arXiv subjects

Ying Li

Publications and source records attributed to Ying Li.

At least 19 recordsLinked to original sources

arXiv2026

Brilliance: toward autonomous computational research in particle accelerator physics

Computational research in physics requires continuous interplay between simulation and scientific interpretation, demanding substantial specialist effort. Here we introduce Brilliance, a research agent that combines large language models, reusable physics knowledge and numerical simulation to conduct adaptive investigations in accelerator physics. Given a single user prompt, Brilliance designs a fourth-generation synchrotron light source that meets coupled emittance and nonlinear-acceptance requirements. Further exploration yields a design combining sub-10-pm emittance with a large dynamic aperture under realistic engineering constraints, suggesting a route towards hard-X-ray diffraction-limited operation, although the momentum-acceptance target remains unmet. Brilliance also calibrates a model of an operating light source against measured responses, accurately inferring its lattice and calibration errors. Across these studies, Brilliance autonomously revises its computational strategies in response to numerical evidence. These results demonstrate a route towards delegating sustained scientific investigations to agents equipped with domain expertise and numerical tools.

physics.acc-ph↗

arXiv2026

When to Intervene? An Empirical Study of Errors and Self-Recovery in LLM Agents

LLM agents interact with Web services to perform tasks on users' behalf. Errors during these interactions can disrupt execution and prevent task completion, motivating external intervention. However, agents may self-recover through continued execution, raising the question of when to intervene and when to wait for self-recovery. We present an empirical study of errors and self-recovery in LLM agents. We construct ReInBench, a dataset containing 1,259 agent traces and 1,403 error events across computer-use and Web tasks. Our study examines how often and how quickly agents self-recover, how self-recovery relates to subsequent execution and task completion, and whether self-recovery can be predicted when an error occurs. We find that 725 error events self-recover, often within a few steps, with self-recovery performance differing across Execution LLMs. Self-recovery can also help agents recover from subsequent errors of the same type and support overall task completion. Intervention should therefore prioritize errors unlikely to self-recover, while allowing agents to continue when self-recovery is likely. Building on these empirical findings, we propose ReCoF, a method that predicts whether an error will self-recover to inform whether additional intervention is needed. ReCoF achieves an average relative improvement of 9.02% in Macro-F1 over the strongest baseline on ReInBench. Our code and data can be found at https://method-notes.github.io/ReCoF/.

cs.AI↗

arXiv2026

LAW: Unifying Language, Action, and World Modeling for Agentic Physical Intelligence

World-action models jointly predict actions and future world states, but long-horizon manipulation further requires language reasoning to plan subtask objectives, assess execution progress, and guide action selection. We present LAW, a closed-loop VLM--WAM framework that unifies language reasoning, action generation, and world modeling through semantic subtask progress. Within the LAW agent, a hybrid-expert vision-language model plans the remaining subtasks, estimates the values of and fuses actions proposed by a world-action model, and asynchronously assesses execution status to support continued execution and subtask transitions. Within the world-action model, Mixture-of-Transformer Experts integrates modality-specific and shared computation for joint video--action modeling. Dual-stream hierarchical visual memory maintains temporal context for both models, while subtask-local value learning aligns action evaluation with semantic objectives and preserves supervision from successful prefixes of failed trajectories. LAW achieves 94.21% average success across 50 RoboTwin tasks. Replacing the WAM backbone with OpenWAM-$alpha$ further raises the average success rate to 95.56%, establishing a new open-source state of the art on RoboTwin. Despite having only 2B parameters, our VLM also outperforms variants using Embodied-R1.5 and GPT-5.5 High. On three long-horizon real-robot tasks, LAW achieves 65% average success, compared with 28.33% for $pi_{0.5}$ and 25% for Fast-WAM.

cs.RO↗

arXiv2026

Continuous-depth Deep Gaussian Processes

Deep Gaussian processes (DGPs) are appealing Bayesian regression models, but their standard discrete-depth parameterization often becomes harder to optimize as depth grows. We argue that depth is better treated as a latent continuous evolution than as a long discrete stack. We therefore propose a continuous-depth formulation in which the latent representation follows a stochastic differential equation with a depth-indexed drift family equipped with a GP-based prior. This reframes the core design choice as the prior placed on that family, rather than adopting a purely neural continuous-depth parameterization as in neural ODE or neural SDE models. We study two instantiations of this idea: CDGP, a direct GP prior over the state--depth drift, and FlowDGP, a flow-evolved prior that captures richer depth dependence at lower training cost. To motivate the redesign, we analyze how the single-sample DGP objective can become increasingly sensitive under repeated layer composition, while residual DGP is a useful first repair that still remains a finite discrete construction. We also give pathwise sensitivity analysis for the continuous-depth formulation. Experiments on synthetic, benchmark, and large-scale regression tasks show that residual DGP improves over plain DGP, while CDGP and FlowDGP provide a stronger overall story across the regimes we study.

stat.ML↗

arXiv2026

Flow-based Spectral Kernel Learning for Nonstationary Attention

Kernelized attention provides a useful perspective for understanding self-attention as a similarity-based aggregation mechanism, but existing approaches often rely on fixed or stationary kernels, limiting their ability to adapt the similarity structure to complex token interactions. In this paper, we propose flow-based spectral kernel attention (FSKA), a framework for learning expressive nonstationary attention kernels. FSKA formulates kernelized attention through spectral density learning and parameterizes the resulting bivariate spectral density with normalizing flows. This design enables flexible and regularizable kernel learning while remaining compatible with scalable linear-attention computation. We further show that expressive kernel learning enables a simplified attention design with removed query--key projections and fixed orthogonal value projections, resulting in a more parameter-efficient architecture. Controlled experiments on classification, ablation, and scalability profiling show that FSKA achieves competitive or improved performance over existing attention mechanisms, while reducing projection-related trainable parameters by approximately $40\%$--$47\%$ compared with most baselines and maintaining favorable linear-attention memory scaling. Code Available: https://github.com/ziyang18/FSKA.

stat.ML↗

arXiv2026

Rate and depth of two-axis twisting by pair exchange under the self-Kerr constraint

In this paper, we study two bosonic modes that exchange pairs of quanta, which realize two-axis twisting exactly: within a sector of fixed particle number the exchange operator is proportional to the two-axis twisting generator, with a fixed factor of two. Because any real nonlinear medium supplies a local Kerr nonlinearity alongside this operator, the two merge into one twisting Hamiltonian governed by a single dimensionless ratio $r=χ_s/χ_e$ of the Kerr strength to the exchange strength. Two conditions then decide how much of the ideal behavior survives this constraint, and they must be met separately. The first concerns speed: the exponential squeezing rate follows in closed form, vanishes exactly at $r=\pm1$, and recovers as the square root of the distance from that point. The second concerns depth: the smallest attainable squeezing parameter depends only on the anisotropy of the mean-field saddle, attaining the full two-axis-twisting value at $r=0$ and $r=\pm3$ regardless of the rate, so that settings sharing the same anisotropy but differing twofold in rate reach the same depth. The two conditions address different targets, and a system may squeeze quickly yet terminate at a shallow optimum; only $r=0$ is optimal in both respects. The closed-form Gaussian solution reveals a further systematic bias: because the early-time decay rate of the variance exceeds twice the exponential rate, a fit to short-time data overestimates the rate by more than a factor of two. Since an isotropic cubic nonlinearity imposes $r=1$, the value at which the rate vanishes, the two strengths must be mismatched, and a moderate mismatch already restores most of the optimal depth. Finally, the constraint is invisible at low excitation: at two quanta the Kerr term shifts both coupled states equally, so the two Hamiltonians are indistinguishable until four quanta.

quant-ph↗

arXiv2026

Low-cost algorithm-to-execution framework for surface-code quantum computing

Fault-tolerant quantum computing is progressing from demonstrations of individual logical components toward the execution of useful quantum algorithms on large-scale error-corrected processors. Realizing this transition requires co-design across algorithms, fault-tolerant implementations, and hardware resources. Here we develop an algorithm-to-execution framework for low-cost surface-code quantum computing that carries the structure of a computation from hierarchical logical descriptions to circuit-specific spatial organization, non-Clifford resource planning, and physical resource evaluation. We apply the framework to twenty benchmark circuits spanning seven algorithm families and to a hierarchically reconstructed application-scale elliptic-curve discrete-logarithm workload. Under the stated error and resource-supply models, combining circuit-specific layouts with hybrid non-Clifford implementation reduces space--time volume by up to 12.70x relative to an all-synthesis Litinski fast-block baseline. Circuit-specific layouts reduce both modeled runtime and space--time volume in eighteen benchmarks, while the remaining two retain the fast layout. These results show that the physical cost of fault-tolerant computation emerges from the interaction between algorithm structure, error-corrected implementation, and hardware organization rather than from aggregate logical resource counts alone.

quant-ph↗

arXiv2026

Dynamic K-Head: Difficulty-Supervised Budget Selection for Continuous Latent Reasoning

Chain-of-thought (CoT) improves reasoning but incurs substantial autoregressive rationale cost. Continuous CoT moves intermediate computation into recurrent hidden states, yet typically uses a fixed number of updates even though the best budget varies across questions and accuracy need not improve monotonically with $K$. We introduce Dynamic K-Head, a question-conditioned, pre-rollout budget selector. It is jointly trained with the reasoner under selected budgets, using difficulty supervision based on scores assigned from the question alone. We also show that latent-update count can misrepresent end-to-end cost because a short latent rollout can shift computation into a longer explicit continuation. We therefore count latent updates, implementation-specific special tokens, and all decoded text. Instantiated with CoDI on Llama 3.2 1B and GSM8K, Dynamic K-Head reaches 54.44% exact-match accuracy at 12.03 average reasoning/output tokens, compared with 52.31% at 15.21 tokens for fixed-$K=6$ CoDI---a 2.12-point gain with 20.9% lower token cost. Across three out-of-distribution arithmetic datasets, it reduces cost by 19.3--25.7% while improving accuracy by 0.30--2.22 points. These results demonstrate that difficulty-supervised budget selection can improve both accuracy and efficiency, while stable budget selection remains a central challenge.

cs.LG↗

arXiv2026

ReCast: Attribution-Oriented Step Representation Learning for LLM-Based Agent Systems

In LLM-based agent systems, failures can originate from early steps whose effects propagate through subsequent interactions, making their origins difficult to identify. To trace such failures back to their origin, failure attribution has been formulated as the task of identifying the earliest step responsible for the failure. Recent methods leverage LLM internal signals for failure attribution, typically using hidden states as step representations. We therefore conduct an empirical study to evaluate how effectively these representations distinguish root-cause steps from other steps and find limited separation. Motivated by this observation, we propose ReCast, a step representation learning method that transforms hidden states from a frozen LLM into attribution-oriented step representations. ReCast first selects attribution-relevant layers, then constructs complementary pattern and deviation features, and finally learns contextualized step representations through an encoder trained with contrastive and ranking objectives. We also introduce ReCast-2K, a training dataset for failure attribution. ReCast achieves the best Hit@1 across four benchmarks, surpassing the strongest baseline by 5.65 and 9.19 pp on Who&When Algorithm and Handcrafted, respectively. Code is available at https://anonymous.4open.science/r/ReCast-5FB6 .

cs.AI↗

arXiv2026

Charging Dicke and Tavis--Cummings quantum batteries with a $\mathcal{PT}$-symmetric lossy--gain cavity pair

We study how a parity--time-symmetric photonic environment changes the charging of Dicke and Tavis--Cummings quantum batteries. For atoms prepared in the collective ground state the excitation number grows as the square of the charging time, with a coefficient set by the initial mean photon number; the counter-rotating terms of the Dicke model add one to four times that number. Conservation of the total excitation number turns this into an exact ceiling on the stored Tavis--Cummings energy, the transition energy times the smaller of the photon and atom numbers, met numerically to within two per cent, which the Dicke terms overshoot by a few per cent at ultrastrong coupling. We then couple a battery to a lossy--gain pair, keeping the parity--time symmetry intact by putting an identical battery in each cavity; the closed moment system puts the exceptional point at the loss rate. Below it the gain cavity fills and the gain-side battery charges while the lossy-side one stays empty; above it photons are shuttled away and dissipated. Solving the full Lindblad equation with batched Monte--Carlo wave functions, we find that the gain-side battery gains thirty-nine, twenty-six and twenty-eight per cent in stored energy, power and ergotropy in the broken phase and loses sixty-five, thirty-five and sixty-five per cent deep in the unbroken phase, where the lossy-side battery overtakes it. Restoring the counter-rotating terms leaves all of this unchanged within the statistical error; what they add is a reactive oscillation at twice the transition frequency that does no net work but nearly doubles the peak of the instantaneous power. The extra energy is of lower quality, the recoverable fraction dropping from about ninety-eight to ninety per cent. The optimum sits on the broken side, close to but not at the exceptional point.

quant-ph↗

arXiv2026

Random Feature Gaussian Process Attention: Linear-Time Probabilistic Attention with Calibrated Uncertainty

Transformers provide a state-of-the-art modeling framework, yet poor calibration limits their reliability in safety-critical applications. A promising direction addresses this issue by interpreting attention as a Gaussian process (GP) posterior, which enables principled uncertainty calibration but incurs cubic complexity in sequence length due to the inversion of the kernel; although decoupled GP variants reduced the cost to quadratic, the computation remains prohibitive in practice. In this paper, we propose the plug-and-play random Fourier feature Gaussian process attention (RFF-GPA) module, which represents the attention as a GP with a stationary kernel approximated by random Fourier features. This low-rank approximation results in linear-time complexity for approximating the posterior mean and variance, making it far more scalable compared to previous work. Empirical results on multiple real-world datasets show that our attention module improves calibration while maintaining predictive accuracy, and simultaneously reduces computational complexity to linear in the sequence length.

cs.LG↗

arXiv2026

Playing social deduction games with reinforcement fine-tuned large language models

Reinforcement fine-tuning (RFT) is increasingly used in applications where large language models (LLMs) interact with humans and other agents. Here we use social deduction games to study how RFT changes LLMs' social behaviour. We let fine-tuned and base LLM agents play hidden-role games that require hidden-state inference, social reading and vote steering. Our results show that LLM agents do not reliably acquire social-deduction ability by directly optimizing terminal win--loss outcomes, suggesting that final game results provide a sparse and noisy signal for socially interactive learning. However, RFT is particularly effective at improving social reading, including tasks that require agents to infer hidden roles from public discussion, update beliefs over time and predict other agents' future decisions. We further show that RFT can also improve social influence, including tasks that require agents to steer votes, team approvals and collective decisions, although these gains depend more strongly on behaviourally specific rewards and structured interaction settings. Finally, we show that LLMs' ability to play social deduction games can be further improved through multi-agent social-cognitive reinforcement fine-tuning, which combines social-reading and social-influence signals during same-side multi-agent training. These learned behaviours also receive more favourable human evaluations of strategic competence, persuasiveness and social usefulness. Together, these results enrich our understanding of how RFT changes LLMs' social behaviour and provide a step toward a behavioural learning theory for machine social intelligence.

cs.CL↗

arXiv2026

AudioJev: Direct Audio Decisions with Order-Calibrated Probabilities

Audio decisions often depend on evidence that a transcript does not preserve, while their probability estimates can depend on how answer options are ordered. AudioJev maps a waveform, question and supplied alternatives directly to a candidate distribution through one shared full-parameter model. We define order calibration as preserving an answer's probability under meaning-preserving option permutations. Random-derangement SKL training pairs each question with a reordered view in which every alternative changes position, supervises both answers, and aligns the two distributions before applying a symmetric KL penalty. Inference retains a single candidate-scoring forward, with no calibration head or order ensemble. Across three training seeds, AudioJev reaches 68.88%/55.33% mean accuracy on complete MMAU/MMAR and reduces random-order SKL by 43.8%/60.5% relative to the single-view removal ablation. The same model handles intent, environmental sound, note properties, speech activity and conversational transitions. Paired removal ablations and multi-order evaluation measure predictive accuracy and probability stability together, establishing a direct audio interface whose calibration objective acts on candidate meaning rather than presentation position. Inference code and model weights are available at https://github.com/SihanLv/AudioJev-Inference and https://huggingface.co/shlv/AudioJev.

cs.SD↗

arXiv2026

Optimal initial states for quantum Fisher information in linearized cavity optomechanics

We find the optimal initial states of a linearized cavity-optomechanical system for estimating the single-photon coupling by quantum Fisher information. In the red-detuned (beam-splitter) regime the interaction exchanges excitations between the two modes, and for pure inputs the Fisher information is four times the variance of the excitation-exchange operator, reducing the optimization to variance maximization with a closed-form solution. With the mechanical mode in a Fock state the probe is degenerate: any state saturating the photon-number constraint is optimal, and a Fock reference with many phonons enhances the per-photon sensitivity linearly. For an arbitrary fixed mechanical reference it becomes a phase-matched squeezed vacuum, from an exact bound on that coherence obtained by a Lagrange-dual argument. Jointly optimizing both modes at fixed total excitation number, the quantum Fisher information obeys a Heisenberg bound quadratic in the energy, saturated by equal superpositions of the two extremal eigenstates of the generator, equivalently a two-mode NOON state after a balanced beam splitter; balanced product Fock states reach only a linear enhancement. Exact diagonalization in truncated Fock spaces confirms all the results, the familiar single-photon input being the simplest saturating case. Three extensions are worked out: the blue sideband has an unbounded two-mode-squeezing spectrum, so the entangled optimum becomes impossible and the balanced product Fock state is exactly optimal at definite energy; under cavity loss the information accumulation turns linear on the cavity-lifetime timescale, at a renewal-optimal duration of about two and a half lifetimes; and for a state-of-the-art quantum-coherent coupling experiment the per-hour relative precision lies between one part in a million and ten parts in a million, number and thermal references being the most robust.

quant-ph↗

arXiv2026

NHO: A Neural Hamiltonian Operator for Anchor-based Region Localization and Dense Correspondence

Non-rigid partial-to-full shape correspondence from sparse anchors requires identifying the corresponding region on the full surface and recovering dense correspondences between the partial shape and that region. We present NHO, which combines sparse anchors with the intrinsic geometry of the partial shape to learn a neural Hamiltonian operator whose localized eigenspace encodes both the region support and intrinsic coordinates for dense correspondence. NHO parameterizes the Hamiltonian potential as an intrinsic neural field and optimizes it using anchor evidence together with spectral and geometric constraints. To resolve the spatial ambiguity left by sparse anchors, we introduce reciprocal refinement between operator estimation and correspondence recovery. At each round, the current eigenspace provides spectral coordinates and restricts matching to its induced support, while geometrically reliable correspondences provide additional evidence for updating the potential. After refinement, aggregated eigenfunction energy yields the final localization, and the recovered map initializes dense correspondence refinement. Experiments demonstrate competitive accuracy on both tasks and robustness to uniform scaling and rotation.

cs.CV↗

arXiv2026

Simulation-Free Learning of GP-SDEs from Irregular Observations

Gaussian process stochastic differential equations (GP-SDEs) provide a flexible Bayesian model for unknown continuous-time state dynamics with uncertainty quantification, but learning and inference from noisy and irregular observations remain computationally challenging. To address this issue, we propose GP-SDE Matching, a simulation-free variational framework for Bayesian GP drift learning and continuous-time state smoothing. We analytically marginalize the sparse GP posterior to derive a tractable drift-matching objective that accounts for both the posterior mean and uncertainty of the unknown drift. To handle irregular observations, we further introduce an irregular-time-aware variational state posterior that incorporates the actual observation times during both encoding and continuous-time marginal querying. Experiments on the stochastic Lorenz--63 system demonstrate substantially improved drift recovery and state reconstruction under irregular observations, while five system identification benchmarks show robust forecasting under increasing observation sparsity and competitive performance against existing latent-SDE and state-space methods.

cs.LG↗

arXiv2026

Efficient-WAM: A 1B-Parameter World-Action Model with Low-Cost Future Imagination

World-Action Models (WAMs) have emerged as a promising paradigm for embodied control by coupling future visual prediction with action generation. However, many existing WAMs rely on photorealistic future prediction, which incurs high inference latency and makes real-time robot deployment difficult. We introduce Efficient-WAM, a World-Action Model motivated by the idea that a compact future branch can still support effective action generation, even at lower visual fidelity. Efficient-WAM improves inference efficiency via a compact video expert transferred from WAN-2.2-5B, token-sparse video latents, and asymmetric video-action denoising that allocates fewer sampling steps to video than to actions. Instead of prioritizing visual fidelity, Efficient-WAM treats future video prediction as a compact guidance signal for action generation. Experiments on RoboTwin 2.0 and real-world manipulation tasks show that Efficient-WAM maintains strong action performance despite visibly coarse future predictions. Our 1B-parameter model achieves 98 ms per action chunk during physical deployment, over 30x faster than the Motus baseline with comparable task success.

cs.RO↗

arXiv2026

Gradient estimates for the fractional $p$-Laplacian in the superquadratic regime

We establish gradient potential estimates for solutions obtained as limits of approximations (SOLA) to the fractional $p$-Laplace equation with finite signed Radon measure data. Under the assumptions $n\ge2$, $p>2$, $0 p-1$, every SOLA belongs to $W^{1,p-1}_{\mathrm{loc}}$, and its weak gradient satisfies a Wolff potential estimate at every Lebesgue point of the weak gradient. Under the additional condition $sp>n$, the solution has a continuous representative that is Fréchet differentiable at every point where the potential is finite. These results give a partial answer to questions raised by Diening and Nowak [Ann. PDE, \textbf{11}(2025)] and by Diening, Kim, Lee and Nowak~[J. Eur. Math. Soc. (JEMS), 2025] concerning gradient potential estimates for the fractional $p$-Laplacian. The proof combines homogeneous affine decay with constants independent of the affine slope and comparison estimates to obtain an affine excess recurrence. Iteration then yields the Wolff potential bound.

math.AP↗