arXiv subjects

Zijia Wang

Publications and source records attributed to Zijia Wang.

At least 19 recordsLinked to original sources

arXiv2026

Calibration Is Not Verification: Answer-Level Conformal Guarantees for Mixture-of-Agents

Web answer synthesis with Mixture-of-Agents (MoA) requires reliable factual support for every atomic claim, yet agreement among agents is not verification. Correlated proposers can repeat a shared misconception, and a correct specialist fact receives little support when the other proposers stay silent. We introduce C-MoA, which decomposes MoA outputs into atomic claims, scores inter-agent semantic support, and calibrates claim retention at the level of whole answers. Under exchangeability and a policy frozen before calibration, it provides distribution-free marginal control of the probability that an answer retains any claim unsupported by the reference labels. Because calibration only selects a cutoff on a fixed ranking and cannot repair shared errors, we also study CONTRA-MoA, an optional extension that adds blinded near-miss tournaments, leave-one-agent-out stability, and availability-aware fusion, and we derive a competence--discrimination law for its verifier. Across 300 splits of the FactScore biography benchmark, the mean rate of fully supported answers (Fact) exceeds all six nominal targets, but only $1.7\%$--$9.8\%$ of supported claims are retained and $30\%$--$75\%$ of answers are empty. At $α=0.10$, mean agreement retains $3.1\%$ of supported claims, about $3.4$ times the $0.9\%$ retained by the best single-proposer score. On the medical Golden Dataset, it retains $37.3\%$ at a mean Fact of $0.898$, slightly below the $0.90$ target, and on NQ mean Fact reaches $0.904$. On a separate balanced FactScore cohort, memory-only counterfactual scoring is close to chance (AUC $0.522$ on 182 claims), and raw maximum fusion lowers the agreement AUC from $0.687$ to $0.652$. These results separate the statistical calibration of factuality from informative claim verification and clarify the role of each in multi-agent Web synthesis.

cs.MA↗

arXiv2026

R$^2$ Flow: Recursive Self-Improvement via Recursive Skill Evolution

LLM-based agents can improve themselves across tasks by reusing and revising the skills they orchestrate into executable procedures. Flow-based training fits this loop: it samples procedures in proportion to reward, and the flow through each skill credits it for the next library revision. Three obstacles stand in the way of making this self-improvement reliable: flow training suffers strategy collapse over tree-structured histories; nonnegative flow-based credit rewards frequent use as if it were benefit; and library edits rest on the task reward the policy optimizes. We introduce R$^2$ Flow, a recursive self-improvement framework that alternates policy learning, independent verification, and versioned skill-library updates on a shared-state orchestration graph. The graph merges histories that differ only in the order of independent steps, allowing flow training to pool evidence across equivalent executions. A flow-share readout of the trained flow, invariant to the backward policy, and a separate signed utility rank which skills to change, verifier evidence decides whether an edit is warranted, and a residual-variance plateau sets when to update. Committed edits reshape the graph the next policy learns on, realizing recursive skill evolution. Across question answering, mathematical reasoning, interactive decision making, and code generation, R$^2$ Flow improves task accuracy and library-edit precision over heuristic orchestration, reinforcement learning, and skill-evolution baselines, and transfers across executors. Code is available at https://github.com/beita6969/r2flow.

cs.AI↗

arXiv2026

EvoSteer: Online Self-Evolving Graph Orchestration via Reference-Anchored Credit Assignment

In recent years, LLM-based multi-agent systems have been widely applied to orchestrate tool-using agents into executable communication graphs. However, existing self-evolving orchestration still faces key challenges, including post-hoc evolution that revises the team only after the trajectory ends, credit diffusion that gives every action the same terminal advantage under confounded baselines, and skill admission that is uncalibrated and never retired. To address these challenges, we propose EvoSteer, a new paradigm of Online Self-Evolving Graph Orchestration -- the orchestrator builds a running team and repairs its plausible but failing steps from execution features and a learned value estimate. To support this paradigm, we introduce Anchored Trajectory Balance (AnchorTB), a regression-style flow-matching loss that assigns each orchestration action a coefficient by balancing subtrajectories against a frozen reference. Built on the learned flow, we further propose Validated Skill Admission, in which a candidate skill is tried before promotion and promoted only if paired evidence passes a sequential test under a shared nominal testing budget. Moreover, AnchorTB combines measured task-level reference reward statistics with prefix-dependent corrections. Experimental results on twelve datasets show that EvoSteer significantly outperforms baselines across question answering, mathematical reasoning, code generation, and interactive decision making. Our code is available at https://github.com/beita6969/evosteer.

cs.AI↗

arXiv2026

CollabFlow: Recursive Self-Improvement of Agent Collaboration

Recursive self-improvement (RSI) lets a system improve from its own outcomes; in LLM-based multi-agent systems, Agents refine one another within a task, and outcomes improve how they collaborate across tasks. However, existing multi-agent collaboration leaves this loop open: collaboration is pre-defined at the operator level, topology-only learning keeps verbatim exchange that propagates errors, and reward maximization on a system's own outcomes concentrates on a few teams. To address these challenges, we propose CollabFlow, an RSI system of Learned Agent Collaboration: a trainable Collab-Director constructs teams of complete Agents, a frozen executor runs them, and each round's outcomes retrain the director. Within each round, the edges of a collaboration graph carry protocols of Evidence-Conditioned Communication: a receiver adopts a differing answer only when the sender's evidence is stronger by a margin, so the director learns who communicates and how. Across rounds, we further propose Collaborative Trajectory Balance (CTB), a flow-based objective that credits each team once across its construction orders and targets a reward-proportional distribution over teams, so several good teams stay in play. We also bound how far this self-generated target moves between rounds, which shrinks as records accumulate. On twelve datasets, CollabFlow outperforms all baselines and keeps improving across rounds. Code is available at https://anonymous.4open.science/r/CollabFlow-631E.

cs.MA↗

arXiv2026

JusticeAxis: Benchmarking Legal Judgment between Rigid Rule Application and Ungrounded Discretion

A sound judgment applies the law to established facts and weighs the circumstances in which they arose. However, existing methods swing between rigid statute matching and ungrounded discretion, benchmarks score a label or a rubric, and the experience that would supply the balance stays unverified. We formalize legal judgment as a reference-anchored task, whose object is a single decision that stays tied to the statute and to the circumstances at once. We introduce JusticeAxis, 256 real-world criminal cases from 18 countries with audio, image, and text evidence, and three lawyer-written judgments for every case: the recorded one and one for each failure. We further propose JusticeAgent, a harness whose element agents establish the facts and whose judge agent applies the law under skills carrying experience of the circumstances. Skills are distilled from execution trajectories and admitted only under Bayesian credible bounds. Experiments show that failure turns direction with scale: open-weight backbones drift to unsupported grounds, frontier models to the statutory default. We further verify that JusticeAgent, as a simple yet effective plugin, carries a frozen open-weight backbone to commercial level. Project resources are available at https://github.com/beita6969/JusticeAxis.

cs.AI↗

arXiv2026

What Visual Generators Need from Teachers: Rethinking Representation Alignment

Representation alignment speeds up diffusion transformer training by pulling an intermediate block of the model (student) toward features of a frozen pretrained encoder (teacher). Which teacher layer to align, and for how long, is still set by convention, and each alternative costs a training run. We find that alignment helps where the student cannot linearly recover the teacher's features, not where it already resembles them. Since a deep teacher layer is largely predictable from the one below, we isolate what each layer adds, its increment, and measure how much of it an unaligned student recovers. The student fills the teacher's hierarchy from the bottom up and stalls near the top, which we call hierarchy filling: even after 400K steps it recovers almost none of the deepest. The recoverability gap is the unrecovered share of an increment, read from one unaligned checkpoint. In short runs that each align one teacher layer at one block, the gap nearly reproduces their ranking by FID improvement, and CKA, a measure of feature similarity, largely reverses it. Representation Alignment and Recoverability Estimation (RARE) picks the teacher layer with the largest gap before training. During training, it tracks each token's remaining distance to that layer, the online counterpart of the gap, weights tokens by it, and phases out the loss once the average distance stops falling. With SiT-B/2 on ImageNet $256\times256$, RARE reaches an FID of 18.02 without guidance and 4.46 with it, ahead of seven alignment baselines including REPA, iREPA and HASTE. It also trains in 14% fewer GPU-hours than iREPA. Its FID stays below iREPA's across model scales, teachers, datasets and backbones.

cs.CV↗

arXiv2026

Programmable and scalable on-chip WDM processing platform enabled by cascaded FSR-free resonators

Wavelength-division multiplexing (WDM) is pivotal for expanding the capacity and functionality of photonic systems, from communications to computing. However, on-chip implementation for broadband operation is fundamentally limited by the narrow free spectral range (FSR) and inefficient drop transmission of conventional optical microcavities. Here, we transcend this limit with a scalable integrated WDM processing platform based on cascaded dual-sided coupled Fabry-Pérot add-drop resonators. This unit achieves a record-large FSR-free bandwidth (> 300 nm) and low insertion loss (< 0.5 dB), while enabling independent manipulation of monochromatic light in both the wavelength and intensity domains. Leveraging these performances, we architect programmable cores with up to 20 channels in a single bus waveguide to demonstrate application versatility by configuring as: a high-capacity on-chip WDM communication fabric (2.4-Tbps single-link WDM transmission and O-to-C-band multi-subband (de)triplexing), a reconfigurable optical processor (broadband and hitless wavelength-selective switching), and a parallel computing accelerator (theoretically capable of 1.28-TOPS convolution operations, 16-channel modulation). This work establishes a unified and versatile WDM processing platform that bridges high-speed optical communication with in-line computing, charting a scalable and backward-compatible path toward petabit-per-second-class links and hundreds of TOPS of on-chip computation, resolving a critical bottleneck for next-generation photonic systems.

physics.optics↗

arXiv2026

InfluenceField: A Differentiable Field with Interventionally Identifiable Causal Structure for Multimodal World Modeling

Multimodal large language models often capture visual-linguistic correlations but struggle to predict how local visual interventions propagate and affect downstream answers. We introduce InfluenceField, an intervention-aware latent field inserted between the visual encoder and language decoder. It lifts patch features into a continuous spatial representation, propagates directed influence over multiple steps, and predicts local intervention effects through a shared transition operator. Training jointly optimizes language modeling, cross-environment invariance, counterfactual rollout supervision, and structural regularization. For a nonlinear finite-basis population model, we show that target-aligned interventional supervision, together with a one-step separation condition on the transition, restricts admissible representations to within-location reparameterizations, so that the directed dependency graph of the full transition is recovered exactly. A linear specialization gives an exact partial-coverage characterization and a finite-loss stability bound, and the field analysis derives the spatial profile of coefficient interventions together with a shared-channel calibration result. On CausalVQA, InfluenceField improves overall accuracy over its backbone by 13.1 percentage points, with the largest gains on the planning and hypothetical categories. Capacity-matched baselines and structural controls attribute the gains in robustness and factual-counterfactual consistency to the causal objectives rather than to added capacity.

cs.LG↗

arXiv2026

From fragmented data to actionable design: Physics-calibrated learning for plastic upcycling

Thermochemical upgrading of plastic waste is a key upcycling pathway, yet the experimental literature is fragmented by heterogeneous conditions and incomplete reporting. Complete-case learning would retain only 10.99% of the curated experiments, while target imputation can introduce biased supervision. Here we develop a Physics-Calibrated, Missingness-Gated, and Load-Balanced Mixture-of-Experts (PC-MG-MoE) framework that converts structured missingness into an informative learning signal. PC-MG-MoE learns directly from partially observed experiments without target imputation, reconstructs physically consistent product distributions, accommodates cross-laboratory heterogeneity, and provides interpretable model behaviour rather than black-box prediction alone. Under stringent source-grouped validation, it achieved the lowest aggregate absolute error among the evaluated models, supporting engineering screening under cross-laboratory heterogeneity. Wet-lab experiments provide an external comparison, showing key composition-dependent trends. Implemented as an interactive web-based workflow, PC-MG-MoE enables forward screening, physics-grounded constrained inverse design, targeted experimental planning that supports reduced experimental workload and trial-and-error, and laboratory-specific adaptation with new platform-specific data. This work establishes a transferable framework for converting fragmented literature data into experimentally actionable guidance for model-guided plastic upcycling and broader thermochemical systems.

cs.LG↗

arXiv2026

Sustainability and Artificial Intelligence: Necessary, Challenging, and Promising Intersections

Both digital economy and digital technology researchers increasingly recognize the need to better address the role that artificial intelligence (AI) plays in shaping the evolution of the environmental, social and governance aspects of development. It appears that sustainability and AI research converge on the features of wicked problems that are complex, interconnected and dynamic. Building off such convergence, this article aims to map out the necessary, challenging, and promising intersections by providing an overview of the state of art research. Based on 541 bibliographic data collected from the Web of Science (WoS) database, the findings reveal the increasingly central body of work on green and sustainable science and technology in bridging various disciplines, main journals and key topics and concepts. The findings reveal how such interactions can be necessary, challenging, and promising. The article concludes with few general arguments regarding how to diversify and expand the community of practice regarding AI for sustainable development, especially in the areas of expected AI application areas and institutions.

cs.SI↗

arXiv2026

Enabling Extensible Embodied Capabilities with Tools

Most existing embodied intelligence methods formulate perception, reasoning, planning, and control within a unified parameterized policy. Yet these capabilities are inherently hierarchical and heterogeneous, making them difficult to reliably learn and modularize within a single model. We propose a capability externalization approach that decouples heterogeneous capabilities into independently optimized tools, dynamically invoked at inference time. To this end, we introduce Embodied Tool Protocol (ETP), a standardized protocol for embodied tool registration, discovery, invocation, and execution, and curate 100+ validated tools spanning perception, cognition, reasoning, and execution as the tool base. Building on this, we construct EmbodiedToolBench to evaluate both whether tool augmentation improves embodied performance and how well current models use tools across tool-necessity recognition, tool selection, tool execution, and tool-chain composition. Experiments across simulation and real-world platforms confirm that capability externalization consistently improves embodied performance (avg. gain 31% on EB-ALFRED and 36% on EB-Navigation), yet reveal a clear boundary: gains are substantial for cognition and perception but are limited for execution-type capabilities. Moreover, our analysis reveals that knowing when, which, and how to invoke tools remains a persistent challenge across all models, thereby highlighting embodied tool competence as a critical direction for future research.

cs.RO↗

arXiv2025

WritingBench: A Comprehensive Benchmark for Generative Writing

Recent advancements in large language models (LLMs) have significantly enhanced text generation capabilities, yet evaluating their performance in generative writing remains a challenge. Existing benchmarks primarily focus on generic text generation or limited in writing tasks, failing to capture the diverse requirements of high-quality written contents across various domains. To bridge this gap, we present WritingBench, a comprehensive benchmark designed to evaluate LLMs across 6 core writing domains and 100 subdomains. We further propose a query-dependent evaluation framework that empowers LLMs to dynamically generate instance-specific assessment criteria. This framework is complemented by a fine-tuned critic model for criteria-aware scoring, enabling evaluations in style, format and length. The framework's validity is further demonstrated by its data curation capability, which enables a 7B-parameter model to outperform the performance of GPT-4o in writing. We open-source the benchmark, along with evaluation tools and modular framework components, to advance the development of LLMs in writing.

cs.AI↗

arXiv2024

A new paradigm of mortality modeling via individual vitality dynamics

The significance of mortality modeling extends across multiple research areas, ranging from life insurance valuation to optimal lifetime decision-making. Existing approaches, such as mortality laws and factor-based models, often fall short in capturing the complexity of individual mortality, hindering their ability to address specific research needs. To overcome these limitations, this paper introduces a novel approach to mortality modeling centered on the dynamics of individual vitality. A four-component framework is developed to account for initial conditions, natural aging processes, stochastic fluctuations, and accidental events over an individual's lifetime. We demonstrate the framework's analytical capabilities across various settings and explore its practical implications in solving life insurance problems and deriving optimal lifetime decisions. Our results show that the proposed framework not only encompasses existing mortality models but also provides individualized mortality outcomes and offers an intuitive explanation for survival biases.

stat.AP↗

arXiv2024

On-Chip Vectorial Structured Light Manipulation via Inverse Design

On-chip structured light, with potentially infinite complexity, has emerged as a linchpin in the realm of integrated photonics. However, the realization of arbitrarily tailoring a multitude of light field dimensions in complex media remains a challenge1, Through associating physical light fields and mathematical function spaces by introducing a mapping operator, we proposed a data-driven inverse design method to precisely manipulate between any two structured light fields in the on-chip high-dimensional Hilbert space. To illustrate, light field conversion in on-chip topological photonics was achieved. High-performance topological coupling devices with minimal insertion loss and customizable topological routing devices were designed and realized. Our method provides a new paradigm to enable precise manipulation over the on-chip vectorial structured light and paves the way for the realization of complex photonic functions.

physics.optics↗

arXiv2024

StyleMamba : State Space Model for Efficient Text-driven Image Style Transfer

We present StyleMamba, an efficient image style transfer framework that translates text prompts into corresponding visual styles while preserving the content integrity of the original images. Existing text-guided stylization requires hundreds of training iterations and takes a lot of computing resources. To speed up the process, we propose a conditional State Space Model for Efficient Text-driven Image Style Transfer, dubbed StyleMamba, that sequentially aligns the image features to the target text prompts. To enhance the local and global style consistency between text and image, we propose masked and second-order directional losses to optimize the stylization direction to significantly reduce the training iterations by 5 times and the inference time by 3 times. Extensive experiments and qualitative evaluation confirm the robust and superior stylization performance of our methods compared to the existing baselines.

cs.CV↗

arXiv2024

Uncertainty-aware self-training with expectation maximization basis transformation

Self-training is a powerful approach to deep learning. The key process is to find a pseudo-label for modeling. However, previous self-training algorithms suffer from the over-confidence issue brought by the hard labels, even some confidence-related regularizers cannot comprehensively catch the uncertainty. Therefore, we propose a new self-training framework to combine uncertainty information of both model and dataset. Specifically, we propose to use Expectation-Maximization (EM) to smooth the labels and comprehensively estimate the uncertainty information. We further design a basis extraction network to estimate the initial basis from the dataset. The obtained basis with uncertainty can be filtered based on uncertainty information. It can then be transformed into the real hard label to iteratively update the model and basis in the retraining process. Experiments on image classification and semantic segmentation show the advantages of our methods among confidence-aware self-training algorithms with 1-3 percentage improvement on different datasets.

cs.CV↗

arXiv2024

Potential Energy based Mixture Model for Noisy Label Learning

Training deep neural networks (DNNs) from noisy labels is an important and challenging task. However, most existing approaches focus on the corrupted labels and ignore the importance of inherent data structure. To bridge the gap between noisy labels and data, inspired by the concept of potential energy in physics, we propose a novel Potential Energy based Mixture Model (PEMM) for noise-labels learning. We innovate a distance-based classifier with the potential energy regularization on its class centers. Embedding our proposed classifier with existing deep learning backbones, we can have robust networks with better feature representations. They can preserve intrinsic structures from the data, resulting in a superior noisy tolerance. We conducted extensive experiments to analyze the efficiency of our proposed model on several real-world datasets. Quantitative results show that it can achieve state-of-the-art performance.

cs.LG↗

arXiv2023

Arbitrary point cloud upsampling via Dual Back-Projection Network

Point clouds acquired from 3D sensors are usually sparse and noisy. Point cloud upsampling is an approach to increase the density of the point cloud so that detailed geometric information can be restored. In this paper, we propose a Dual Back-Projection network for point cloud upsampling (DBPnet). A Dual Back-Projection is formulated in an up-down-up manner for point cloud upsampling. It not only back projects feature residues but also coordinates residues so that the network better captures the point correlations in the feature and space domains, achieving lower reconstruction errors on both uniform and non-uniform sparse point clouds. Our proposed method is also generalizable for arbitrary upsampling tasks (e.g. 4x, 5.5x). Experimental results show that the proposed method achieves the lowest point set matching losses with respect to the benchmark. In addition, the success of our approach demonstrates that generative networks are not necessarily needed for non-uniform point clouds.

cs.CV↗