arXiv subjects

Ziyuan Li

Publications and source records attributed to Ziyuan Li.

At least 19 recordsLinked to original sources

arXiv2026

Spin Hamiltonian for Quantum Simulation of Minimally Truncated $3+1$D SU(2) Lattice Gauge Theory and Testing Eigenstate Thermalization

We construct the spatial-gauge-singlet Hamiltonian of pure SU(2) lattice gauge theory in $3+1$ dimensions ($3+1$D) on qubits under the minimal truncation $j_{\rm max}=1/2$, building on previous work on the hyperhoneycomb lattice, and test the eigenstate thermalization hypothesis (ETH). By enforcing Gauss's law and exploiting the redundancy among gauge-invariant loop operators, we find the minimal set of independent loop variables that can be bijectively mapped onto qubits, and write the Hamiltonian as a sum of Pauli strings. An $n\times n\times n$ lattice with fixed boundary conditions requires $2(n-1)^3+3(n-1)^2$ qubits. We exactly diagonalize the Hamiltonian on a $2\times 3\times 3$ lattice and identify the coupling range of quantum ergodicity from the level statistics. At the ergodic coupling $g=0.3$, we calculate the matrix elements of local energy density operators and the entanglement entropies of all eigenstates and find them consistent with the ETH. These results show that the $3+1$D truncated SU(2) lattice gauge theory thermalizes in accordance with the ETH and provide a benchmark for future real-time quantum simulations of non-equilibrium dynamics.

hep-lat↗

arXiv2026

From Debate to Decision: Conformal Social Choice for Safe Multi-Agent Deliberation

Multi-agent debate improves LLM reasoning, yet agreement is not evidence of correctness. When agents converge on a wrong answer through social reinforcement, consensus stopping commits that error to automated action with no recourse. We introduce Conformal Social Choice, a post-hoc layer that converts debate outputs into calibrated act-versus-escalate decisions. Verbalized probability distributions from heterogeneous agents are aggregated via a linear opinion pool and calibrated with split conformal prediction, yielding fixed-round marginal coverage: at any pre-specified round, the correct answer is included with probability ${\geq}\,1{-}α$, without assumptions on individual model calibration. A hierarchical action policy maps singletons to autonomous action and larger sets to human escalation. Across eight MMLU-Pro domains and two additional benchmarks, fixed-round coverage stays close to target. The key finding is not that debate becomes more accurate, but that the conformal layer makes its failures actionable: at the final round, 81.9\% of wrong-consensus cases are flagged by non-singleton sets at $α{=}0.05$. Separately, among cases that reach a singleton before the final round, accuracy reaches 90.0--96.8\% on MMLU-Pro (up to 22.1pp above consensus stopping): a selection effect, not a reasoning improvement. Simple probability- and entropy-threshold rules attain comparable interception; conformal's distinct value is a finite-sample fixed-round coverage guarantee together with explicit candidate sets for escalation. Adaptive first-singleton stopping is evaluated empirically and not covered by this guarantee.

cs.AI↗

arXiv2026

LoopWAM: Closed-Loop Generalist Policy Improvement from Deployment Rollouts

Scaling expert demonstrations alone does not enable generalist robot policies to improve from deployment. Real-world rollouts contain successes, suboptimal behavior, and failures: imitating every recorded action can reinforce poor behavior, while filtering out failures discards evidence of physical interactions and their consequences. We introduce LoopWAM, a closed-loop world-action model that learns directly from this full quality spectrum, using rollouts from both itself and other deployed policies. Source-dependent noise sampling unifies behavior learning and action-conditioned world modeling, while quality steering and quality-weighted supervision retain full video supervision across qualities and regulate action imitation. Video and action tokens interact throughout parallel denoising, allowing learned consequences to inform action generation directly. On RoboChallenge Table30-V2, three deployment-and-training rounds raise success from 23.67% to 43.33%, a 19.7-percentage-point gain, with a final score of 56.80. LoopWAM also achieves 94.5% average success on RoboTwin 2.0 and 49.5% overall success on RoboCasa365 without additional embodied pretraining, obtaining the best overall results among the compared methods across all three benchmarks. On challenging shirt folding task, two rounds of deployment-driven improvement after single-task post-training raise success from 13.20% to 55.56%. These results demonstrate closed-loop improvement from mixed-quality experience under both multi-task learning and single-task post-training. Project page: https://loop-wam.github.io/

cs.RO↗

arXiv2026

EviGround-CT: Sentence-to-Volume Grounding for Verifying LLM-Generated Chest CT Reports

Chest computed tomography (CT) reports generated by large language models (LLMs) can describe positive findings that lack localized evidence in the scan. We propose EviGround-CT to verify these findings after report generation. A deterministic extractor identifies positive finding sentences, and a 3D grounding model predicts an evidence mask for each sentence. Using the full sentence retains anatomical location, laterality, and descriptive modifiers in the grounding query. We evaluate report generation on CT-RATE, and grounding and verification on a held-out 100-case ReXGroundingCT subset. Using the available annotations, sentence-grounded verification raises evidence-supported precision (ESP) from 0.263 to 0.365 across 137 extracted in-scope positives, compared with category conditioning. The corresponding unsupported positive finding rate falls from 0.737 to 0.635. The paired case-level bootstrap ESP difference is +0.102 (95% confidence interval: [0.028, 0.181]). The predicted masks associate individual report statements with spatial evidence that can be inspected alongside the CT.

cs.CL↗

arXiv2026

Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents

We changed the agent: did it actually get better? Every self-improving agent loop answers this hundreds of times, and every answer comes from a verifier. On open-ended tasks none exists, so the loop is handed a hand-written rubric or a bare LLM judge grading output from a model like itself, inviting reward hacking and shared blind spots. We make the verifier the evolving object: an inspectable expression over small, mostly deterministic drawback detectors, synthesized from clustered failures, gated at birth, and selected for agreement with a ten-item anchored reference set plus consensus over unlabeled outputs, never for the agent's score. On MBPP+ it gains +0.21 held-out agreement over the hand-authored seed composition, on every seed, and ends ahead of the bare LLM judge it contains. One finding should change how co-evolved verifiers are validated: removing the anchor guards collapses the verifier into a vacuous always-pass grader, yet that collapsed verifier trains skills just as well. Downstream task score cannot certify a self-evolved verifier. Score does answer sufficiency, and there an evolved verifier can substitute: Double Ratchet, pairing the verifier with a lifecycle-managed skill loop, retains 88-110% of the lift that ground truth or a rubric buys the same loop, across code generation, enterprise text-to-SQL, and reference-free report generation. When evolved skills gamed the report rubric, an outer judge caught it and one added detector repaired it; the judge itself was wrong until given the task contract.

cs.AI↗

arXiv2026

Self-powered InAs nanowire detector arrays for extended-SWIR spectrometry at room temperature

Spectral sensing in the extended shortwave infrared (e-SWIR) is important for molecular analysis, infrared imaging, and machine vision, motivating the development of compact spectrometers for broader applications. However, conventional commercial off-the-shelf spectrometers in this wavelength region are expensive and bulky due to their reliance on external dispersive optics/filters and/or cryogenic accessories. Other emerging computational spectrometers are based on Si and InGaAs photodetectors that remain focused on the visible and near-infrared, with few detector platforms operating in the e-SWIR regime that simultaneously provide broadband sensitivity, low-noise room-temperature operation, and diverse spectral signatures for accurate identification and reconstruction. Here, we report a room-temperature e-SWIR computational spectrometer based on InAs/InP core-shell nanowire photodetector arrays with geometry-encoded spectral responses. The detectors exhibit self-powered broadband photoresponse across the 1--3 $μ$m range, with responsivity up to 0.215 A W$^{-1}$, detectivity up to $1.6 \times 10^{9}$ cm Hz$^{1/2}$ W$^{-1}$, and microsecond response times. The excellent detector performance is leveraged to demonstrate filter-free spectral reconstruction using a compact multipixel photodetector array device. This enables high-accuracy molecular absorption spectrum reconstruction and hyperspectral imaging. Our results indicate that InAs nanowire arrays are a promising platform for compact computational spectrometry and imaging in the e-SWIR at room temperature.

physics.optics↗

arXiv2026

The Blind Curator: How a Biased Judge Silently Disables Skill Retirement in Self-Evolving Agents

A self-evolving agent retires its bad skills by watching them fail, so what happens when the judge cannot see the failures? Skill retirement is the structural constraint that keeps a growing library from drifting below the no-skill baseline, but its guarantee assumes an unbiased reward, which is false for the LLM judges that reference-free tasks require. We show that a biased judge does not merely add noise; it \emph{silently switches off the curator}. We make this precise with a corrupted-reward analysis, then a behavioral study on a reference-free report-writing testbed with a code-generation cross-check, injecting corruption on top of a deterministic reward to isolate the causal channel. Symmetric noise leaves retirement intact, but \emph{false-pass} bias (failures slipping through as passes) disables contribution-based retirement past a sharp threshold (here a false-pass rate of $0.45$) that no amount of data can cross. Separating genuine retirement from cap-eviction churn shows this \emph{mechanism} failure is universal, holding across domains and failure rates and sparing only near-zero-false-pass, verifier-like graders. The downstream \emph{outcome}, though, is regime-dependent: eval quality degrades only where the same corruption also starves skill synthesis, and otherwise holds steady, so the disabled curator is \emph{silent}, surfacing in no aggregate metric. The contribution is a behavioral safety result, not a performance one. A cheap defect-injection audit then tells an operator, before deployment, which side of the threshold their judge occupies.

cs.AI↗

arXiv2026

Ratchet: How Reliable Must an LLM Judge Be to Retire a Skill?

A large language model (LLM) agent that writes and edits its own skill library must also decide which skills to keep, from one noisy scalar per skill. The answer is exact: a judge scoring failures as passes at rate $(1-τ)/2$ or above retires nothing, at any sample size, for eviction margin $τ$. Audits find that machinery is rarely built: LLM-written skills are worth $+0.0$ percentage points (pp) against a no-skill control, human-written ones $+16.2$pp. Unmaintained, a library enters \emph{library drift}, growing until injecting a skill scores worse than injecting nothing. \textbf{Ratchet} repairs this: it evicts each skill on its measured contribution, caps the library at width $C$, and constrains synthesis, lifting held-out $pass@1$ by $+0.328$ on a hard MBPP+ slice. The matching non-divergence bound is finite for exactly two reasons, $C$ and $τ$. Our contribution is the condition this repair carries and no deployed system states. In reference-free domains the scalar comes from an LLM judge, whose two error directions, modelled as a binary channel, behave nothing alike. Passes scored as failures cost sample efficiency, which more trials buy back; failures scored as passes displace the eviction statistic, and no correction inside the rule recovers it. End-task score is a poor alarm, moving by at most a fifth of the governed lift and not monotonically in the rate. We prove both edges of the certifiable region, confirm them in a running loop, and place a judge on a known side in one offline pass.

cs.AI↗

arXiv2026

PURe: A Plug-and-Play Product-Unit Residual Module for Vision Networks

Modern vision networks are dominated by additive local transformations, whereas explicit multiplicative local interactions remain underexplored. Product units offer a direct approach to modeling such interactions, but their use in deep architectures has been limited by optimization instability. In this work, we propose PURe, a product-unit Residual Module for deep vision networks. PURe is built around a 2D product unit with a real-valued log-domain formulation that makes multiplicative local aggregation practical within deep residual hierarchies. The resulting module serves as a drop-in replacement for native residual units. We instantiate PURe in residual CNNs for image classification and in 2D residual encoder--decoder networks for slice-based segmentation on volumetric CT data. Across Galaxy10 DECaLS, ImageNet, and CIFAR-10, PURe consistently improves residual CNNs and yields a more favorable accuracy--parameter trade-off, allowing moderately deep models to match or surpass substantially deeper ResNet baselines with much smaller parameter budgets. On the AMOS benchmark, PURe also improves slice-based CT segmentation under 3D case-level evaluation. These results show that explicit multiplicative local interaction is a practical and effective design primitive for deep residual vision networks.

cs.CV↗

arXiv2026

The Utility of Sparse Error Detection in Quantum Simulations

The recent success of error detecting codes points toward their potential application to fault-tolerant simulations of nature. In this work, we examine the utility of sparse error detection for simulating lattice gauge theories using quantum computers. In particular, we study the time evolution of the lattice Schwinger model embedded into the Iceberg code family, $[[N+2, N, 2]]$, as well as the Hypercube code family, $[[2^N, N, 2]]$. The lattice of electrons and positrons in the axial gauge is embedded into a single code block or into multiple code blocks, and this work finds that large codeblocks are advantageous in the absence of connectivity constraints. Noisy classical simulations with realistic near-term error rates, infrequent syndrome measurements and physics-aware postselection are found to improve observable estimation. Under realistic noise rates for near-term quantum computers, this work finds that sparse error detection in quantum simulations has the potential to improve accuracy of observable estimation. Additional rounds of error detection are found to systematically drive errors in observables to the noise floor set by the code. These findings suggest that incorporating minimal implementations of fault tolerance in the near-term will enhance the performance of quantum simulations in nuclear physics and high-energy physics.

quant-ph↗

arXiv2026

Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents

Self-evolving agent systems create, revise, and retire their own skills, but every such loop assumes a reliable evaluation metric already exists. In many real applications none does. We show the metric itself can be the evolving object: our loop searches compositions of small typed drawback detectors under a full evolutionary lifecycle, selecting for agreement with a ten-item anchored reference set and regularizing by consensus over unlabeled outputs. What evolves is the function that grades one output, never the fixed task sets it is scored on, and what comes out is an inspectable expression rather than an opaque judge. It is also valid: on code generation it gains 0.21 agreement with hidden ground truth on a locked set that metric selection never reads (paired $p=0.014$), beating the bare LLM judge it contains. Validity is where safety lives: removing the anchor guards collapses the metric into a vacuous always-pass detector while removing the detector lifecycle does not, inverting the lesson from skill evolution. That collapse warns this line of work that downstream task score cannot validate a self-evolved evaluator, since the collapsed metric trains skills just as well. Task score answers only sufficiency, and an evolved metric suffices: \emph{Double Ratchet}, co-evolving the metric with a lifecycle-managed skill loop, retains 88--110\% of the lift ground truth or a hand-written rubric buys, across MBPP+, Spider~2.0-Snow, and report generation. When evolved skills gamed the report rubric, an independent judge caught it and one added detector repaired it.

cs.AI↗

arXiv2026

Library Drift: Diagnosing and Fixing a Silent Failure Mode in Self-Evolving LLM Skill Libraries

Self-evolving skill libraries face a silent failure mode we term \emph{library drift}: unbounded skill accumulation without outcome-driven lifecycle management causes retrieval degradation, false-positive injections, and performance stagnation. Recent evaluation confirms the symptom (LLM-authored skills deliver +0.0pp gain while human-curated ones deliver +16.2pp (SkillsBench)), yet the underlying mechanism has not been isolated. We provide (1) a reproducible trigger: ablations that isolate drift: one disables skill injection (flat floor, +0.002), one imposes premature retirement (active harm, $-$0.019); (2) trace-level diagnostics: an append-only evidence log with per-skill contribution scores, attribution verdicts, and router engagement metrics that make the failure visible before it reaches end-task scores; and (3) a verified fix: a minimal governance recipe (outcome-driven retirement + bounded active-cap + meta-skill authoring prior) that lifts held-out pass@1 from a 0.258 baseline to a late-window mean of 0.584 (rolling gain $+$0.328) on MBPP+ hard-100 over 100 rounds. Eight ablations decompose which governance mechanisms are load-bearing and which are subsumed, providing a concrete playbook for diagnosing library drift in any self-evolving agent.

cs.AI↗

arXiv2026

Realizing Error Suppression in Partially Fault-Tolerant Quantum Simulations with IBM Quantum Computers

Quantum error-detecting codes offer a near-term path for improving the performance of quantum simulations on noisy hardware. Using IBM's superconducting quantum computer ibm_boston, we show that partially fault-tolerant encoded quantum simulations of the Ising model in 1+1D and 2+1D outperform their unencoded counterparts in estimating local observables. To represent 42 logical qubits on the heavy-hex quantum processor, 21 blocks of the [[4, 2, 2]] Iceberg code and up to 136 physical qubits are used. By pairing fault-tolerant syndrome extraction with non-fault-tolerant logical operations, this scheme preserves many of the benefits of error detection while avoiding the overhead typically required for a fully fault-tolerant logical gate set. The encoding's square logical connectivity, together with the freedom to place logical qubits within each block, enables simulations of a 2D spatial lattice with lower circuit depth than the unencoded implementation requires. We introduce Observable-Ranked Postselection, a selective-filtering technique based on syndrome correlations that recovers reliable results without the prohibitive shot loss of full syndrome postselection. Under the cumulative effect of device errors, this encoding improves local-observable accuracy over the unencoded baseline by 2-6% at intermediate times in 1+1D simulations, growing with circuit depth to over 200% in 2+1D at the latest times studied.

quant-ph↗

arXiv2026

EVA-Net: Subject-Independent EEG Motor Decoding with Video-Derived Motor Priors

Practical non-invasive Brain-Computer Interface (BCI) systems require EEG decoders with strong cross-subject generalization and minimal calibration. However, inter-subject variability and signal non-stationarity often entangle motor semantics with subject-specific noise, limiting subject-independent decoding. Recent multimodal approaches use text as a semantic anchor, yet text provides sparse and static supervision for inherently dynamic motor processes. To address this issue, we propose EVA-Net, a two-stage framework that uses action videos as semantic priors for subject-independent EEG motor decoding. In the first stage, EEG and video features are aligned in a shared space using cross-modal and supervised contrastive objectives to reduce subject-specific variation. In the second stage, video category prototypes and knowledge distillation transfer video-derived priors to an EEG-only classifier without adding inference overhead. Experiments on two public datasets show that EVA-Net achieves strong subject-independent decoding performance, including an 8.66% LOSO accuracy gain on EEGMMI. Ablation results further suggest that video provides a more effective semantic anchor than the text baseline considered in this work.

cs.AI↗

arXiv2026

Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents

As LLM agents scale to long-horizon, multi-session deployments, efficiently managing accumulated experience becomes a critical bottleneck. Agent memory systems and agent skill discovery both address this challenge, extracting reusable knowledge from interaction traces, yet a citation analysis of 1{,}136 references across 22 primary papers reveals a cross-community citation rate below 1\%. We propose the \emph{Experience Compression Spectrum}, a unifying framework that positions memory, skills, and rules as points along a single axis of increasing compression (5--20$\times$ for episodic memory, 50--500$\times$ for procedural skills, 1{,}000$\times$+ for declarative rules), directly reducing context consumption, retrieval latency, and compute overhead. Mapping 20+ systems onto this spectrum reveals that every system operates at a fixed, predetermined compression level: none supports adaptive cross-level compression, a gap we term the \emph{missing diagonal}. We further show that specialization alone is insufficient (both communities independently solve shared sub-problems without exchanging solutions), that evaluation methods are tightly coupled to compression levels, that transferability increases with compression at the cost of specificity, and that knowledge lifecycle management remains largely neglected. We articulate open problems and design principles for scalable, full-spectrum agent learning systems.

cs.AI↗

arXiv2026

PU-UNet: Stable Multiplicative Interactions for Medical Image Segmentation

Many dense prediction networks rely on additive feature transformations and model higher-order feature interactions only implicitly. Product units provide an explicit mechanism for multiplicative feature modeling, but their logarithmic--exponential formulation can cause numerical instability, which has limited their use in deep dense prediction networks. In this work, we propose Product-Unit U-Net (PU-UNet), a residual U-Net that integrates stable product-unit residual blocks into rich low-resolution stages for medical image segmentation. The proposed formulation combines smooth positivity mapping with log-domain clipping, enabling stable multiplicative feature learning with negligible computational overhead. On ISIC 2018, Kvasir-SEG, and BUSI, PU-UNet achieves Dice scores of 0.942, 0.959, and up to 0.925, respectively. Compared with a matched Residual U-Net baseline, PU-UNet consistently improves Dice and IoU while keeping parameters, FLOPs, and inference latency nearly unchanged, and reduces the image-level false-positive rate on normal BUSI cases from 0.077 to zero. Ablation studies suggest that the gains are associated with product-unit interactions, are strongest under low-resolution placement, and benefit from the proposed stabilization design. These results suggest that stable product-unit residual learning can be an effective way to enhance U-Net-style segmentation networks with explicit multiplicative interactions.

cs.CV↗

arXiv2026

Product units in gated recurrent units improve nuclear-mass prediction

The prediction of masses of atomic nuclei using machine learning can complement theoretical models and advance the exploration of poorly known domains of the nuclear chart. We propose a machine learning technique based on gated recurrent units (GRU), which have demonstrated competitive performance in nuclear-mass prediction by exploiting long-term dependencies. By integrating multiplicative interactions and product-unit transformations within recurrent units, we report significant improvements in nuclear-mass prediction. Computations are performed in the complex domain to jointly capture amplitude and phase dynamics. For interpolation and temporal-extrapolation tasks based on the atomic mass evaluation (AME2016 and AME2020), the complex additive-multiplicative product-unit gated recurrent unit (AM-PU-GRU) model consistently achieves the lowest prediction errors, with an interpolation RMSE of 0.227 $\pm$ 0.004 MeV and an extrapolation RMSE of 0.179 $\pm$ 0.015 MeV. These results surpass other state-of-the-art machine learning models and also outperform the real-valued GRU baseline and product-unit ablation variants, while remaining robust to different theoretical priors, including WS4 and SEMF. Our findings establish complex-valued product-unit recurrent networks as a new benchmark for sequence-based nuclear-mass prediction.

cs.LG↗

arXiv2026

Modeling Nonlinear Feature Interactions with Product-Unit Residual Networks

Understanding nonlinear feature interactions is crucial in science and engineering, yet standard multilayer perceptrons (MLPs) often capture such interactions only implicitly, leading to entangled representations that can impair robustness and interpretability. We investigate product-unit residual networks (PURe) that integrate multiplicative product units with residual connections to explicitly model cross-feature couplings while stabilizing optimization. We conduct a systematic evaluation on an interaction-driven synthetic benchmark and two real-world datasets, assessing predictive accuracy, robustness to Gaussian feature noise, and performance under limited training data, and we compare real- and complex-valued variants under a matched parameter budget. Beyond accuracy, SHapley Additive exPlanations (SHAP)-based interaction analyses show that PURe learns more concentrated and structurally coherent interaction patterns than MLP baselines. Overall, PURe achieves competitive or improved performance, better robustness and sample efficiency in low-data regimes, and enhanced interaction-level interpretability.

cs.LG↗