q2CRBench-3, a benchmark derived from guideline development records for three diseases. Experiments show that Quicker produces precise question decomposition, expert-aligned retrieval, and near-comprehensive screening. Quicker assistance improved the accuracy of extracted study data, and its recommendations were more comprehensive and coherent than clinician-written ones. In system-level testing, Quicker working with one participant reduced recommendation development to 20-40 min. Overall, the findings demonstrate Quicker's potential to enhance the speed and reliability of evidence-based clinical decision-making.

cs.CL↗

arXiv2026

Spin Hall effect in van der Waals ferromagnet Fe$_{5}$GeTe$_{2}$

We investigate the spin Hall effect (SHE) in a van der Waals (vdW) ferromagnet Fe$_{5}$GeTe$_{2}$ (FGT) with a Curie temperature $T_{\rm C}$ of 310 K utilizing the spin-torque ferromagnetic resonance method. In synchronization with the emergence of the ferromagnetic phase resulting in the anomalous Hall effect (AHE), a noticeable enhancement in the SHE was observed below $T_{\rm C}$. On the other hand, the SHE shows a different temperature dependence from the AHE: the effective spin Hall conductivity is clearly enhanced with decreasing temperature unlike the anomalous Hall conductivity, reflecting the variation of band-structure accompanied by the complicated magnetic ordering of the FGT. The results provide a deep understanding of the SHE in magnetic materials to open a new route for novel functionalities in vdW materials-based spintronic devices.

cond-mat.mes-hall↗

arXiv2025

UniAct: Unified Motion Generation and Action Streaming for Humanoid Robots

A long-standing objective in humanoid robotics is the realization of versatile agents capable of following diverse multimodal instructions with human-level flexibility. Despite advances in humanoid control, bridging high-level multimodal perception with whole-body execution remains a significant bottleneck. Existing methods often struggle to translate heterogeneous instructions -- such as language, music, and trajectories -- into stable, real-time actions. Here we show that UniAct, a two-stage framework integrating a fine-tuned MLLM with a causal streaming pipeline, enables humanoid robots to execute multimodal instructions with sub-500 ms latency. By unifying inputs through a shared discrete codebook via FSQ, UniAct ensures cross-modal alignment while constraining motions to a physically grounded manifold. This approach yields a 19% improvement in the success rate of zero-shot tracking of imperfect reference motions. We validate UniAct on UniMoCap, our 20-hour humanoid motion benchmark, demonstrating robust generalization across diverse real-world scenarios. Our results mark a critical step toward responsive, general-purpose humanoid assistants capable of seamless interaction through unified perception and control.

cs.CV↗

arXiv2025

Unified Software Engineering Agent as AI Software Engineer

The growth of Large Language Model (LLM) technology has raised expectations for automated coding. However, software engineering is more than coding and is concerned with activities including maintenance and evolution of a project. In this context, the concept of LLM agents has gained traction, which utilize LLMs as reasoning engines to invoke external tools autonomously. But is an LLM agent the same as an AI software engineer? In this paper, we seek to understand this question by developing a Unified Software Engineering agent or USEagent. Unlike existing work which builds specialized agents for specific software tasks such as testing, debugging, and repair, our goal is to build a unified agent which can orchestrate and handle multiple capabilities. This gives the agent the promise of handling complex scenarios in software development such as fixing an incomplete patch, adding new features, or taking over code written by others. We envision USEagent as the first draft of a future AI Software Engineer which can be a team member in future software development teams involving both AI and humans. To evaluate the efficacy of USEagent, we build a Unified Software Engineering bench (USEbench) comprising of myriad tasks such as coding, testing, and patching. USEbench is a judicious mixture of tasks from existing benchmarks such as SWE-bench, SWT-bench, and REPOCOD. In an evaluation on USEbench consisting of 1,271 repository-level software engineering tasks, USEagent shows improved efficacy compared to existing general agents such as OpenHands CodeActAgent. There exist gaps in the capabilities of USEagent for certain coding tasks, which provides hints on further developing the AI Software Engineer of the future.

cs.SE↗

arXiv2025

Reinforce-Ada: An Adaptive Sampling Framework under Non-linear RL Objectives

Reinforcement learning (RL) for large language model reasoning is frequently hindered by signal loss, a phenomenon where standard uniform sampling with small group sizes fails to uncover informative learning signals for difficult prompts. We demonstrate that this collapse is a statistical artifact of undersampling rather than an inherent model limitation. To address this systematically, we introduce a theoretical framework based on optimizing a non-linear RL objective (e.g., log-likelihood). We show that this objective naturally induces a weighted gradient estimator that prioritizes difficult prompts, which can be robustly realized through adaptive sampling. Guided by this framework, we propose Reinforce-Ada, a family of algorithms that dynamically allocates inference budgets based on prompt difficulty, effectively scaling up RL compute to where it is needed most. Unlike passive filtering methods that discard low-signal prompts, Reinforce-Ada actively invests compute to recover them. We introduce two efficient realizations: an estimation-based approach and a model-free sequential sampling approach. Extensive experiments across multiple benchmarks show that Reinforce-Ada significantly outperforms uniform baselines like GRPO, recovering lost signals and accelerating convergence by up to $2\times$ while maintaining the same total inference budget. Code is available at https://github.com/RLHFlow/Reinforce-Ada.

cs.LG↗