arXiv subjects

Zihao Li

Publications and source records attributed to Zihao Li.

At least 19 recordsLinked to original sources

arXiv2026

Multidimensional Online Resource Allocation for Large Language Model Services under Nonstationary Demand

We study multidimensional online generalized assignment under nonstationary arrivals, motivated by resource allocation for large language model agent services. Providers may commit model and tool allowances to agent tasks upon admission, reserving budgets for their internal execution loops. Each request is rejected or accepted with an irrevocable assignment to one compatible server, earning reward and consuming a vector of its nonreplenishing budgets. Arrivals are independent with known, potentially time-varying type distributions. Scalar reductions can discard feasible combinations of requests with complementary resource requirements. To maximize expected reward, we develop Vector-Aware Best-Fit, which combines linear-programming-guided candidate routing with calibrated acceptance while retaining vector feasibility. Relative to the expected clairvoyant offline optimum, its guarantee improves on scalar reduction in the multidimensional case and has the best possible inverse-dimensional order. With an appropriate acceptance parameter, the proposed policy attains the optimal competitive ratio on all-heavy instances, where every feasible assignment consumes more than half of at least one resource budget. We further develop two heuristics that retain its performance guarantee. Computational results show that the heuristics perform well against benchmarks in most tested settings, especially when capacity is scarce relative to demand.

cs.GT↗

arXiv2026

Proposing Better Rollouts for On-Policy Distillation

On-policy distillation (OPD) improves language-model post-training by letting the student learn from its own sampled trajectories under teacher supervision, aligning the training process with inference-time behavior. However, the rollout policy in OPD is usually treated as fixed: each round samples directly from the current student. We show that this choice can create a compounding teacher--student mismatch, because it requires the teacher to provide effective supervision on whichever trajectories the student samples, including ones less reachable by the teacher itself. On such trajectories, teacher supervision can become noisy or even misleading, leading to undesirable student patterns such as repetition. Moreover, since the updated student in turn defines the rollout policy of subsequent rounds, these patterns persist and potentially amplify over training. This failure mode suggests that improving OPD requires treating the rollout policy itself as an optimization target, rather than only modifying the distillation objective. We propose Mixture Proposal On-Policy Distillation (MPOPD), which derives a student-anchored, teacher-guided rollout distribution from a local trust-region objective based on relative teacher support. The resulting target is a geometric mixture of the student and teacher policies, and we introduce a proposal policy that approximates this mixture, enabling rollout generation in a single forward pass. Experiments on 16 math, knowledge, and medical reasoning benchmarks show consistent gains over OPD and recent variants. On math reasoning, our method improves average accuracy by 6.91 over OPD on Qwen3-4B-Base while reducing average generation length by 41%, suggesting that rollout policy design provides an effective way to improve OPD.

cs.LG↗

arXiv2026

Choosing models for network-wide freeway speed estimation and forecasting from sparse fixed sensors

Most freeway agencies have fixed detectors on only part of their network but need speeds for all of it. This paper compares methods for estimating current and next-hour speeds on every link from such sparse detectors, using historical probe data for training. The test network is the Dallas--Fort Worth freeway system: 4,127 directional links, of which 643 (15.6\%) carry a side-fire detector. Eighteen methods from five families (non-learned baselines, statistical estimators, tabular learners, road-graph neural networks and models that learn spatial relations) were trained under one protocol with three random seeds. Each neural architecture was also retrained with a parameter-free, direction-aware kernel interpolation of the sensor readings as additional input. Graph neural networks that pass information along the road graph, including a plain graph convolutional recurrent network, reduced the current-hour error on unsensored links by about 17\% relative to the historical average. Models with adaptive graphs, attention or identity embeddings stayed close to the historical average and did not fit the training data well, indicating that they could not learn which detectors matter for which links. With the interpolation features, all ten architectures came within about 0.1 mph of each other. The remaining error depended mainly on coverage: almost all of the improvement was found within about 3 mi of a detector, and very little on tollways without detectors or on links with no road-graph connection to one. The results are summarised as guidance on model choice and detector placement.

cs.LG↗

arXiv2026

Semantic Atoms for Scalable Text Analysis in Data Analytics in the Era of LLM-Based Agents

Much of the information needed for data analysis is buried in text columns: an analyst may infer customer frustration from complaint text and use it to explain resolution time. We call workloads in which processing text columns is central to subsequent data analysis TADA (\emph{Text Analysis in Data Analytics}). In an agentic setting, the user's query dynamically specifies this processing, and answering it interleaves LLM-based semantic steps with exact SQL or Python operations. This mixed execution raises two design questions: which work should be neural and which symbolic, and which semantic outputs later steps can reuse. We address them for agent-planned TADA workflows with two complementary, query-driven semantic atoms. Each atom takes a sequence of text items and a natural-language query specifying the semantic task: \textbf{Tagging} returns one string per input item, while \textbf{Categorization} returns $M$ category objects over $N$ input items. Tagging results are materialized as a row-aligned column and Categorization results as a category table; SQL or Python then performs exact computation over this exposed state. We realize this interface in TADA-Agent, whose atoms run inside generated Python programs and internally handle inputs beyond a reliable context window. We evaluate TADA-Agent on TABench, a benchmark of 40 complete analysis requests over spreadsheet tables spanning workflow composition, input--output structure, and literature-grounded analytical scenarios. Against a semantic-operator baseline and relative to a held-out TADA reference run, TADA-Agent delivers comparable row-level semantic quality, has higher complete-artifact preservation estimates, and uses over an order of magnitude fewer semantic-model requests under the evaluated native execution paths.

cs.CL↗

arXiv2026

From Self-Confidence to Statistical Accuracy: Scaling Human-in-the-Loop Evaluation with Direct and LLM-Assisted Auditing

For large language models (LLMs) and multimodal large language models (MLLMs), model-reported self-confidence is often miscalibrated, yet it may still be useful for efficient evaluation. At the same time, LLM-as-a-judge offers a scalable alternative to direct human auditing, but its practical benefit depends on calibration quality and inference cost. This paper develops a unified statistical framework for both evaluation routes under alternative sampling strategies. For direct auditing, confidence-stratified sampling is used to recover population accuracy from partially verified outputs without requiring calibrated confidence. Finite-sample uncertainty is characterized, and the analysis shows how highly uneven confidence strata affect audit efficiency and label allocation. For judge-assisted evaluation, the judge is modeled through sensitivity and false-positive rate, yielding an error-corrected estimator with calibration uncertainty. The two routes are then compared through human-label and total-cost efficiency boundaries. Experiments on MMLU-Pro and MMMU-Pro show that miscalibrated self-confidence can still support statistically valid stratified auditing. Under highly imbalanced confidence distributions, proportional allocation substantially reduces estimation error relative to equal-per-stratum allocation. For judge-assisted evaluation, only a small subset of target--verifier pairs satisfies the idealized human-label efficiency condition, and automated inference costs can further limit potential savings. These findings clarify the statistical and economic conditions under which confidence-guided auditing and automated verification are useful.

cs.CL↗

arXiv2026

Modality Disentangled Learning for Incomplete Multimodal Emotion Recognition: A Primitive Memory Distillation Perspective

Multimodal Emotion Recognition (MER) systems often suffer from missing modalities in real-world scenarios. Existing methods usually generate, align, or distill missing modalities as a whole, overlooking the heterogeneous nature of the information carried by each modality. Such holistic treatment mixes inferable shared semantics with uncertain modality-specific details, yielding unstable representations and degrading robustness. To address this issue, we propose the Primitive Memory Distillation (PriMD) framework. Unlike existing methods, PriMD takes an intra-modal perspective and focuses on how different types of information within a modality differ in recoverability. PriMD first disentangles cross-modal shared semantics from modality-specific representations, and then discretizes the latter into learnable semantic primitives to construct modality-specific memory banks. When modalities are missing, PriMD is a teacher-student framework that the student model uses the shared semantics of available modalities as queries to dynamically retrieve primitives. It compensates for missing modality-specific information within a constrained memory space and aligns with the teacher model. Extensive experiments on IEMOCAP, CMU-MOSI, and CMU-MOSEI demonstrate that PriMD achieves state-of-the-art performance and consistently stronger robustness across a wide range of missing-modality settings, while mitigating the instability caused by holistic feature inference. Our code and project website are available at https://www.jiaqizhang-sengoku.com/Code/PriMD and https://www.jiaqizhang-sengoku.com/PriMD/ , respectively.

cs.CV↗

arXiv2026

DIGEST: Discovery-Guided LLM Tuning for Symbolic Regression

Symbolic regression (SR) seeks to discover mathematical expressions directly from data, without requiring task-specific scientific descriptions. Despite tremendous progress in both conventional search-based methods and recent LLM-based methods, two critical limitations largely remain. First (Limited Generation Flexibility), conventional search-based methods rely on either fixed search rules or specialized learned generators, severely limiting adaptive candidate generation across diverse expression structures. Second (Underutilized Search Discoveries), most LLM-based methods cannot reliably recover ground-truth expressions because they fall short in effectively leveraging the LLM itself to thoroughly digest past expressions or adaptively guide future search. To address these limitations, we propose DIGEST. To render greater flexibility in generation, DIGEST repurposes a pretrained decoder block from a lightweight LLM into a symbolic policy for Genetic Programming (GP) population seeding, thereby leveraging the broader modeling capacity of the LLM. To more thoroughly utilize search discoveries, high-quality expressions discovered by GP are fed back to update the LLM-derived policy online, thereby enabling the LLM to progressively internalize past search discoveries and translate them into improved subsequent population generation. Experiments show that on two ground-truth benchmarks, DIGEST outperforms the strongest competitor by up to 2.59% in recovery. Among LLM-based baselines, it improves recovery by at least 54.91% while using at most 17.4% of the strongest competitor's mean search time. On 122 BlackBox datasets, DIGEST achieves the highest accuracy while maintaining a moderate model complexity.

cs.LG↗

arXiv2026

Rational Learning One Step Off the Path

We study how endogenous learning shapes beliefs and strategies in extensive-form games. We adopt the overlapping-generations model of \citet{FL1993}, in which finitely lived Bayesian agents are randomly matched, observe only the outcomes of their own matches, and choose dynamically optimal strategies. We study limiting strategy profiles in steady states as agents become arbitrarily long lived and patient. We show that learning disciplines beliefs and strategies in subgames reached after a single deviation from the path of play. In particular, we establish the converse of \citet{FL2006}, showing that such limiting strategy profiles must be path-equivalent to subgame-confirmed equilibria.

econ.TH↗

arXiv2026

ConQEC: Concolic Testing for Quantum Error-Correcting Code Implementations

Quantum error-correcting code (QECC) implementations are hybrid software-control systems: measured syndromes drive decoder dispatch, recovery actions, and logical checks. Existing QECC verification techniques can prove formal obligations, and general quantum testing can expose final failures, but developers still lack a scalable testing workflow that generates concrete faults and explains where a QECC repair path first diverges. This paper presents ConQEC, to our knowledge the first concolic testing framework for QECC implementations. By combining concrete (Pauli) fault replay with symbolic repair-path constraints, ConQEC generates tests that expose not only final logical failures but also fault--syndrome--decoder--recovery (FSDR) diagnostics: the injected fault, observed syndrome, decoder decision, recovery action, oracle result, and earliest divergent stage. Its symbolic encoding of Pauli faults, syndrome predicates, decoder and recovery relations, and oracle checks enables this diagnostic workflow to scale beyond exponentially increasing amplitude-level quantum-state reasoning. We implemented ConQEC and evaluated it on generated source-level QECC families and public QECC artifacts with up to thousands of qubits. ConQEC detects the seeded bug in every experimental case and reports a replay-valid, fault-localizing FSDR witness for each.

quant-ph↗

arXiv2026

MAMMOTH-LyC: Investigating the Role of Galaxy Mergers in a Strong Lyman Continuum Leaker at $z=2.39$

The MAMMOTH-LyC survey is a cycle 30 Hubble Space Telescope (HST) medium program obtaining 18-orbit-deep WFC3/UVIS F225W imaging in two massive galaxy protocluster fields at $z\sim2.2$. We introduce this survey by reporting the discovery of J1244-LyC1, a strong Lyman continuum (LyC) leaker at $z = 2.39$, exhibiting clear merger signatures. J1244-LyC1 has a highly significant ($6σ$) LyC detection, corresponding to an absolute escape fraction of $f_{\mathrm{esc}} \! =\!36\%\pm6\%$ ($1σ$). The LyC emission is spatially resolved into multiple peaks that coincide with the system's disturbed morphology, confirming genuine multi-site LyC leakage. With a stellar mass of $10^{10.2}{M_\odot}$, J1244-LyC1 is both the first confirmed high-redshift LyC-leaking merger and the most massive LyC emitter known to date. We interpret J1244-LyC1 as a merger-driven starburst system in which tidal interactions have disrupted the interstellar medium, creating multiple low-column-density pathways that facilitate LyC escape. This discovery provides the first direct evidence of spatially resolved LyC escape in a merging system, offering new insight into the potential role of major mergers in driving the cosmic reionization.

astro-ph.GA↗

arXiv2026

Geometry-Aware Latent Representation Learning for Drag Coefficient Prediction of Irregular Particles

Predicting the drag coefficient $C_\mathrm{d}$ of irregular particles requires representing their complex three-dimensional geometry in a form suitable for predictive models. Empirical drag correlations and many data-driven models represent particle geometry through a small set of hand-crafted shape descriptors. Geometric differences omitted by these descriptors are unavailable to the predictor. We propose a variational autoencoder-multilayer perceptron (VAE-MLP) framework that learns a latent representation directly from voxelized particle geometry. Geometry reconstruction and drag coefficient prediction jointly guide the learning of this representation. We compare VAE-MLP with a descriptor-based MLP (Descriptor-MLP) and established empirical correlations using CFD data for synthetic particle geometries at Reynolds numbers of 10--200. When the evaluated geometric variations are represented in the training data, the two learned models achieve comparable accuracy for regular ellipsoids, while VAE-MLP yields lower errors for irregular particles. When high-order surface perturbations are excluded from training, VAE-MLP achieves lower overall mean absolute relative error in all five training runs. At the strongest tested perturbation, the mean errors across these runs are 4.97% for VAE-MLP and 9.88% for Descriptor-MLP. These results support learning geometric representations directly from particle geometry to improve drag prediction for the irregular geometries and unseen surface perturbations examined here.

math.NA↗

arXiv2026

DriveWorld3D Evidence Constrained 3D World Modeling and Generative Simulation for Real-World Autonomous Driving

Three dimensional modeling for autonomous driving must continuously estimate road geometry, obstacle occupancy, traffic participant behavior, and observational uncertainty under limited computational resources. Existing reconstruction methods can synthesize realistic novel views, but visual fidelity does not sufficiently characterize the geometric reliability required for driving. Generative world models can expand scenarios and predict future observations, yet may introduce content unsupported by measurements. We propose DriveWorld3D, a framework combining temporal video learning, measurement constraints from multiple sensors, and offline generative simulation. Its online component maintains a unified world state comprising road surfaces, volumetric occupancy, dynamic entities, and road relations, while separately representing physical occupancy, observability, and evidence provenance. Action conditioned future prediction and risk sensitive planning connect this representation to driving decisions. Its offline component reconstructs recorded scenes using neural radiance fields or 3D Gaussian Splatting, then expands long tail conditions using generative models while restricting the use of generated content as geometric ground truth. We specify data partitioning, sensor ablations, fault injection, paired closed loop testing, and real to simulation consistency assessment. The central hypothesis is that explicit evidence management can reduce dangerous false free space predictions and occlusion risk within a matched computational budget while retaining acceptable driving efficiency. All performance claims remain subject to experimental validation. Keywords: autonomous driving; 3D occupancy; sensor fusion; uncertainty; generative world models; NeRF; 3D Gaussian Splatting; closed loop simulation

cs.RO↗

arXiv2026

Mothership: A Two Tier Autonomous Refuse Collection System for Pedestrian Campuses

Refuse collection in pedestrian campuses, parks and mixed-use districts remains manual because compaction trucks cannot reach bins located on walkways, courtyards and building service doors. This paper presents Mothership, a two-tier autonomous refuse collection system in which an electric compaction truck (the mothership) carries two low-speed sidewalk robots (RoboCarts) that fetch standard two-wheel bins, deliver them to the truck, return the empty bins to their original positions and re-dock on the truck for charging. The architecture separates responsibilities by scale: the truck performs road driving, compaction, lifting and energy supply, while the carts perform only sidewalk transport and bin grasping. This split removes the lift-and-tip mechanism from the sidewalk vehicle, keeping it light and low, and relaxes cart-to-truck docking tolerance to the +/-100 mm capture window of a standard automated lifter. We describe the system architecture, the bin-handling and docking interfaces, and a mission-planning formulation that decomposes a route into truck stop points and cart tasks. A discrete-event simulation over synthetic campus layouts with 2-16 bins/ha quantifies throughput as a function of bin density, cart count and stop-point radius. A two-cart configuration serves 20-29 bins/h. Throughput is governed by an interaction between cart count and stop-point radius: a single cart performs best with many close stops, whereas three or four carts need wide stops (80 m) to stay busy, reaching 35-49 bins/h at moderate-to-high densities. The best stop radius therefore has to be chosen together with the fleet size, not fixed in advance.

cs.RO↗

arXiv2026

Learning Higher-Order Interactions from Incomplete Spatiotemporal Data with Mixtures of Hypergraph Laplacians

Traffic sensor data exhibit strong temporal and cross-sensor dependencies, yet long gaps erase both a sensor's local history and the joint observations needed to learn its relationships with other sensors. We introduce MoHL, a mixture of hypergraph Laplacians that learns group-wise relational structure directly from incomplete spatiotemporal data. MoHL discovers overlapping sensor groups from observed co-variation, constructs a Laplacian component for each interaction order, and learns their mixture through holdouts that mimic the missingness pattern. By sharing weights across group-induced couplings, MoHL pools scattered evidence instead of estimating each coupling independently. We characterize the induced weight-sharing structure and analyze the conditions under which sharing reduces estimation risk. Experiments on three traffic networks show that higher-order groups are most beneficial under long gaps, while support-matched comparisons demonstrate the advantage of shared weights when relational evidence is scarce. MoHL achieves the lowest error among the evaluated baseline methods under day-long gaps, with orders of magnitude fewer learned parameters than deep imputers. These results show how group-wise sharing can sustain reliable reconstruction precisely when missingness depletes the evidence needed to learn cross-sensor dependence.

cs.LG↗

arXiv2026

Before We Act: Learning Action-Conditioned Team-Response Beliefs for Decentralized Multi-Robot Manipulation

In multi-robot manipulation, an action's value depends on how teammates respond. We present \method{} (Cooperative Action--Response Belief Estimation), a framework for learning \emph{team-response beliefs} from local history. A predictive belief encoder represents the interaction and conditions a nominal action. Matched reactive and replay continuations separate each candidate's direct advantage from the utility contributed by teammate adaptation. A candidate-conditioned readout learns this response distribution alongside total advantage, and calibrated selection chooses an alternative or retains the nominal action. Our analysis characterizes when response changes the preferred action and how prediction and source errors limit the recoverable gain. With the representation, proposals, and data fixed, explicit response supervision improves candidate ranking and reduces selection regret relative to learning the same information through branch utilities. The benefit is concentrated in decisions where teammate adaptation changes the preferred action. Across RoboFactory, the MARS Control Track, and physical collaboration tasks, CARE achieves the highest average success among the evaluated policies, combining a capable nominal policy with selective coordination adjustments. Together, these results show that CARE turns predictions of teammate adaptation into effective decentralized action selection. Project page: https://jeong-zju.github.io/care/.

cs.RO↗

arXiv2026

CoRE: Learning Collaboration-Role Experts for Decentralized Collaborative Manipulation with One Policy

Collaborative manipulation requires robots to perform complementary actions as interactions unfold. We study single-policy decentralized collaboration: every robot runs the same policy from its visual observations and proprioception, without task prompts, identity labels, or inter-robot messages. The challenge is to learn complementary team behaviors within shared parameters and select appropriate actions from each robot's local observations. We introduce CoRE, which learns Collaboration-Role Experts from pooled multi-task, multi-robot demonstrations. Fused appearance and geometry provide local interaction evidence. Query-conditioned cross-attention experts provide adaptable prediction paths, which a local router combines at each action-chunk position. During training, an action-expert alignment loss supervises expert selection using relative forced-route prediction errors against demonstrations under fixed inputs, without role labels. Across simulation benchmarks, CoRE achieves the highest average performance among evaluated decentralized methods. Physical experiments demonstrate effective collaboration across diverse manipulation tasks and robustness to partner delays and slowdowns. Project page: https://aus.bot/research/core/.

cs.RO↗

arXiv2026

COSMOS-3D: Spectro-photometric Confirmation of Shabgard, a Cool, Metal-poor Brown Dwarf in the COSMOS Field

Brown dwarfs are unique laboratories for studying cool atmospheres, the low-mass end of star formation, and the transition between stars and giant planets. Identifying and characterizing the coldest brown dwarfs remains challenging, particularly at faint magnitudes in deep extragalactic surveys. We investigate COSMOS2025 ID639188, hereafter Shabgard, a compact source previously selected as a brown-dwarf candidate from COSMOS-Web photometry. We use serendipitous JWST/NIRCam F444W wide-field slitless spectroscopy from COSMOS-3D, combined with COSMOS2025 NIRCam photometry and multi-epoch imaging. We extract a local-background-subtracted, profile-weighted $4.05$-$4.85μ{\rm m}$ spectrum and compare the data with six atmosphere-model families. The LOWZ atmosphere-grid fit, using PCA interpolation and MCMC sampling, gives $T_{\rm eff}=718.6^{+10.6}_{-10.2}$ K and $[{\rm M/H}]=-1.00^{+0.07}_{-0.08}$, favoring a cool, subsolar-metallicity atmosphere near the late-T/Y transition. A fixed-atmosphere molecular-opacity test strongly favors a contribution from CH$_4$ to the observed $4$-$5 μ{\rm m}$ spectral shape, although the restricted wavelength coverage prevents a molecule-by-molecule abundance determination. Model-dependent physical estimates span approximately $23$-$52 {\rm M_{\rm Jup}}$ and $140$-$180$ pc. We measure a proper motion of $|μ|=89.2\pm0.4$ mas yr$^{-1}$, while the LOWZ MCMC flux scaling gives a distance of $176.2^{+18.4}_{-17.8}$pc. Together, these results support the classification of Shabgard as a cool, metal-poor Galactic brown dwarf and demonstrate the ability of JWST deep-field surveys to discover and characterize faint substellar objects.

astro-ph.GA↗

arXiv2026

EvoHarness-RL: Learning Runtime Harness Coordination for Self-Evolving Agents

Long-horizon LLM agents increasingly rely on external execution support to maintain state, track progress, recover from failures, and reuse experience across extended interactions. Yet existing harnesses and their use are often tailored to environments and controlled through prompts, heuristics, or system-specific rules, making agent and harness coordination difficult to jointly optimize. We introduce EvoHarness-RL, a unified framework that separates environment-specific harness implementations from a shared policy-facing interface. EvoHarness-RL organizes external support into a Belief, Progress, and Experience (BPE) workspace and exposes four compact harness actions for accessing and updating this state. We first instantiate BPE as an inference-time scaffold and then make harness coordination learnable through supervised initialization followed by cost-aware GRPO. Across heterogeneous long-horizon tasks, EvoHarness-Base improves the average success rate of frontier models by 10.0 percentage points, while EvoHarness-RL outperforms the strongest open-source baseline by 8.5 percentage points, together with higher RL rollout efficiency and stronger generalization to unseen tasks. Our analyses show that training gradually shifts agents from frequent scaffold use toward selective, environment-dependent harness access as the policy becomes more capable, while the external workspace continues to evolve and refine itself to better support task execution and generalization. Together, these results show that long-horizon agents benefit not only from external scaffolding itself, but also from learning how and when to coordinate with external support as part of a cost-aware policy.

cs.LG↗