arXiv subjects

Chenyang Zhao

Publications and source records attributed to Chenyang Zhao.

At least 37 records · Page 2Linked to original sources

arXiv2026

ATTS: Asynchronous Test-Time Scaling via Conformal Prediction

Large language models (LLMs) benefit from test-time scaling but are often hampered by high inference latency. Speculative decoding is a natural way to accelerate the scaling process; however, scaling along both the parallel and sequential dimensions poses significant challenges, including substantial memory-bound execution and synchronization overhead. We introduce ATTS (Asynchronous Test-Time Scaling), a statistically guaranteed adaptive scaling framework that follows the hypothesis testing process to address these challenges. By revisiting arithmetic intensity, ATTS identifies synchronization as the primary bottleneck. It enables asynchronous inference through online calibration and proposes an ordinal classification algorithm that supports a three-stage rejection sampling pipeline, scaling along both the sequential and parallel axes. Across experiments on the MATH, AMC23, AIME24, and AIME25 datasets and across multiple draft-target model families, we show that ATTS delivers up to 56.7x speedup in test-time scaling and a 4.14x throughput improvement, while maintaining accurate control of the rejection rate, reducing latency and memory overhead, and incurring no accuracy loss. By scaling both in parallel and sequential dimensions, we enable the 1.5B/70B draft/target model combination to achieve the performance of the state-of-the-art reasoning model o3-mini (high) on the AIME dataset. We have released the code at https://github.com/menik1126/asynchronous-test-time-scaling.

cs.CL↗

arXiv2026

Balancing Multiple Objectives in Urban Traffic Control with Reinforcement Learning from AI Feedback

Reward design has been one of the central challenges for real world reinforcement learning (RL) deployment, especially in settings with multiple objectives. Preference-based RL offers an appealing alternative by learning from human preferences over pairs of behavioural outcomes. More recently, RL from AI feedback (RLAIF) has demonstrated that large language models (LLMs) can generate preference labels at scale, mitigating the reliance on human annotators. However, existing RLAIF work typically focuses only on single-objective tasks, leaving the open question of how RLAIF handles systems that involve multiple objectives. In such systems trade-offs among conflicting objectives are difficult to specify, and policies risk collapsing into optimizing for a dominant goal. In this paper, we explore the extension of the RLAIF paradigm to multi-objective self-adaptive systems. We show that multi-objective RLAIF can produce policies that yield balanced trade-offs reflecting different user priorities without laborious reward engineering. We argue that integrating RLAIF into multi-objective RL offers a scalable path toward user-aligned policy learning in domains with inherently conflicting objectives.

cs.AI↗

arXiv2026

CM2: Reinforcement Learning with Checklist Rewards for Multi-Turn and Multi-Step Agentic Tool Use

AI agents are increasingly used to solve real-world tasks by reasoning over multi-turn user interactions and invoking external tools. However, applying reinforcement learning to such settings remains difficult: realistic objectives often lack verifiable rewards and instead emphasize open-ended behaviors; moreover, RL for multi-turn, multi-step agentic tool use is still underexplored; and building and maintaining executable tool environments is costly, limiting scale and coverage. We propose CM2, an RL framework that replaces verifiable outcome rewards with checklist rewards. CM2 decomposes each turn's intended behavior into fine-grained binary criteria with explicit evidence grounding and structured metadata, turning open-ended judging into more stable classification-style decisions. To balance stability and informativeness, our method adopts a strategy of sparse reward assignment but dense evaluation criteria. Training is performed in a scalable LLM-simulated tool environment, avoiding heavy engineering for large tool sets. Experiments show that CM2 consistently improves over supervised fine-tuning. Starting from an 8B Base model and training on an 8k-example RL dataset, CM2 improves over the SFT counterpart by 8 points on tau^-Bench, by 10 points on BFCL-V4, and by 12 points on ToolSandbox. The results match or even outperform similarly sized open-source baselines, including the judging model. CM2 thus provides a scalable recipe for optimizing multi-turn, multi-step tool-using agents without relying on verifiable rewards. Code provided by the open-source community: https://github.com/namezhenzhang/CM2-RLCR-Tool-Agent.

cs.AI↗

arXiv2026

On-Chip Erbium-Doped Tantalum Oxide Microring Hybrid Cavity Single-Mode Laser

We demonstrate a high-performance, single-mode Er:Ta2O5 microring laser monolithically integrated on a silicon platform via a customized Damascene process. The Er:Ta2O5 gain medium exhibits a low propagation loss of 0.73 dB/cm and a high intrinsic Q-factor of 5.03 x 105. By utilizing a hybrid cavity_consisting of a microring coupled to a U-shaped waveguide at two symmetric points_we exploit the Vernier effect to achieve robust longitudinal mode selection. Under a non-resonant 1480 nm pumping scheme, the laser yields a side_mode suppression ratio (SMSR) of 53.3 dB and a narrow linewidth of 9.5 pm. A slope efficiency of 2.76 % is achieved_the highest reported to date for Er:Ta2O5 lasers_with a lasing threshold of 3.3 mW. Furthermore, stable single-mode tuning is demonstrated across a temperature range of 18_68 celsius, consistently aligning with theoretical transfer matrix models. This work provides a scalable pathway for high-efficiency, tunable on-chip light sources, bridging the gap for monolithic active-passive integration on the tantalum oxide photonic platform.

physics.optics↗

arXiv2026

MMFormalizer: Multimodal Autoformalization in the Wild

Autoformalization, which translates natural language mathematics into formal statements to enable machine reasoning, faces fundamental challenges in the wild due to the multimodal nature of the physical world, where physics requires inferring hidden constraints (e.g., mass or energy) from visual elements. To address this, we propose MMFormalizer, which extends autoformalization beyond text by integrating adaptive grounding with entities from real-world mathematical and physical domains. MMFormalizer recursively constructs formal propositions from perceptually grounded primitives through recursive grounding and axiom composition, with adaptive recursive termination ensuring that every abstraction is supported by visual evidence and anchored in dimensional or axiomatic grounding. We evaluate MMFormalizer on a new benchmark, PhyX-AF, comprising 115 curated samples from MathVerse, PhyX, Synthetic Geometry, and Analytic Geometry, covering diverse multimodal autoformalization tasks. Results show that frontier models such as GPT-5 and Gemini-3-Pro achieve the highest compile and semantic accuracy, with GPT-5 excelling in physical reasoning, while geometry remains the most challenging domain. Overall, MMFormalizer provides a scalable framework for unified multimodal autoformalization, bridging perception and formal reasoning. To the best of our knowledge, this is the first multimodal autoformalization method capable of handling classical mechanics (derived from the Hamiltonian), as well as relativity, quantum mechanics, and thermodynamics. More details are available on our project page: MMFormalizer.github.io

cs.CL↗

arXiv2026

ScienceDB AI: An LLM-Driven Agentic Recommender System for Large-Scale Scientific Data Sharing Services

The rapid growth of AI for Science (AI4S) has underscored the significance of scientific datasets, leading to the establishment of numerous national scientific data centers and sharing platforms. Despite this progress, efficiently promoting dataset sharing and utilization for scientific research remains challenging. Scientific datasets contain intricate domain-specific knowledge and contexts, rendering traditional collaborative filtering-based recommenders inadequate. Recent advances in Large Language Models (LLMs) offer unprecedented opportunities to build conversational agents capable of deep semantic understanding and personalized recommendations. In response, we present ScienceDB AI, a novel LLM-driven agentic recommender system developed on Science Data Bank (ScienceDB), one of the largest global scientific data-sharing platforms. ScienceDB AI leverages natural language conversations and deep reasoning to accurately recommend datasets aligned with researchers' scientific intents and evolving requirements. The system introduces several innovations: a Scientific Intention Perceptor to extract structured experimental elements from complicated queries, a Structured Memory Compressor to manage multi-turn dialogues effectively, and a Trustworthy Retrieval-Augmented Generation (Trustworthy RAG) framework. The Trustworthy RAG employs a two-stage retrieval mechanism and provides citable dataset references via Citable Scientific Task Record (CSTR) identifiers, enhancing recommendation trustworthiness and reproducibility. Through extensive offline and online experiments using over 10 million real-world datasets, ScienceDB AI has demonstrated significant effectiveness. To our knowledge, ScienceDB AI is the first LLM-driven conversational recommender tailored explicitly for large-scale scientific dataset sharing services. The platform is publicly accessible at: https://ai.scidb.cn/en.

cs.IR↗

arXiv2025

On-chip quadratically nonlinear photodetector

Involving deterministically nonlinear photoresponse in on-chip photodetector is intriguing to develop sophisticated functions in photonic integrated circuits, such as in-sensor computing and optoelectronic mixing, though the corresponding devices are still lack of sufficient investigation. Here, we demonstrate an on-chip quadratically nonlinear photodetector (QNPD) by configuring an InSe p-i-n homojunction on a silicon waveguide. Telecom-band light guiding in the waveguide couples with the InSe evanescently and is frequency up-converted into visible light via InSe's second-harmonic generation (SHG), which is subsequently absorbed by InSe and finally generates photocurrent under the built-in electric field of the p-i-n homojunction. Governed by these sequential processes, the on-chip QNPD presents a quadratic function between photocurrent and optical power. Thanks to the efficient SHG and well-established homojunction in InSe, the QNPD reaches a high normalized responsivity of 37.1 A/W2 and low dark current of 1 pA, representing greatly improved performances among reported nonlinear photodetectors. Benefiting from the extra SHG process, the on-chip QNPD intrinsically incorporates light-light interactions, enabling straightforwardly monitoring all-optically mixing signals electrically. As an example, an array of 16-pixel QNPDs was designed to implement a fully single-shot on-chip autocorrelator without requirement of bulky optics and external cameras, which precisely measures picosecond pulses with high sensitivity of 6.1*10-10 W2.

physics.optics↗

arXiv2025

Characterising Atomic-Scale Surface Disorder on 2D Materials Using Neutral Atoms

Two-dimensional (2D) transition metal dichalcogenides (TMDs), such as MoS2, have the potential to be widely used in electronic devices and sensors due to their high carrier mobility and tunable band structure. In 2D TMD devices, surface and interface cleanness can critically impact the performance and reproducibility. Even sample surfaces prepared under ultra-high vacuum (UHV) can be contaminated, causing disorder. On such samples, trace levels of submonolayer contamination remain largely overlooked, and conventional surface characterisation techniques have limited capability in detecting such adsorbates. Here, we apply scanning helium microscopy (SHeM), a non-destructive and ultra-sensitive technique, to investigate the surface cleanness of 2D MoS2. Our measurements reveal that even minute amounts of adventitious carbon induce atomic-scale disorder across MoS2 surfaces, leading to the disappearance of helium diffraction. By tracking helium reflectivity over time, we quantify the decay of surface order across different microscopic regions and find that flat areas are more susceptible to contamination than regions near edges. These findings highlight the fragility of surface order in 2D materials, even under UHV, and establish SHeM as a powerful tool for non-damaging microscopic 2D material cleanness characterisation. The approach offers a new route to wafer-scale characterisation of 2D material quality.

cond-mat.mtrl-sci↗

arXiv2025

Object-IR: Leveraging Object Consistency and Mesh Deformation for Self-Supervised Image Retargeting

Eliminating geometric distortion in semantically important regions remains an intractable challenge in image retargeting. This paper presents Object-IR, a self-supervised architecture that reformulates image retargeting as a learning-based mesh warping optimization problem, where the mesh deformation is guided by object appearance consistency and geometric-preserving constraints. Given an input image and a target aspect ratio, we initialize a uniform rigid mesh at the output resolution and use a convolutional neural network to predict the motion of each mesh grid and obtain the deformed mesh. The retargeted result is generated by warping the input image according to the rigid mesh in the input image and the deformed mesh in the output resolution. To mitigate geometric distortion, we design a comprehensive objective function incorporating a) object-consistent loss to ensure that the important semantic objects retain their appearance, b) geometric-preserving loss to constrain simple scale transform of the important meshes, and c) boundary loss to enforce a clean rectangular output. Notably, our self-supervised paradigm eliminates the need for manually annotated retargeting datasets by deriving supervision directly from the input's geometric and semantic properties. Extensive evaluations on the RetargetMe benchmark demonstrate that our Object-IR achieves state-of-the-art performance, outperforming existing methods in quantitative metrics and subjective visual quality assessments. The framework efficiently processes arbitrary input resolutions (average inference time: 0.009s for 1024x683 resolution) while maintaining real-time performance on consumer-grade GPUs. The source code will soon be available at https://github.com/tlliao/Object-IR.

cs.CV↗

arXiv2025

Omni-Mol: Multitask Molecular Model for Any-to-any Modalities

In the molecular domain, numerous studies have explored the use of multimodal large language models (LLMs) to construct a general-purpose, multi-task molecular model. However, these efforts are still far from achieving a truly universal molecular model. We identify three key challenges in this endeavor: (1) Existing molecular task datasets are typically small in scale and lack comprehensive domain coverage. (2) Tasks from different molecular subfields are difficult to effectively learn jointly through LLMs due to significant distributional shifts and competition among tasks, which introduces instability in the learning process. (3) Both inter-task and intra-task molecular representations demand different intrinsic dimensions in the language space, making it challenging to balance between redundancy and insufficiency in language model representations. To address these challenges, we innovatively categorize existing small-molecule tasks into four types: Mol2Mol, Mol2Text, Mol2Num, and Text2Mol. We then collect a dataset encompassing over 16 tasks with more than 1.4 million samples, making it the largest molecular instruction-tuning dataset to date. Leveraging the extensive pretraining of LLMs on existing chemical literature, we propose a novel multimodal LLM framework, named Omni-Mol, which unifies all small-molecule tasks and supports both molecular generation and understanding. The core of Omni-Mol is our proposed MoGE, which dynamically adapts to the intrinsic rank of different tasks. This mixture-of-experts architecture enhances the model's ability to handle diverse tasks and modalities effectively. Our model achieves unified instruction tuning across 16 tasks and attains state-of-the-art performance on 13 of them. Extensive experiments further demonstrate the scalability and versatility of Omni-Mol.

cs.LG↗

arXiv2025

The Role of Computing Resources in Publishing Foundation Model Research

Cutting-edge research in Artificial Intelligence (AI) requires considerable resources, including Graphics Processing Units (GPUs), data, and human resources. In this paper, we evaluate of the relationship between these resources and the scientific advancement of foundation models (FM). We reviewed 6517 FM papers published between 2022 to 2024, and surveyed 229 first-authors to the impact of computing resources on scientific output. We find that increased computing is correlated with national funding allocations and citations, but our findings don't observe the strong correlations with research environment (academic or industrial), domain, or study methodology. We advise that individuals and institutions focus on creating shared and affordable computing opportunities to lower the entry barrier for under-resourced researchers. These steps can help expand participation in FM research, foster diversity of ideas and contributors, and sustain innovation and progress in AI. The data will be available at: https://mit-calc.csail.mit.edu/

cs.CY↗

arXiv2025

APRIL: Active Partial Rollouts in Reinforcement Learning to Tame Long-tail Generation

Reinforcement learning (RL) has become a cornerstone in advancing large-scale pre-trained language models (LLMs). Successive generations, including GPT-o series, DeepSeek-R1, Kimi-K1.5, Grok 4, and GLM-4.5, have relied on large-scale RL training to enhance reasoning and coding capabilities. To meet the community's growing RL needs, numerous RL frameworks have been proposed. However, RL training remains computationally expensive, with rollout generation accounting for more than 90% of total runtime. In addition, its efficiency is often constrained by the long-tail distribution of rollout response lengths, where a few lengthy responses stall entire batches, leaving GPUs idle and underutilized. As model and rollout sizes continue to grow, this bottleneck increasingly limits scalability. To address this challenge, we propose Active Partial Rollouts in Reinforcement Learning (APRIL), which mitigates long-tail inefficiency. In the rollout phase, APRIL over-provisions rollout requests, terminates once the target number of responses is reached, and recycles incomplete responses for continuation in future steps. This strategy ensures that no rollouts are discarded while substantially reducing GPU idle time. Experiments show that APRIL improves rollout throughput by 22.5% on average (at most 44%) across commonly used RL algorithms (GRPO, DAPO, GSPO), accelerates convergence, and achieves 2.1% on average(at most 8%) higher final accuracy across tasks. Moreover, APRIL is both framework and hardware agnostic, already integrated into the slime RL framework, and deployable on NVIDIA and AMD GPUs alike. Taken together, this work unifies system-level and algorithmic considerations in proposing APRIL, with the aim of advancing RL training efficiency and inspiring further optimizations in RL systems. Our codebase is available at https://github.com/RLsys-Foundation/APRIL

cs.LG↗

arXiv2025

VLASCD: A Visual Language Action Model for Simultaneous Chatting and Decision Making

Recent large pretrained models such as LLMs (e.g., GPT series) and VLAs (e.g., OpenVLA) have achieved notable progress on multimodal tasks, yet they are built upon a multi-input single-output (MISO) paradigm. We show that this paradigm fundamentally limits performance in multi-input multi-output (MIMO) scenarios, where parallel task execution is required. In MISO architectures, tasks compete for a shared output channel, creating mutual exclusion effects that cause unbalanced optimization and degraded performance. To address this gap, we introduce MIMO-VLA (VLASCD), a unified training framework that enables concurrent multi-task outputs, exemplified by simultaneous dialogue generation and decision-making. Inspired by human cognition, MIMO-VLA eliminates interference between tasks and supports efficient parallel processing. Experiments on the CARLA autonomous driving platform demonstrate that MIMO-VLA substantially outperforms state-of-the-art MISO-based LLMs, reinforcement learning models, and VLAs in MIMO settings, establishing a new direction for multimodal and multitask learning.

cs.AI↗

arXiv2025

The GameTheory package for Macaulay2

We describe the GameTheory package version 1.0 for computing equilibria in game theory available since version 1.25.05 of Macaulay2. We briefly explain the four equilibrium notions, Nash, correlated, dependency, and conditional independence, and demonstrate their implementation in the package with examples.

math.AG↗

arXiv2025

Leveraging Local Patch Alignment to Seam-cutting for Large Parallax Image Stitching

Seam cutting has shown significant effectiveness in the composition phase of image stitching, particularly for scenarios involving parallax. However, conventional implementations typically position seam-cutting as a downstream process contingent upon successful image alignment. This approach inherently assumes the existence of locally aligned regions where visually plausible seams can be established. Current alignment methods frequently fail to satisfy this prerequisite in large parallax scenarios despite considerable research efforts dedicated to improving alignment accuracy. In this paper, we propose an alignment-compensation paradigm that dissociates seam quality from initial alignment accuracy by integrating a Local Patch Alignment Module (LPAM) into the seam-cutting pipeline. Concretely, given the aligned images with an estimated initial seam, our method first identifies low-quality pixels along the seam through a seam quality assessment, then performs localized SIFT-flow alignment on the critical patches enclosing these pixels. Finally, we recomposite the aligned patches using adaptive seam-cutting and merge them into the original aligned images to generate the final mosaic. Comprehensive experiments on large parallax stitching datasets demonstrate that LPAM significantly enhances stitching quality while maintaining computational efficiency. The code is available at https://github.com/tlliao/LPAM_seam-cutting.

cs.CV↗

arXiv2025

Heliometric stereo: a new frontier in surface profilometry

Accurate and reliable measurements of three-dimensional surface structures are important for a broad range of technological and research applications, including materials science, nanotechnology, and biomedical research. Scanning helium microscopy (SHeM) uses low-energy (64 meV) neutral helium atoms as the imaging probe particles, providing a highly sensitive and delicate approach to measuring surface topography. To date, topographic SHeM measurements have been largely qualitative, but with the advent of the heliometric stereo method - a technique that combines multiple images to create a 3D representation of a surface - quantitative maps of surface topography may now be acquired with SHeM. Here, we present and discuss two different implementations of heliometric stereo on two separate instruments, a single detector SHeM and a multiple-detector SHeM. Both implementations show good accuracy (5% and 10% respectively) for recovering the shape of a surface. Additionally, we discuss where heliometric stereo is most applicable, identify contrast features that can limit its accuracy, and discuss how to mitigate these limitations with careful design and sample choices that be readily implemented on current instruments.

physics.ins-det↗

arXiv2025

ParallelComp: Parallel Long-Context Compressor for Length Extrapolation

Extrapolating ultra-long contexts (text length >128K) remains a major challenge for large language models (LLMs), as most training-free extrapolation methods are not only severely limited by memory bottlenecks, but also suffer from the attention sink, which restricts their scalability and effectiveness in practice. In this work, we propose ParallelComp, a parallel long-context compression method that effectively overcomes the memory bottleneck, enabling 8B-parameter LLMs to extrapolate from 8K to 128K tokens on a single A100 80GB GPU in a training-free setting. ParallelComp splits the input into chunks, dynamically evicting redundant chunks and irrelevant tokens, supported by a parallel KV cache eviction mechanism. Importantly, we present a systematic theoretical and empirical analysis of attention biases in parallel attention-including the attention sink, recency bias, and middle bias-and reveal that these biases exhibit distinctive patterns under ultra-long context settings. We further design a KV cache eviction technique to mitigate this phenomenon. Experimental results show that ParallelComp enables an 8B model (trained on 8K context) to achieve 91.17% of GPT-4's performance under ultra-long contexts, outperforming closed-source models such as Claude-2 and Kimi-Chat. We achieve a 1.76x improvement in chunk throughput, thereby achieving a 23.50x acceleration in the prefill stage with negligible performance loss and pave the way for scalable and robust ultra-long contexts extrapolation in LLMs. We release the code at https://github.com/menik1126/ParallelComp.

cs.CL↗

arXiv2025

Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting

Point detection has been developed to locate pedestrians in crowded scenes by training a counter through a point-to-point (P2P) supervision scheme. Despite its excellent localization and counting performance, training a point-based counter still faces challenges concerning annotation labor: hundreds to thousands of points are required to annotate a single sample capturing a dense crowd. In this paper, we integrate point-based methods into a semi-supervised counting framework based on pseudo-labeling, enabling the training of a counter with only a few annotated samples supplemented by a large volume of pseudo-labeled data. However, during implementation, the training encounters issues as the confidence for pseudo-labels fails to be propagated to background pixels via the P2P. To tackle this challenge, we devise a point-specific activation map (PSAM) to visually interpret the phenomena occurring during the ill-posed training. Observations from the PSAM suggest that the feature map is excessively activated by the loss for unlabeled data, causing the decoder to misinterpret these over-activations as pedestrians. To mitigate this issue, we propose a point-to-region (P2R) scheme to substitute P2P, which segments out local regions rather than detects a point corresponding to a pedestrian for supervision. Consequently, pixels in the local region can share the same confidence with the corresponding pseudo points. Experimental results in both semi-supervised counting and unsupervised domain adaptation highlight the advantages of our method, illustrating P2R can resolve issues identified in PSAM. The code is available at https://github.com/Elin24/P2RLoss.

cs.CV↗