Searcharxiv⌕ Search

arXiv subjects

Yu Zhang

Publications and source records attributed to Yu Zhang.

At least 109 records · Page 6Linked to original sources
arXiv2026

CogEvol: Towards Efficient and Reliable Learning Environment Generation

We present CogEvol, a family of models trained specifically for Learning Environment Generation: turning a course brief into a finished learning artifact (structured-JSON slides or self-contained interactive HTML pages) in a single pass. Across 220k production requests, CogEvol completes a slide in a median of 17 seconds and an interactive page in 59, replacing minutes-long multi-turn agent scaffolding. Reliability is enforced rather than hoped for: a production-grounded data pipeline turns real failures into 53,687 verified SFT samples, and a hybrid rule-plus-VLM reward drives GRPO-based RL, hardened after we caught and fixed a reward-hacking episode that produced visually convincing but unplayable games. CogEvol-27B scores 83.7 on slide quality and 63.7 on a 500-case interactive-HTML benchmark with 26.9x fewer parameters than flagship coding models, and, in collaboration with the OpenMAIC team, serves their live production traffic. CogEvol-4B is released openly under the Apache 2.0 license at https://github.com/CogEvol/CogEvol-4B; external flagships are measured on the same suites under the identical harness. Scaffold editing cuts interactive-page generation cost by a further ~76%, and the full stack runs on domestic Ascend accelerators at application-level parity with A800 GPUs, lowering the unit cost of AI-native education at scale.

cs.CL↗
arXiv2026

Probing the dark axion portal via $J/ψ$ decays at BESIII and STCF

Large numbers of $J/ψ$ mesons can be resonantly produced at BESIII and STCF at the center-of-mass energy $\sqrt{s}=3.097$~GeV. Such $J/ψ$ mesons may undergo rare decays into an axionlike particle (ALP) $a$ and a dark photon $γ'$ through the dark axion portal. We investigate the exclusion reach of the existing BESIII dataset and the projected sensitivity of STCF, focusing on the mono-photon signature. We perform Monte Carlo simulations of the signal and estimate both the irreducible $J/ψ\toγν\barν$ background and the continuum and resonant three-photon reducible backgrounds. Detector-induced migration of three-photon events into the mono-photon signal region is parameterized by the event-level leakage probability $ε_{3γ}^{\rm leak}=0,10^{-6},10^{-4},10^{-3}$. In the zero- and sufficiently small-leakage benchmarks, the existing BESIII dataset could probe previously unexplored parameter space, while STCF can further improve the coupling reach. Increasing leakage probabilities progressively weaken these projections, highlighting the importance of detector-level three-photon rejection in realizing the projected reach.

hep-ph↗
arXiv2026

Figures as Programs: Recursive Generation of Editable Scientific Figures

Scientific methodology figures are essential for communicating complex methods clearly, yet creating them remains labor-intensive and typically requires multiple rounds of refinement. Recent image-generation models can synthesize visually appealing raster figures, but producing a human-satisfactory result in a single generation step remains difficult. Moreover, precise edits to raster figures are challenging for both humans and models. We formulate scientific figure generation as recursive SVG program construction and propose \textsc{FigTree}, a \textit{multi-agent} system that automatically transforms a scientific paper into a structured vector figure. \textsc{FigTree} grounds figure content in the source paper, decomposes a figure into a hierarchy of local regions, generates each region as a short SVG program, and assembles the resulting fragments. A render-critic refinement loop jointly inspects the rendered figure and its underlying program, enabling visual defects to be traced to specific statements and accurately repaired. We conduct extensive evaluations of \textsc{FigTree} on figure quality and editability, showing that \textsc{FigTree} produces high-quality figures, while also enabling more effective editing than existing raster-based methods.

cs.AI↗
arXiv2026

HybridWorldSim: A Scalable and Controllable High-fidelity Simulator for Autonomous Driving

Realistic and controllable simulation is critical for advancing end-to-end autonomous driving, yet existing approaches often struggle to support novel view synthesis under large viewpoint changes or to ensure geometric consistency. We introduce HybridWorldSim, a hybrid simulation framework that integrates multi-traversal neural reconstruction for static backgrounds with generative modeling for dynamic agents. This unified design addresses key limitations of previous methods, enabling the creation of diverse and high-fidelity driving scenarios with reliable visual and spatial consistency. To facilitate robust benchmarking, we further release a new multi-traversal dataset MIRROR that captures a wide range of routes and environmental conditions across different cities. Extensive experiments demonstrate that HybridWorldSim surpasses previous state-of-the-art methods, providing a practical and scalable solution for high-fidelity simulation and a valuable resource for research and development in autonomous driving.

cs.CV↗
arXiv2026

SetMIR: Multi-Interest Retrieval as Set Prediction

Embedding-based retrieval is at the core of industrial recommender systems, but a single user embedding is often too limited to capture a user's diverse interests. Multi-interest retrieval addresses this by using multiple user embeddings, yet existing methods still suffer from two issues: interest collapse, where different embeddings learn the same interest, and static dispatch, where serving uses a fixed retrieval budget even when some embeddings are unnecessary. We propose SetMIR, which treats multi-interest retrieval as a set prediction problem. SetMIR encodes a user's behavior history with a transformer and uses K learnable queries to decode a set of user interests, each producing a retrieval embedding and a presence score. During training, Hungarian matching assigns targets to queries one-to-one, so matched queries learn distinct interests and the presence head learns which queries are active. At serving time, SetMIR uses presence scores and query-level Non-Maximum Suppression (NMS) to issue only active, non-redundant ANN queries. On Snap's Dynamic Product Ads (DPA) data, SetMIR outperforms four learned multi-interest retrievers on every metric while issuing 33% fewer ANN queries per request. Deployed as a new retrieval source in the DPA production stack, SetMIR lifts overall CVR by 3.1%, while lifting CTR by 44% and CVR by 51% over the item-to-item retrieval source with the same item embeddings, ANN index, and retrieval quota.

cs.IR↗
arXiv2026

CAMIE: Co-Engagement-Aware Multimodal Item Embeddings for Snap Dynamic Product Ads Retrieval

Item-to-item (I2I) retrieval is a core primitive in large-scale recommendation and advertising systems. In production Snap Dynamic Product Ads (DPA), I2I retrieval faces two challenges: separate visual, textual, and multimodal encoders fragment the retrieval stack, and content-only training does not align embeddings with the co-engagement behavior that drives downstream conversions. We present CAMIE, a co-engagement-aware multimodal item embedding framework for Snap DPA retrieval. CAMIE builds on LLM/MLLM backbones, using their native multimodal interfaces to represent item images and metadata in a shared embedding space. It then fine-tunes the backbone on co-engaged item pairs mined from user journeys with a symmetric in-batch InfoNCE objective. Offline, CAMIE outperforms the strongest commercial multimodal embedding model on Recall@10 and serves text-only retrieval from the same checkpoint with minimal quality loss. Online, CAMIE serves as a drop-in replacement for two deployed content-based I2I encoders, delivering +0.390% CTR / +10.832% CVR over the multimodal control, +18.958% CTR / +13.12% CVR over the text control, and +0.211% CTR / +1.911% CVR on overall DPA traffic. CAMIE is deployed in production.

cs.IR↗
arXiv2026

Two-loop QCD corrections to $CP$-even Higgs boson decays into $Vγ$ in the Type-I Two-Higgs-Doublet Model

We investigate the QCD corrections to the loop-induced decays of the heavy $CP$-even Higgs boson, $H \to Vγ$ ($V = Z, γ$), within the framework of the Type-I Two-Higgs-Doublet Model (THDM). Owing to the similarity of the underlying loop topologies, the two decay channels exhibit closely correlated behaviour. After imposing relevant phenomenological constraints, we evaluate the decay widths for 5000 viable parameter points at next-to-leading order (NLO) in QCD under the alignment limit. The $H\toγγ$ width exceeds that of $H\to Zγ$ by roughly a factor of four. Both are dominated by the top-quark loop: the $W$ contribution vanishes in the alignment limit, and the charged-Higgs loop is suppressed throughout most of the viable parameter region. The corrections exhibit a pronounced dependence on the heavy Higgs boson mass $m_H$, attaining a maximum in the vicinity of the $t\bar{t}$ production threshold. To elucidate the individual parameter dependences, dedicated benchmark scenarios are analysed, revealing that the decay widths diminish monotonically with increasing $\tanβ$ while growing with $m_H$. The QCD corrections for both decay modes typically span approximately $-20\%$ to $+20\%$, with larger corrections in the non-alignment scenario: in the $H\toγγ$ channel they reach up to $+38\%$ at large $\tanβ$, and in the $H\to Zγ$ channel they cross zero to become positive. The $CP$ selection rule forces the $t\bar{t}$ pair into a $P$-wave configuration at threshold, which strongly suppresses the Coulomb corrections and secures the perturbative stability of the fixed-order prediction. A comparison with ATLAS limits shows that current data do not yet constrain the parameter space, whereas the $\mathcal{O}(15\%)$ NLO corrections may become relevant for interpreting exclusion bounds at the HL-LHC.

hep-ph↗
arXiv2026

MemeBridge: A Dataset for Benchmarking and Mitigating the Bidirectional Cultural Gap in Meme Interpretation

Communicating across cultures is inherently challenging, especially through culturally dense and ambiguous formats like memes. While people expect large language models (LLMs) to hold promise for bridging such gaps, existing benchmark datasets often fail to capture the cultural context necessary for accurate interpretation. To address this, we introduce MemeBridge, a curated dataset centered on U.S.-originated memes, designed to capture two complementary perspectives: (1) how Chinese participants interpret these memes, and (2) how U.S. participants anticipate how people from other cultures might misunderstand them. Here, context refers to implicit cultural knowledge, including background beliefs, norms, and shared assumptions that shape meme comprehension. The dataset was constructed via a multi-stage crowdsourcing pipeline with rigorous validation, including human agreement checks and GPT-based classification verification. Each meme is annotated with sentiment, emotion, cultural significance, and knowledge type, providing rich supervision for downstream tasks. Notably, we observe that the anticipated misunderstandings from U.S. participants are often inaccurate, highlighting the asymmetries in cultural understanding and the challenges of adopting perspectives beyond one's own. This bidirectional framing, which focuses on both expression and perception, enables more nuanced benchmarking of cross-cultural comprehension. Our probing of multiple LLMs reveals that while models developed in different cultural contexts exhibit partial cross-cultural understanding, they often struggle with sophisticated interpretations. By contrast, fine-tuning with MemeBridge improves model performance, underscoring the value of culturally grounded resources for training and evaluating LLMs in globally diverse settings.

cs.CL↗
arXiv2026

Low-energy Muon-Nucleon scattering experiment: LUNE (White Paper)

The HIAF will provide high-intensity, high-quality muon beams with momenta from 0.5 to 7.5 GeV/c. This energy range is uniquely suited for precision muon scattering, bridging the gap between low-energy electron facilities and future high-energy lepton-ion colliders. In particular, HIAF will enable precision measurements with both positive and negative muon beams over a broad kinematic range, complementing existing electron-scattering facilities such as JLab, EicC and EIC. Based on HIAF muon source, the LUNE Collaboration has been established to address several fundamental questions in nuclear and particle physics, including the proton charge radius puzzle, nucleon electromagnetic structure, and the dynamics of quantum electrodynamics and hadronic interactions. The program proceeds in two phases, from elastic scattering to nucleon structure and beyond-Standard-Model searches. The experiment is expected to determine the proton charge radius with a precision of approximately 1.0\% using elastic muon-proton scattering. It will also perform systematic measurements of the proton electromagnetic form factors with both $μ^+$ and $μ^-$ beams, enabling precise studies of two-photon exchange effects and stringent tests of quantum electrodynamics. Beyond elastic scattering, LUNE will investigate TMD, gravitational form factors, and nuclear charge radii, providing new insights into the 3D structure of nucleons and nuclei. The experiment will further address important topics including Coulomb-distortion corrections, nuclear medium effects, and possible signatures of physics beyond the Standard Model. This white paper presents the scientific motivation, detector concept, expected performance, and long-term strategy of LUNE.

hep-ex↗
arXiv2026

PCBnet: A Dataset and Automatic Construction of SPICE Netlists from Schematic Images

Printed circuit boards (PCBs) are fundamental to modern electronic systems, yet AI-driven PCB design automation remains constrained by the lack of large-scale paired schematic-netlist datasets. PCB schematics are particularly challenging due to diverse component types, complex wiring topologies, and noisy textual annotations. To address this gap, we present PCBnet, a large-scale PCB schematic dataset comprising over 300 real-world designs with annotated pins and paired SPICE netlists. It contains more than 50,000 component instances, 150,000 wires, 100,000 text regions, and 400,000 characters. We further develop an automated schematic-to-netlist pipeline that combines visual recognition, topology construction, and domain-knowledge-guided multi-agent correction. The proposed method achieves 94.54% component detection mAP, 98.57% text recognition accuracy, and 84.47% end-to-end connectivity accuracy. PCBnet provides a benchmark and data foundation for future AI-driven PCB design automation.

cs.CV↗
arXiv2026

Astar: Learning to Propose Evolution Directions for Self-Evolving Industrial AI Systems

Modern AI systems advance through continuous iteration: a loop of proposing evolution directions, implementing code, training, and evaluation. While the latter three stages are increasingly automated, the starting point --- proposing effective evolution directions --- remains a critical bottleneck that still relies heavily on senior experts. In this work, we explore whether AI can take over this role. We find that general-purpose LLMs, even the advanced GPT-5.5, offer only generic and misaligned suggestions: the required expertise is accumulated through experience rather than explicitly codified, and thus hard to inject directly. To this end, we propose Astar, a training-based approach that learns a specialized evolution-guiding model from the abundant iteration histories of industrial systems. Realizing this idea, however, raises four challenges: sparse supervision, noisy data, a vast direction space, and prohibitively expensive verification. We address them along two fronts. On the data side, we design a pipeline that turns noisy historical commits into a large, clean evolutionary corpus via pairwise sample expansion and noise filtering. On the model side, we train the model through mid-training, SFT, and RL, guiding evolution direction generation with hierarchical hints and using the reward model in RL as a fast surrogate evaluator. Astar has been deployed in Alibaba's Lazada advertising system for evolution direction proposal. Astar-8B achieves a single-proposal success rate of 0.6786 in real-execution evaluation, far exceeding human experts (0.3229) and the strongest general-purpose LLM (0.3071). More importantly, Astar closes the loop and enables fully automatic iteration: it guided 20 consecutive iterations over two weeks, improving offline Hitrate@200 by 23.6%, while an online A/B test yielded relative lifts of 4.86% in GMV and 1.82% in advertising revenue.

cs.IR↗
arXiv2026

CSAVocoder: A Causal Spatial Audio Vocoder Towards Real-Time Spatial Audio Generation

Spatial audio vocoders are able to convert mel-spectrograms produced by generative models into spatial audio waveforms. Most neural vocoders are designed for monaural audio, and direct extensions to spatial audio can degrade spatial quality by ignoring inter-channel cues. We present CSAVocoder, a causal GAN-based spatial audio vocoder that jointly optimizes waveform fidelity and spatial rendering. Our framework introduces a Spatial Adaptor that fuses multi-channel mel-spectrograms with dynamic source-listener pose information, together with a spatial consistency discriminator that supervises inter-channel cues. To meet real-time requirements, we design a strictly causal, stateful generator that supports efficient streaming inference with constant memory overhead. Experiments on large-scale spatial audio datasets show that CSAVocoder improves spatial fidelity at competitive audio quality and real-time performance.

eess.AS↗
arXiv2026

A General-Purpose Framework for Chemical Reaction Representation with Atomic Correspondence and Flexible Condition Adaptation

Motivation: Organic synthesis is fundamental to the chemical industry, particularly in domains such as pharmaceutical development. While artificial intelligence offers powerful tools for modeling chemical reactions, current approaches are primarily limited to two paradigms: those that rely on hand-crafted, domain-specific features, and those that apply generic deep learning models through simplistic concatenation or aggregation of reaction components. The former often struggles to scale effectively with increasing data volumes, while the latter relies on simple input- or feature-level concatenation to combine different reaction components. Such simplistic aggregation prevents these models from directly capturing the structural transformations between reactants and products, and also makes it difficult to adapt or extend them to datasets that include non-molecular reaction conditions without modifying their model architectures. Results: This paper introduces Align-React, a novel chemical reaction representation learning framework designed for a wide range of organic reaction tasks. Our approach integrates atomic correspondence between reactants and products to discern precise molecular transformations, thereby improving the model's comprehension of molecular transformation patterns. We incorporate an adapter structure to embed reaction conditions into the representation, enhancing adaptability across varied datasets and tasks. Furthermore, a Reaction-Center-Aware attention mechanism is proposed to enable the model to focus on critical functional groups, yielding more powerful and informative representations. Evaluated across multiple downstream tasks, our model demonstrates superior performance, significantly outperforming existing chemical reaction representation learning architectures on most benchmark datasets.

cs.LG↗
arXiv2026

Reproducible Evaluation of MoE Expert Caching: Replay Semantics, Workload Contamination, and Operating Regimes

Mixture-of-Experts (MoE) models have outgrown accelerator memory, and offloading expert weights to host memory is now standard. This makes expert cache management an attractive lever: a policy that raised the hit rate would cut expert traffic per token. Evaluating that is a measurement problem, and we find the measurement fragile. With a trace-driven, event-atomic simulator over three MoE models (40, 64, 128 experts), we isolate three evaluation axes that change conclusions, not just numbers. Replay semantics: under a fused-event traffic contract, an inconsistent per-access replay inflates recency-based policies by 27-29% while leaving frequency-based and static ones within 4%, inverting the policy ranking. Workload contamination: probe sets using one instruction template per category produce verbatim-identical generation prefixes; a matched-pair rendering intervention moves the measured early-window effect by 19.4-31.9 points and reverses which workloads look most cache-friendly. Operating regimes: normalized miss fractions do not transfer across models, so the per-step expert union relative to per-layer capacity must be reported -- yet permuting only the temporal order of an identical event stream moves the offline-optimal gap from 44.9% to 30.8%, so it is not sufficient. Corrected, a stable gap to the offline optimum remains (44.2-45.9% over 13 frozen workload compositions). A forced-admission oracle attributes 84.3-96.6% of it to knowing which resident expert is used furthest in the future. A causal next-use predictor, used as an eviction rule, recovers -11.4% of the gap; it picks an optimal victim 3.4% of the time, against 2.4% for a random resident block and 20.6-22.1% for LRU and LFRU. Our position is narrow: in our evaluated settings a large offline-optimal gap substantially overstates the gains recovered by representative lightweight causal mechanisms.

cs.LG↗
arXiv2026

Towards Comprehensive Basketball Understanding

Understanding a basketball game requires recognizing events, localizing actions, identifying players, and relating these to structured game knowledge. Existing benchmarks primarily evaluate these abilities one at a time, leaving the interactions among these abilities under-explored. We introduce BasketballBench, a multimodal benchmark comprising 7,980 questions across ten tasks in text, image, and video. It is built from the 2025-2026 NBA season and includes official playby-play, rosters and profiles for 530 active players, and 2,501 possession-level broadcast clips. We further propose BasketballSkills, an agent that composes eight basketball-specific perception and retrieval tools under four reusable skills that specify tool order, evidence bindings, and stopping conditions. Experiments show that current MLLMs struggle particularly on questions requiring the integration of multiple capabilities, whereas BasketballSkills outperforms them, highlighting the effectiveness of explicitly composing domain-specific capabilities for comprehensive basketball understanding.

cs.CV↗
arXiv2026

MARL-Based Sequential RIS Auctions: A Physical-Layer Security Analysis

Reconfigurable intelligent surfaces (RISs) hold great potential to enhance coverage, spectral efficiency, and communication security by intelligently configuring their reflecting elements. When owned by a neutral RIS operator, these elements can be offered as resources for which legitimate receivers and eavesdroppers compete. This paper investigates such competition and evaluates its impact on the physical-layer security performance of legitimate receivers. To model the competition, we develop a sequential RIS auction (SRA) framework, in which a bundle of RIS elements is auctioned in each round through a first-price sealed-bid mechanism, with each bidder submitting its bid based on the achievable rate gain and remaining budget. We then formulate the sequential bidding process as a Markov game by specifying its states, actions, rewards, and state transitions. To solve the game, we propose a multi-bidder deep deterministic policy gradient (MADDPG)-based multi-bidder reinforcement learning (MARL) approach under centralized training and decentralized execution (CTDE), enabling legitimate receivers and eavesdroppers to learn bidding strategies that maximize their long-term economic surplus. Numerical results show that, under the considered eavesdropper bidding strategies, the RL-based strategy enables legitimate receivers to achieve the highest secrecy rate per unit cost, outperforming random and fixed strategies and approaching the ideal physical-layer upper bound.

cs.CR↗
arXiv2026

The Imitator Game: Benchmarking Robot Imitative Ability Beyond Action Prediction

Humans imitate at the level of intent: given a demonstration, we infer its goal and carry it out with whatever tools, objects, and layouts are at hand. Current robot policies instead learn observation-to-action mappings from visual inputs and language instructions, without explicitly inferring the demonstrated task. Learning from human video thus remains largely trajectory-level: models can replay motions in near-identical scenes, but still struggle to imitate what the demonstrator intends rather than merely what they do. We introduce The Imitator Game, a four-level benchmark (L0-L3) that progressively widens the gap between the human demonstration and the robot's own scene, isolating where trajectory replay ceases to suffice and task understanding becomes necessary. We pair it with IG-10K, the largest environment-aligned paired human-robot dataset to date and the only one instantiated across all four levels in both real and simulated settings (20,000+ paired episodes, 50+ tasks, 6 domains), and Imitator Arena, an open platform for blind A/B human evaluation. Across nine state-of-the-art models, performance is stable from L0 to L2 but collapses at L3, identifying functional substitution - achieving the same intent through a different object affordance - as the decisive barrier to intent-level imitation. Human-video-conditioned models outperform caption-conditioned ones, yet every model falls below 13% zero-shot success on unseen tasks; fine-tuning IG-10K-pretrained models with only $10$ paired human-robot demonstrations yields large gains that grow with pretraining scale. The project website and access to Imitator Arena are available at https://imitator-game.github.io.

cs.RO↗
arXiv2026

Learning Spatially Adaptive Structural Coordination for Underwater Salient Object Detection

Underwater salient object detection (USOD) has attracted increasing attention for underwater scene understanding and vision-guided robotic applications. However, the spatially non-uniform degradation in underwater images causes spatially varying reliability of structural cues: boundary-sensitive responses can enhance object contours but are vulnerable to degradation-induced noise, whereas region-coherent responses improve semantic completeness but may blur object boundaries. Existing methods rarely explicitly consider the spatial variation in structural cue reliability under underwater image degradation. To address this problem, this work proposes SASC-USOD, a novel framework for learning spatially adaptive structural coordination in USOD. The proposed framework constructs two complementary structural representations with different characteristics. A boundary-sensitive representation is obtained by combining fixed Laplacian filtering with a learnable local-detail transformation to enhance discriminative boundary information, while a region-coherent representation is generated through dual-range anisotropic large-kernel contextual aggregation to capture long-range structural consistency. A spatial coordination module is then introduced to estimate the relative reliability of these structural representations and adaptively coordinate their contributions according to image content. Extensive experiments on the USOD10K and USOD benchmarks demonstrate that SASC-USOD consistently outperforms existing methods, reducing MAE by 4.07\% and 23.53\% compared with the strongest competing method, respectively. Moreover, its lightweight variant runs at 21 FPS on an NVIDIA Jetson TX2 NX, demonstrating its capability for onboard underwater robotic perception.

cs.CV↗