arXiv subjects

Wei Xia

Publications and source records attributed to Wei Xia.

At least 19 recordsLinked to original sources

arXiv2026

Skeleton Coding: Program Structure as a Specification for Agents

Coding agents are typically provided end-to-end behavioral specifications that the code they produce should satisfy, such as tests and input--output examples. How many such specifications must a developer provide, and which ones, for the agent to produce the developer's intended program? We argue that this cannot be determined a priori, for it depends on the structure of the implementation, which is left to the agent to decide. An agent first commits to a structure---a decomposition into data types, modules, and functions---which fixes a class of candidate programs; behavioral specifications then select a program within that class. A set of tests that suffices for one structure may leave another badly underdetermined, requiring different edge cases or exponentially more examples. Since the structure is chosen by the agent, developers cannot ensure their specification is complete without reverse-engineering it from thousands of lines of generated code. To resolve this, we introduce Skeleton Coding, in which developers jointly specify the structure---in high-level, code-like format---alongside behavioral specifications. Theoretically, we model AI-assisted coding as a control problem, and show that skeletons can reduce exponentially the number of tests developers must supply to steer an agent toward an acceptable program. Empirically, across four models and three benchmarks, the best-performing skeleton specifications reduce mean test error by 48.4% relative to the best end-to-end behavioral specifications. Even when the agent can query a complete reference implementation, adding a skeleton reduces mean test error relatively by up to 27.0%. Skeletons also narrow the average gap between small and large models by about 52.4% while lowering implementation cost by 15--26x, and reduce code erosion by 24.7% on average.

cs.SE↗

arXiv2026

When Should LLMs Reason? Adaptive Inference Routing via Entropy Dynamics

Chain-of-thought (CoT) reasoning can improve LLM performance, but its benefits vary across tasks, models, and individual queries. On some tasks, CoT provides marginal or negative accuracy gains while substantially increasing token consumption. We investigate whether early decoding dynamics can help identify when explicit reasoning is useful. Our analysis shows that tasks with larger CoT gains tend to exhibit decreasing early-stage entropy, while low-gain tasks more often show oscillatory or increasing trajectories. These patterns motivate an empirical regime interpretation of decoding behavior. We propose \textbf{EDRM} (Entropy Dynamics-based Reasoning Manifold), a lightweight routing framework whose default heuristic requires no router training. EDRM summarizes early entropy trajectories with a compact, interpretable representation and selects among Direct, Standard, and CoT decoding. Across 15 benchmarks and 7 LLMs, its global and instance-level variants offer \textbf{16--55\%} token savings at comparable accuracy. The Full instance-level configuration combines routing with a short Direct fallback branch; our heuristic-router ablation reports up to \textbf{4.9 percentage points} of additional accuracy from oracle answer selection over routing without fallback. These results support selective reasoning based on early decoding signals and demonstrate the potential of entropy-based control for efficient LLM inference.

cs.LG↗

arXiv2026

Beyond Clean Accuracy: The Practical Challenges and Benefits of Pixel-Based Language Models

Pixel-based language models process rendered text without relying on a fixed subword vocabulary, offering a promising approach to noisy and multilingual inputs. However, their practical value depends on whether they can be reliably adapted to downstream tasks. To this end, we systematically examine this question through fine-tuning behavior, typographical robustness, and zero-shot crosslingual transfer. Under our tested settings, pixel-based models require longer training and broader labelled-data coverage than text-based models, while delayed learning and learning-rate-dependent failures make adaptation less reliable. Our analyses further suggest that task-relevant information is less readily accessible in pretrained pixel representations, placing greater demands on fine-tuning. Once successfully adapted, however, pixel-based models lose substantially less accuracy under typographical noise and achieve higher target-language accuracy than English-pretrained text-based counterparts on two crosslingual benchmarks, despite weaker clean English performance. Together, these findings reveal a central trade-off: replacing tokenization with rendering reduces sensitivity to corrupted and unfamiliar inputs, but shifts a central bottleneck to downstream adaptation.

cs.CL↗

arXiv2026

Resolutions of compact balanced spaces

Let $X$ be a compact reduced balanced complex space of pure dimension $n\geq2$. We prove that every resolution of singularities of $X$ is a balanced manifold; in particular, $X$ belongs to the class $\mathscr{B}$. This removes the additional cohomological assumptions appearing in earlier results of Fu-Meng-Xia. The main new ingredient is the linear independence of the Aeppli classes of the exceptional prime divisors of a projective bimeromorphic morphism onto a normal complex space. This follows from the vanishing of curve degrees for Aeppli-trivial divisors and the relative negativity lemma. For non-normal spaces, strict positivity is used to eliminate divisors mapping onto codimension-one subsets before passing to the normalization. We also recall a classical surface example of Grauert showing that the inclusion of compact balanced spaces in $\mathscr{B}$ is strict.

math.CV↗

arXiv2026

EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering

Test-time context optimization adapts frozen language models through textual feedback, but ordinary conditional generation provides no separate control over how strongly to reinforce the distributional changes induced by a context update. We introduce EMO, a training-free framework inspired by expectation--maximization that alternates question-specific context estimation (E-step) and distribution refinement (M-step). The E-step revises a system context using textual gradients from an external judge without access to gold answers. The M-step contrasts next-token distributions under the revised and preceding contexts at a shared prefix to construct a confidence-constrained logit correction. Feedback on the steered response drives the next context update, coupling adaptation across rounds with intervention within each generation. Theoretically, we connect parameter and context optimization through a shared target-KL objective and additive logit representations, and derive the steering vector from context-induced logit displacements. Under context-improvement and faithful-steering assumptions, we prove a one-round decrease in target KL divergence for sufficiently small steering strength. Under greedy decoding across eight models and six benchmarks, EMO improves the zero-shot baseline in 39 of 40 aggregated settings, averaging $+9.10$ percentage points. With the same judge, it adds $+1.53$ points on average over context-only optimization (TextGrad), with consistent MMLU-Pro gains across models. At matched target-model generation budgets, accuracy advantages over best-of-$N$ are strongest on Llama, while oracle coverage is higher in most comparisons despite fewer candidates. The coverage--selection gap motivates allocating test-time computation jointly to candidate generation and verification.

cs.CL↗

arXiv2026

DiarizationLM: Speaker Diarization Post-Processing with Large Language Models

In this paper, we introduce DiarizationLM, a framework to leverage large language models (LLM) to post-process the outputs from a speaker diarization system. Various goals can be achieved with the proposed framework, such as improving the readability of the diarized transcript, or reducing the word diarization error rate (WDER). In this framework, the outputs of the automatic speech recognition (ASR) and speaker diarization systems are represented as a compact textual format, which is included in the prompt to an optionally finetuned LLM. The outputs of the LLM can be used as the refined diarization results with the desired enhancement. As a post-processing step, this framework can be easily applied to any off-the-shelf ASR and speaker diarization systems without retraining existing components. Our experiments show that a finetuned PaLM 2-S model can reduce the WDER by rel. 55.5% on the Fisher telephone conversation dataset, and rel. 44.9% on the Callhome English dataset.

eess.AS↗

arXiv2026

e1: Learning Adaptive Control of Reasoning Effort

Increasing the thinking budget of AI models can significantly improve accuracy, but not all questions warrant the same amount of reasoning. Users may prefer to allocate different amounts of reasoning effort depending on how they value output quality versus latency and cost. To leverage this tradeoff effectively, users need fine-grained control over the amount of thinking used for a particular query, but few approaches enable such control. Existing methods require users to specify the absolute number of desired tokens, but this requires knowing the difficulty of the problem beforehand to appropriately set the token budget for a query. To address these issues, we propose \textit{Adaptive Effort Control}, a self-adaptive reinforcement learning method that trains models to use a user-specified fraction of tokens relative to the current average chain-of-thought length for each query. This approach eliminates dataset- and phase-specific hyperparameter tuning. Users can dynamically adjust the cost-accuracy trade-off through a continuous effort parameter specified at inference time. We observe that the model automatically learns to allocate resources proportionally to the task difficulty. Across model scales ranging from 1.5B to 32B parameters, our approach reduces chain-of-thought length on math tasks by 1.9-3.7x at matched accuracy relative to the base model used for RL training.

cs.AI↗

arXiv2026

On the Clock: Towards Punctual and Productive Time-Budgeted AI Agents

We study whether small LLM agents can operate effectively under explicit wall-clock time budgets by both respecting the allocated runtime and using available time productively. We evaluate Qwen3.6-27B on five competitions from MLE-Bench Lite and Qwen3-4B on Zork I (Jericho), two agentic benchmarks where additional computational time can meaningfully improve performance. In the simplest setting, where the budget is stated only in the prompt, agents fail to translate the stated budget into controlled use of time. These failures arise from gaps in time awareness, since the harness provides no timing feedback, but also because they cannot reliably anticipate the duration of actions, and do not have a learned mapping from available time to an appropriate strategy. We investigate two complementary classes of interventions: harness-based mechanisms that expose timing information and enforce deadlines, and reinforcement learning with budget-aware rewards. Injecting timing information through the harness substantially improves budget adherence for Qwen3.6-27B without measurable loss in performance, while enforcement hooks tighten adherence further. RL with GRPO achieves near-perfect budget adherence on Zork I and generalizes to held-out budgets not seen during training, but does not improve task performance over the untrained harness on MLE-Bench. Once agents are made to respect the budget, they still fail to use additional time to improve task performance. RL-trained policies learn when to stop but often fill extra time with repeated actions, and GRPO training on multiple budgets tends to collapse toward the strategy learned for the shortest budget. Our results reveal a gap between time adherence and productive time allocation, which remains a central challenge for budget-conditioned agents.

cs.AI↗

arXiv2026

Theory of the Edge-of-Chaos Advantage in Quantum Reservoir Computing

Numerical studies suggest that quantum reservoir computing performs optimally near the edge of chaos, a behavior commonly attributed to a balance between memory retention and nonlinear information processing. Here we develop an explicit quantum reservoir model and rigorously analyze its chaotic regime and integrable point. In the chaotic regime, we show that Gaussian-unitary-ensemble statistics lead to rapid memory loss: learning performance decays exponentially with the storage time required by a temporal task. This decay limits the ability of chaotic reservoirs to learn long-range temporal dependencies. At the integrable point, described by a free-fermion model, we prove that amplitude encoding and a linear readout of fermionic two-point correlations cannot learn a representative class of nonlinear tasks coupling present and past inputs. Introducing interactions alleviates this obstruction. By identifying distinct limitations on memory retention in the chaotic regime and nonlinear processing at the integrable point, our analysis provides a theoretical basis for the observed edge-of-chaos advantage and guides the design of quantum reservoirs for temporal learning.

quant-ph↗

arXiv2026

Higher-order pruning of experts in mixture-of-experts language models

Mixture-of-Experts (MoE) language models suffer from large parameter counts, which create a significant memory bottleneck. Expert pruning is the most direct approach for reducing this parameter count, yet existing methods make pruning decisions for each expert independently, and assume experts' contributions are purely additive. In reality, expert usage in MoEs is inherently cooperative. We derive HOPE (Higher-Order Pruning of Experts), a second-order pruning objective which provably minimizes an upper bound on the error resulting from pruning. We show that REAP (a state-of-the-art first-order pruning method) is a special case of HOPE where interaction terms are ignored. Across three frontier MoE models (up to 122B parameters), two distinct calibration sets, and multiple benchmarks (including math, instruction following, coding, and an agentic suite), we demonstrate that HOPE produces better pruning decisions than existing methods, and its advantage is most pronounced at high pruning rates and on challenging agentic workloads. At 50% pruning, HOPE outperforms all baselines and achieves an average rank of 1.58 out of 5 methods (versus 2.42 for the next-best method, REAP), with gains of up to +6.1% on agentic coding. Over all conditions, HOPE again achieves the best average rank and surpasses every other method in the majority of head-to-head comparisons. By preserving cooperative expert structure that first-order methods ignore, HOPE enables aggressive compression with minimal degradation, particularly on complex tasks where diverse expert combinations are invoked over long sequences.

cs.LG↗

arXiv2026

What Matters in Designing World Action Models: An Empirical Study

World Action Models (WAMs) have emerged as a promising paradigm for generalizable robot control. Despite the growing number of WAM systems, existing works often introduce unified systems that bundle together multiple design choices, such as architecture and training strategy, making it difficult to isolate individual contributions and systematically compare alternative designs. In this work, we present a controlled study that disentangles these design choices and analyzes not only their empirical effects, but also how and why they shape WAMs. More specifically, we focus on three fundamental questions in building WAMs: (1) what causal structure should govern the interaction between world modeling and action generation? (2) in which latent space should world modeling be performed? and (3) how do different world-action modeling objectives affect model behavior and performance? Through structurally controlled experiments on three representative benchmarks, RoboCasa-GR1, LIBERO, and LIBERO-Plus, we systematically compare six causal structures, eight latent representations, and four training objectives, covering popular design choices in existing WAMs. We further validate our key findings on real-robot data from the DROID dataset. We hope to provide a systematic understanding of how core design choices affect world-action modeling and what principles can guide the development of future WAM systems.

cs.RO↗

arXiv2026

Energy asymptotics of holomorphic functions with application to Calderón-Zygmund theory in $\mathbb{C}$

The Calderón-Zygmund theory establishes the boundedness of singular integral operators on $L^p$ spaces for $1 < p < \infty$, yet it encounters a failure at the endpoint $p = 1$. While radial counterexamples in $\mathbb{R}^n$ are well-documented, Pan-Shao-Wang-Wu \cite{psww2026} has showed that every nonconstant holomorphic function provides a counterexample to the Poisson equation within the Calderón-Zygmund framework, with the singular locus being a complex subvariety of codimension one. In this paper, we focus on the complex one-dimensional case and establish stronger results. We prove asymptotic formulas with explicit constants for both the level-set integral and the sublevel-set energy. Then we give simplified proofs of the universal counterexamples to Calderón-Zygmund theory at $p = 1$ in $\mathbb{C}$. Additionally, we construct a new family of counterexamples at the endpoint $p = \infty$, showing that the failure of $W^{2,\infty}$-regularity is also a universal phenomenon in complex one dimension.

math.CV↗

arXiv2026

AffectAgent: Collaborative Multi-Agent Reasoning for Retrieval-Augmented Multimodal Emotion Recognition

LLM-based multimodal emotion recognition relies on static parametric memory and often hallucinates when interpreting nuanced affective states. In this paper, given that single-round retrieval-augmented generation is highly susceptible to modal ambiguity and therefore struggles to capture complex affective dependencies across modalities, we introduce AffectAgent, an affect-oriented multi-agent retrieval-augmented generation framework that leverages collaborative decision-making among agents for fine-grained affective understanding. Specifically, AffectAgent comprises three jointly optimized specialized agents, namely a query planner, an evidence filter, and an emotion generator, which collaboratively perform analytical reasoning to retrieve cross-modal samples, assess evidence, and generate predictions. These agents are optimized end-to-end using Multi-Agent Proximal Policy Optimization (MAPPO) with a shared affective reward to ensure consistent emotion understanding. Furthermore, we introduce Modality-Balancing Mixture of Experts (MB-MoE) and Retrieval-Augmented Adaptive Fusion (RAAF), where MB-MoE dynamically regulates the contributions of different modalities to mitigate representation mismatch caused by cross-modal heterogeneity, while RAAF enhances semantic completion under missing-modality conditions by incorporating retrieved audiovisual embeddings. Extensive experiments on MER-UniBench demonstrate that AffectAgent achieves superior performance across complex scenarios. Our code will be released at: https://github.com/Wz1h1NG/AffectAgent.

cs.CV↗

arXiv2026

DeceptionX: From Multimodal Evidence to Explainable Deception Detection

Deception detection is a critical and highly challenging task within affective computing and behavioral analysis. Existing deep learning methods typically treat this task as a straightforward classification problem; however, this black-box approach lacks interpretability and fails to capture the complex logical deduction processes utilized by human experts when identifying lies. While Multimodal Large Language Models (MLLMs) have shown potential, applying them effectively requires a bridge between low-level audiovisual cues and high-level logical reasoning. In this paper, we propose DeceptionX, a novel MLLM framework that shifts the paradigm of deception detection from black-box classification to an interpretable Observe-Think-Summarize reasoning process. To address the scarcity of high-quality reasoning data, we first constructed DeceptChain, a high-quality dataset developed through a human-in-the-loop process. This dataset synthesizes fine-grained visual and auditory evidence (such as micro-expressions and vocal tremors) into structured chain-of-thought reasoning data. Furthermore, we propose a three-stage training pipeline and a Discrepancy-Aware Redundancy Elimination~(DARE) strategy for DeceptionX to further enhance the model's generalization capabilities. Extensive experiments demonstrate that DeceptionX not only outperforms existing MLLM baselines and state-of-the-art methods on standard real-world benchmarks but also provides transparent, expert-level reasoning paths, bridging the critical gap between accuracy and interpretability in multimodal deception detection.

cs.CV↗

arXiv2026

Re-FORC: Adaptive Reward Prediction for Efficient Chain-of-Thought Reasoning

We propose Re-FORC, an adaptive reward prediction method that, given a query, enables prediction of the expected future rewards as a function of the number of future thinking tokens. Re-FORC trains a lightweight adapter on reasoning models, demonstrating improved prediction with longer reasoning and larger models. Re-FORC enables: 1) early stopping of unpromising reasoning chains, reducing compute by up to 26\% compared to fixed-budget cutoffs, while maintaining accuracy, 2) optimized model and thinking length selection that outperforms the largest model alone--- reaching 1.7 percentage points higher peak accuracy while needing up to 12\% less compute to match the largest model's accuracy, 3) adaptive test-time scaling, which increases accuracy by 9.9 percentage points (on average at maximum compute) over confidence-based baselines. Re-FORC allows dynamic reasoning with length control via cost-per-token thresholds while estimating computation time upfront.

cs.AI↗

arXiv2026

GMoT: Gated Motion-Aware Tokenization for Fine-Grained Micro-Gesture Video Reasoning with Multimodal LLMs

Micro-gesture recognition demands the detection of fleeting, spatially localized movements that are frequently overwhelmed by dominant static appearances and background noise. While Multimodal Large Language Models (MLLMs) excel at general video understanding, they inherently struggle with subtle kinematics and often rely on static posture priors. To this end, we propose GMoT, a Gated Motion-Aware Tokenization module that explicitly distills sparse kinematic evidence into a compact sequence prior to temporal modeling. GMoT dynamically spotlights action-relevant regions via spatially weighted pooling, extracts adjacent-frame temporal differencing to capture precise motion energy, and adaptively fuses these cues into the visual stream using a conservatively initialized semantic gate. To transition from simple classification to evidence-grounded reasoning, we further introduce a progressive reward-guided policy refinement paradigm, supported by a semi-supervised annotation pipeline that generates anatomically focused captions. Beyond achieving the best Top-1 accuracy among the compared methods on iMiGUE (67.32\%) and SMG (73.11\%), improving the Qwen3-VL-8B baseline by +6.80 and +3.11 points, our framework introduces Body-Region Grounding (BRG) Recall as an anatomical-grounding proxy conditioned on correct predictions, together with an overlapping-label cross-domain transfer protocol between iMiGUE and SMG. Extensive evaluations demonstrate that our GMoT-augmented model improves in-domain accuracy, retains clear gains under label-preserving corruptions, and improves accuracy-oriented cross-domain transfer under explicit small-split caveats while maintaining high anatomical grounding in its generated rationales.

cs.CV↗

arXiv2026

Quantum magic and non-commutativity as computational resources in quantum reservoir computing

Quantum reservoir computing (QRC) provides a hardware-efficient paradigm for temporal information processing on near-term quantum devices. Despite rapid experimental progress, a rigorous understanding of the structural conditions required for its scalable quantum-enhanced performance remains lacking. Here, we develop a theoretical framework in Pauli-Liouville space that provides a unified analytical treatment of the echo state property (ESP), nonlinear expressive power, and quantum resources. We first analyze the widely used qubit-resetting scheme and establish that quantum magic generated by reservoir dynamics is a necessary condition for effective computation, a requirement more fundamental than ESP. However, we prove that this architecture faces inherent expressivity limitations: all nonlinear processing originates exclusively from the classical encoding map, imposing an unavoidable trade-off between nonlinearity and memory capacity. To circumvent this structural bottleneck, we rigorously analyze Hamiltonian encoding, in which temporal inputs are embedded directly into the continuous dynamics generator. We show that the ESP is natively guaranteed by the Liouvillian spectral gap, decoupling it from quantum magic. Crucially, for any non-trivial drive Hamiltonian, the discrete-time update map exhibits a transcendental, infinite-order nonlinear dependence on the instantaneous input. Moreover, the intrinsic non-commutativity of the open-system generators governs the temporal coupling of these nonlinearities, producing highly non-separable processing of the input history. Our results establish a rigorous theoretical hierarchy of QRC architectures and provide prescriptive design principles for experiments targeting genuine quantum advantages in temporal processing.

quant-ph↗

arXiv2026

Chiral-Structured Superconductors TrX4 (Tr = Rh, Ir; X = Ge, Si): A Platform for Mixed-Parity Pairing and Topological States

Chiral-structured superconductors, with simultaneous broken mirror and inversion symmetries, promote unconventional superconductivity through parity-mixing mechanisms. Yet a few bulk chiral-structured superconductors are known, partly due to the difficulty in directly determining their atomic-scale chirality. Here we report three chiral-structured superconductors, , RhGe4, IrGe4, and IrSi4, synthesized under high pressure, with Tc values of about 1.6 K, 1.1 K, and 2.5 K, respectively.Using atomic resolution Cs-corrected scanning transmission electron microscopy (STEM) combined with X-ray diffraction characterizations, we directly confirm their chiral structure (space group P3121). This real space imaging approach overcomes ambiguities in traditional diffraction based methods. These materials exhibit type-II superconductivity, and the enhancement of spin-orbit coupling (SOC) leads to the emergence of mixed parity pairing. Calculations also reveal symmetry protected Weyl points near the Fermi level, which is robust against the SOC. Our work not only expands the family of chiral-structured superconductors but also demonstrates the indispensable role of STEM in directly determining chiral crystal structures. These materials thus offer a clean platform to explore the interplay among structural chirality, SOC, mixed parity superconductivity, and topological quantum phenomena.

cond-mat.supr-con↗