Searcharxiv⌕ Search

arXiv subjects

Xinpeng Li

Publications and source records attributed to Xinpeng Li.

At least 19 recordsLinked to original sources

Are Dense Rewards Worth Their Tokens in GRPO?

RLVR trains language models to reason by rewarding correct final answers. A widely used algorithm, group-relative policy optimization (GRPO), gives every token of a solution the same credit, so dense process rewards have been proposed to credit intermediate steps, for example with a Monte-Carlo reward that scores each step by sampling continuations from it. Comparing such a reward with GRPO per training update, or vetting it with a gradient test on a single batch, leaves out what it costs: in our setting, the continuations alone take 91% of all generated tokens. We introduce a matched-compute protocol that gives every training intervention the same budget, measured in generated tokens and training GPU-hours, and tests it in preregistered studies with controls for side effects. Across 1,668 runs on six models, matching compute reverses the ranking. On Qwen2.5-1.5B, the Monte-Carlo reward leads standard GRPO per update, yet ends 14.3 percentage points lower at equal tokens and 7.8 points lower at equal training time, and none of four dense-reward variants outperforms GRPO even with as much training time. A per-token gradient test orders these rewards as training does but overestimates each by one to two points, as much as the gains it is meant to detect. In contrast, training on the prompts whose predicted success is nearest 1/2 improves accuracy at matched compute, by 1.7 points on Qwen3-0.6B and 0.8 on Qwen2.5-1.5B. These gains replicate on fresh seeds and unseen problems, and a simple rule, that the candidate prompts are predicted to fail more often than not, predicted where they appear in five of six preregistered tests. Both gains come early in training, and standard GRPO with twice the initial learning rate comes within half a point of them. We argue that training interventions should be validated at matched compute, which overturns some apparent gains and confirms others.

cs.CL↗

Towards Trustworthy Physical Intelligence: From Theory to Practice Across Life Cycle

Physical intelligence refers to intelligence systems that understand, reason about and act in accordance with the physical world and its underlying laws, dynamics, and constraints. Unlike conventional AI systems, physical intelligence interacts continuously with uncertain physical environments, and its actions produce consequences that are physically irreversible. As existing trustworthy AI frameworks have been developed primarily for digital AI systems, they do not fully capture the distinctive challenges of Physical Intelligence, such as physical safety, cyber-physical security, and physical manufacturing process. To address this gap, our paper develop a foundation for understanding, operationalizing, and governing trustworthy physical intelligence. First, we characterize the fundamental capabilities of physical intelligence. Second, we investigate the role of physical laws and principles in AI and analyze the associated challenges of physical intelligence. Third, we provide a unified structure of the end-to-end physical intelligence life cycle across five core stages and introduce the Trustworthy Physical Intelligence Operationalization (T-PAIO) framework. Fourth, we formulate the Trustworthy Physical Intelligence (T-PAI) Principles as a theoretical framework that synthesizes key trustworthiness dimensions and establishes a conceptual foundation for the governance of physical intelligence systems.

cs.AI↗

LEANGUARD: Lean-Verified Guardrails for Tool-Using Agents

Tool-using language-model agents often execute actions that their domain policies forbid. We present LeanGuard, a runtime that checks every proposed tool call against a policy written in Lean4 before the call reaches the environment. Policies combine state predicates with scoped past-time temporal operators over the execution history, evaluation is fail-closed, and the decision procedure is proved equivalent to a LeanLTL semantics. A denied call can be corrected in two ways. With feedback, the agent receives the violated rule and a hint; with proof-guided repair, Lean metaprogramming searches for a nearby call that preserves the user's intent and proves that the policy admits it, without the model. On Tau-Bench with five Qwen models (0.8B-27B), unguarded agents execute at least 33 calls that the Lean policy forbids, while LeanGuard executes none; with feedback, task success does not change significantly. On DTap, LeanGuard reduces direct prompt-injection success from 57.9% to 5.9%, and in telecom, where benign tasks need no writes, benign success is unchanged. Feedback reduces repeated blocked calls and raises task success for four of five models over a bare denial. On replayed denials, repair proves an admissible edit for 88 of 252 denials caused by argument values, in a median of 2,ms.

cs.AI↗

Generalized Wishart processes under nonlinear expectation

We develop a theory of generalized Wishart processes within the framework of $G$-expectation, with $G$-Dyson Brownian motion and the $G$-Wishart process as two typical examples. In contrast to the classical setting, we show that $G$-Dyson Brownian motion may exhibit collisions and provide a sufficient condition ensuring the non-collision of the eigenvalues of generalized Wishart processes. We then derive the equations governing the high-dimensional limits of the empirical spectral measures under $G$-expectation, leading to the corresponding semicircle law for $G$-Dyson Brownian motion and Marchenko--Pastur law for the $G$-Wishart process.

math.PR↗

What Can't Jev Decide? Probing the Limits of One-Pass Decision Models

Jev and its open replicas choose among the options a caller supplies with a single forward pass, fast enough to decide inside an agent loop. We ask what such a model cannot decide, and find a simple limit: Jev cannot make a decision that an LLM makes only by writing out intermediate steps. On a benchmark that varies one factor at a time, Jev finds and selects records without error up to 20k tokens and 64 options, but once the answer depends on two or more updates of one user, accuracy falls, to 53-80% at sixteen updates, because similar records interfere. Given a stale summary of the state, Jev follows it. On real decision items graded by how many thinking tokens a reference LLM needs, Jev answers 97.5% of the items that one forward pass of that LLM solves, but only 74% of the others. Jev also does not know its limit: its confidence flags the errors it makes itself but not those it inherits from a stale input, and neither an explicit nor a trained hand-off option catches the latter. The open models we evaluate share these limits. Checking the state such models rely on requires a model that reasons.

cs.CL↗

Correcting WHERE, Preserving HOW: Compositional Generalization for Vision-Language-Action Models via Referential Guidance

While Vision-Language-Action (VLA) models enable flexible action generation, their generalization across diverse environmental elements, including manipulated objects, destinations, and backgrounds, is limited by the lack of diversity in robotic training data. Trained end-to-end on such data, VLAs tend to exploit visual shortcuts, associating actions with task-irrelevant visual features rather than the intended task semantics. These shortcuts block recomposition of elements already seen by the policy, that is, compositional generalization. Existing approaches mitigate such entanglement through task-relevant perception or targeted data diversification, but offer no explicit mechanism for unseen recomposition and require backbone-specific modifications with retraining. We observe that under such recomposition, VLAs often fail at global grounding while retaining local manipulation skills that recover near the correct target in familiar configurations. Therefore, we propose Referential Guidance (ReGuide), a training-free wrapper that, given object poses from a grounding module, combines semantic and geometric rebinding to guide the end-effector into demonstration-supported configurations of the instructed referent, where the frozen policy can resume execution. Experiments in simulation across multiple VLA backbones as well as on a real robot show that ReGuide improves success rates under compositional shifts by up to 56.8 and 75.0 percentage points, respectively, while preserving standard-task performance.

cs.RO↗

Small-supercell and Small-dataset Training Strategy of Machine Learning Interatomic Potentials for Point Defects

Machine learning interatomic potentials (MLIPs) can treat large-scale material systems with near first-principles accuracy and have been widely used to accelerate point-defect simulations. However, the training of MLIPs usually relies on large amounts of DFT data. This issue is particularly pronounced for charged defects, for which DFT calculations of large supercells are required to avoid long-range Coulomb interactions and finite-size effects, making the construction of datasets computationally expensive. In this work, we propose an efficient MLIP training scheme for neutral and lowly charged point defects based on small supercells (less than 100 atoms) and limited number of DFT calculations. The scheme requires only four DFT structural relaxations to construct the training dataset and the trained MLIPs dedicated for the defect can predict the defect formation energies in larger supercells (over 200 atoms) with small errors (mostly smaller than 0.3 eV). Using defects in GaN, SiO$_2$, and $\mathrm{Cu}_2\mathrm{ZnSnS}_4$ as representative examples, we evaluate the extrapolation capability of this scheme for predicting defect total energies and structural relaxations across different supercell sizes. The results show that an MLIP trained only on single small defect supercell produces severe errors when treating large supercells. Incorporating defect data from multiple supercell sizes improves the predictive performance of the MLIP, while adding pristine defect-free bulk supercells further enhances the accuracy. These results provide practical guidance for training MLIP models of defect systems with low computational costs.

cond-mat.mtrl-sci↗

A Decomposed Bilevel Search for Variable-Metric Proximal Gradient Methods

Variable-metric proximal methods accelerate composite convex optimization, but the scaled proximal map induced by a quasi-Newton metric rarely has a closed form. We develop \emph{Decomposed Bilevel Search} (DBS), based on a diagonal-plus-rank-one factor \(X=D+uv^\top\) whose diagonal scaling satisfies a weak secant equation. The induced inverse metric \(B^{-1}=XX^\top\) recovers zero-memory DFP/BFGS-type Broyden members, while the factor form reduces each scaled proximal step to a two-dimensional monotone residual system. Each residual evaluation requires one diagonal-metric proximal map, and a certified bilevel solve reaches target accuracy \(ε\) in \(\mathcal O((d+T_p)\log^2(1/ε))\) work, where \(T_p\) is the cost of that proximal map. Under strong convexity, the outer method converges linearly with exact and inexact inner solves. In the scalar specialization \(D=αI\), the oracle uses only ordinary proximal evaluations of the regularizer and no generalized Jacobian or active-set information. Experiments on ordered-weighted \(\ell_1\) (SLOPE/OWL) and group-lasso logistic regression evaluate both the inner oracle and the full outer method. The oracle solves high-dimensional SLOPE scaled proximal subproblems of condition number up to \(4\times10^6\) using a few hundred ordinary proximal evaluations. In SLOPE outer benchmarks, the warm-started oracle keeps the scaled-proximal overhead controlled and DBS reaches stringent targets with far fewer gradient evaluations than Lipschitz-normalized FISTA; on \texttt{real-sim} this becomes a clear target-time advantage. On group-lasso logistic regression, DBS is reliable on synthetic correlated instances and fastest on a real-data grouped-copy instance.

math.OC↗

Imagining Recovery: Inference-Time Counterfactual Realignment for Vision-Language-Action Models

Vision-language-action (VLA) models have improved the flexibility and generality of robotic manipulation, yet they remain fragile to online disruptions, such as changes in task goal, scene configuration, or robot state. Existing recovery methods often require failure data, policy retraining, or external corrective agents, introducing additional data requirements and execution risks. We propose Counterfactual Realignment (CoRe), a training-free framework that recovers a frozen VLA at inference time without failure data. Upon detecting a deviation, CoRe imagines how the policy would continue toward the current goal from a recent viable state, using synthesized observations in place of physical execution, and then minimally realigns the robot and scene to rejoin this imagined continuation before returning control to the policy. Recovery is therefore planned without physical trial-and-error, preserves completed task progress, and handles both mid-episode instruction changes and physical perturbations in a unified manner. Extensive experiments across multiple simulators, VLA backbones, and real-world settings show that CoRe improves success rates by up to 85.0 percentage points to near-nominal levels while reducing physical restorations by 42.2%, without policy fine-tuning or failure-specific recovery training.

cs.RO↗

Path-Lock Expert: Separating Reasoning Mode in Hybrid Thinking via Architecture-Level Separation

Hybrid-thinking language models expose explicit /think and /no_think modes, but current designs do not separate them cleanly. Even in /no_think mode, models often emit long and self-reflective responses, causing reasoning leakage. Existing work reduces this issue through better data curation and multi-stage training, yet leakage remains because both modes are still encoded in the same feed-forward parameters. We propose Path-Lock Expert (PLE), an architecture-level solution that replaces the single MLP in each decoder layer with two semantically locked experts, one for /think and one for /no_think, while keeping attention, embeddings, normalization, and the language-model head shared. A deterministic control-token router selects exactly one expert path for the entire sequence, so inference preserves the dense model's per-token computation pattern and each expert receives mode-pure updates during supervised fine-tuning. Across math and science reasoning benchmarks, PLE maintains strong /think performance while producing a substantially stronger mode separation, a /no_think mode with higher accuracy and far less reasoning leakage. On Qwen3-4B, for example, compared to the SFT-only baseline on AIME24, PLE generates 17x fewer reflective tokens (6.01 vs. 0.35 per answer) and 2x shorter outputs (8665 vs. 4101 tokens), and improves /no_think accuracy from 35.33% to 44.67%, while maintaining /think-mode performance (61.33% vs. 60.00%). These results suggest that controllable hybrid thinking is fundamentally an architectural problem, and separating mode-specific feed-forward pathways is a simple and effective solution.

cs.CL↗

STRIDE: Strategic Trajectory Reasoning via Discriminative Estimation for Verifiable Reinforcement Learning

Reinforcement Learning with Verifiable Rewards (RLVR) has become an effective post-training paradigm for improving the reasoning abilities of large language models. However, existing RLVR methods typically rely on final-answer correctness to assign trajectory-level rewards, providing sparse supervision and treating all tokens uniformly regardless of their actual contribution to reasoning. Although recent studies introduce intermediate signals such as process rewards, high-entropy tokens, and semantic uncertainty, these signals are often not inherently verifiable and may fail to distinguish beneficial strategic patterns from harmful ones. To address this limitation, we propose STRIDE (Strategic Trajectory Reasoning with Discriminative Estimation), a fine-grained RLVR framework that derives strategic reasoning supervision from verifiable outcomes. STRIDE contrasts successful and failed trajectories within each response group to estimate the outcome-discriminative preference of each $n$-gram strategic pattern, and further combines this signal with reasoning saliency entropy to identify decision-relevant strategic patterns. These patterns are assigned differentiated advantage values during RL optimization, enabling more precise credit assignment while preserving the verifiability of RLVR. Extensive experiments demonstrate that STRIDE consistently improves reasoning performance across diverse models, tasks, and extended settings, including VLMs and agent-based systems.

cs.AI↗

Mid-Think: Training-Free Intermediate-Budget Reasoning via Token-Level Triggers

Hybrid reasoning language models are commonly controlled through high-level Think/No-think instructions to regulate reasoning behavior, yet we found that such mode switching is largely driven by a small set of trigger tokens rather than the instructions themselves. Through attention analysis and controlled prompting experiments, we show that a leading ``Okay'' token induces reasoning behavior, while the newline pattern following `` '' suppresses it. Based on this observation, we propose Mid-Think, a simple training-free prompting format that combines these triggers to achieve intermediate-budget reasoning, consistently outperforming fixed-token and prompt-based baselines in terms of the accuracy-length trade-off. Furthermore, applying Mid-Think to RL training after SFT reduces training time by approximately 15% while improving final performance of Qwen3-8B on AIME from 69.8% to 72.4% and on GPQA from 58.5% to 61.1%, demonstrating its effectiveness for both inference-time control and RL-based reasoning training.

cs.CL↗

InterQ: Communication-Aware Scheduling Across Modular QPUs with Classical and Quantum Links

As quantum computing scales toward practical workloads, future systems are expected to move beyond single monolithic processors toward modular architectures that connect multiple QPUs. Different platforms realize this modularity through different communication models: superconducting systems rely on real-time classical links and dynamic-circuit coordination, trapped-ion systems use photonic interconnects for remote entanglement, and neutral-atom systems provide strong intra-core connectivity with proposed optical links for inter-core communication. This heterogeneity makes communication-aware scheduling essential for shared modular quantum cloud environments. We present InterQ, a communication-aware scheduler for modular QPU architectures with heterogeneous communication models. InterQ jointly considers qubit capacity, placement, parallel execution, and communication-driven dependencies across distributed subcircuits, while enabling adaptive circuit cutting to reduce makespan while balancing fidelity and communication overhead. The framework distinguishes classical-link execution, where measurement and feedforward impose synchronization constraints, from quantum-link execution, where entanglement distribution and state transfer determine coordination cost. Using a unified simulation framework to compare superconducting, trapped-ion, and neutral-atom modular systems, InterQ shows how communication models and scheduler-driven cutting decisions affect throughput, latency, and fidelity. Across evaluated workloads, InterQ exposes an architecture-dependent tradeoff: neutral-atom modular QPUs achieve the highest fidelity, superconducting systems minimize runtime, and trapped-ion systems provide a balanced intermediate profile across fidelity and makespan.

quant-ph↗

Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models

Vision-Language-Action (VLA) models achieve remarkable flexibility and generalization beyond classical control paradigms. However, most prevailing VLAs are trained under a single-frame observation paradigm, which leaves them structurally blind to temporal dynamics. Consequently, these models degrade severely in non-stationary scenarios, even when trained or finetuned on dynamic datasets. Existing approaches either require expensive retraining or suffer from latency bottlenecks and poor temporal consistency across action chunks. We propose Pace-and-Path Correction, a training-free, closed-form inference-time operator that wraps any chunked-action VLA. From a single quadratic cost, joint minimization yields a unified solution that decomposes orthogonally into two distinct channels. The pace channel compresses execution along the planned direction, while the path channel applies an orthogonal spatial offset, jointly absorbing the perceived dynamics within the chunk window. We evaluate our approach on a comprehensive diagnostic benchmark MoveBench designed to isolate motion as the sole controlled variable. Empirical results demonstrate that our framework consistently outperforms state-of-the-art training-free wrappers and dynamic-adaptive methods and improves success rates by up to 28.8% and 25.9% in absolute terms over foundational VLA models in dynamic-only and static-dynamic mixed environments, respectively.

cs.RO↗

Benchmarking and Enabling Efficient Chinese Medical Retrieval via Asymmetric Encoders

Effective medical text retrieval requires both high accuracy and low latency. While LLM-based embedding models possess powerful retrieval capabilities, their prohibitive latency and high computational cost limit their application in real-time scenarios. Furthermore, the lack of comprehensive and high-fidelity benchmarks hinders progress in Chinese medical text retrieval. In this work, we introduce the Chinese Medical Text Embedding Benchmark (CMedTEB), a benchmark spanning three kinds of practical embedding tasks: retrieval, reranking, and semantic textual similarity (STS). Distinct from purely automated datasets, CMedTEB is curated via a rigorous multi-LLM voting pipeline validated by clinical experts, ensuring gold-standard label quality while effectively mitigating annotation noise. On this foundation, we propose the Chinese Medical Asymmetric REtriever (CARE), an asymmetric architecture that pairs a lightweight BERT-style encoder for online query encoding with a powerful LLM-based encoder for offline document encoding. However, optimizing such an asymmetric retriever with two structurally different encoders presents distinctive challenges. To address this, we introduce a novel two-stage training strategy that progressively bridges the query and document representations. Extensive experiments demonstrate that CARE surpasses state-of-the-art symmetric models on CMedTEB, achieving superior retrieval performance without increasing inference latency.

cs.IR↗

QuMod: Parallel Quantum Job Scheduling on Modular QPUs using Circuit Cutting

The quantum computing community is increasingly positioning quantum processors as accelerators within classical HPC workflows, analogous to GPUs and TPUs. However, many real-world applications require scaling to hundreds or thousands of physical qubits to realize logical qubits via error correction. To reach these scales, hardware vendors employing diverse technologies -- such as trapped ions, photonics, neutral atoms, and superconducting circuits -- are moving beyond single, monolithic QPUs toward modular architectures connected via interconnects. For example, IonQ has proposed photonic links for scaling, while IBM has demonstrated a modular QPU architecture by classically linking two 127-qubit devices. Using dynamic circuits, Bell-pair-based teleportation, and circuit cutting, they have shown how to execute a large quantum circuit that cannot fit on a single QPU. As interest in quantum computing grows, cloud providers must ensure fair and efficient resource allocation for multiple users sharing such modular systems. Classical interconnection of QPUs introduces new scheduling challenges, particularly when multiple jobs execute in parallel. In this work, we develop a multi-programmable scheduler for modular quantum systems that jointly considers qubit mapping, parallel circuit execution, measurement synchronization across subcircuits, and teleportation operations between QPUs using dynamic circuits.

quant-ph↗

AgentCE-Bench: Agent Configurable Evaluation with Scalable Horizons and Controllable Difficulty under Lightweight Environments

Existing Agent benchmarks suffer from two critical limitations: high environment interaction overhead (up to 41\% of total evaluation time) and imbalanced task horizon and difficulty distributions that make aggregate scores unreliable. To address these issues, we propose AgentCE-Bench built around a unified grid-based planning task, where agents must fill hidden slots in a partially completed schedule subject to both local slot constraints and global constraints. Our benchmark offers fine-grained control through two orthogonal axes: \textbf{Scalable Horizons}, controlled by the number of hidden slots $H$, and \textbf{Controllable Difficulty}, governed by a decoy budget $B$ that determines the number of globally misleading decoy candidates. Crucially, all tool calls are resolved via static JSON files under a \textbf{Lightweight Environment} design, eliminating setup overhead and enabling fast, reproducible evaluation suitable for training-time validation. We first validate that $H$ and $B$ provide reliable control over task horizon and difficulty, and that AgentCE-Bench exhibits strong domain consistency and model discriminability. We then conduct comprehensive experiments across 13 models of diverse sizes and families over 6 domains, revealing significant cross-model performance variation and confirming that AgentCE-Bench provides interpretable and controllable evaluation of agent reasoning.

cs.AI↗

Towards Online Multi-Modal Social Interaction Understanding

In this paper, we introduce a new problem, Online-MMSI, where the model must perform multimodal social interaction understanding (MMSI) using only historical information. Given a recorded video and a multi-party dialogue, the AI assistant is required to immediately identify the speaker's referent, which is critical for real-world human-AI interaction. Without access to future conversational context, both humans and models experience substantial performance degradation when moving from offline to online settings. To tackle the challenges, we propose Online-MMSI-VLM, a novel framework based on multimodal large language models. The core innovations of our approach lie in two components: (1) multi-party conversation forecasting, which predicts upcoming speaker turns and utterances in a coarse-to-fine manner; and (2) socially-aware visual prompting, which highlights salient social cues in each video frame using bounding boxes and body keypoints. Our model achieves state-of-the-art results on three tasks across two datasets, significantly outperforming the baseline and demonstrating the effectiveness of Online-MMSI-VLM. Project page: https://sampson-lee.github.io/online-mmsi-project-page.

cs.CV↗