Search arXiv⌕ Search

arXiv subjects

Liang Chen

Publications and source records attributed to Liang Chen.

At least 73 records · Page 4Linked to original sources
arXiv2026

BabyVision: Visual Reasoning Beyond Language

While humans develop core visual skills long before acquiring language, contemporary Multimodal LLMs (MLLMs) still rely heavily on linguistic priors to compensate for their fragile visual understanding. We uncovered a crucial fact: state-of-the-art MLLMs consistently fail on basic visual tasks that humans, even 3-year-olds, can solve effortlessly. To systematically investigate this gap, we introduce BabyVision, a benchmark designed to assess core visual abilities independent of linguistic knowledge for MLLMs. BabyVision spans a wide range of tasks, with 388 items divided into 22 subclasses across four key categories. Empirical results and human evaluation reveal that leading MLLMs perform significantly below human baselines. Gemini3-Pro-Preview scores 49.7, lagging behind 6-year-old humans and falling well behind the average adult score of 94.1. These results show despite excelling in knowledge-heavy evaluations, current MLLMs still lack fundamental visual primitives. Progress in BabyVision represents a step toward human-level visual perception and reasoning capabilities. We also explore solving visual reasoning with generation models by proposing BabyVision-Gen and automatic evaluation toolkit. Our code and benchmark data are released at https://github.com/UniPat-AI/BabyVision for reproduction.

cs.CV↗
arXiv2026

BCIJelly: An integrated ecosystem for brain-computer interface research

Brain-computer interface (BCI) research relies on multistage computational pipelines, yet progress remains constrained by fragmented data formats, heterogeneous decoder implementations and hardware-specific deployment toolchains, and researchers lack an integrated workflow. Here, we fill this gap with BCIJelly, a unified computational ecosystem that integrates 18 curated BCI datasets, 15 benchmark decoders and an algorithmic library of 80 reusable modules, an automated architecture search (AAS) procedure, and hardware-aware deployment through the toChip pipeline within a single Python framework. AAS constructs task-specific decoders without manual architecture design. It is further extended into a closed-loop mode guided by a large language model (LLM), which uses task specifications, module descriptions and search history to support multitask and cross-species decoding. The toChip pipeline compiles trained decoders for execution on neuromorphic chips, enabling energy-efficient deployment for BCI systems. An accompanying visualization software provides a graphical interface to the full workflow, making BCIJelly accessible without programming. We validate BCIJelly across five BCI paradigms (motor, visual, speech, emotion and auditory) with recordings from humans, macaques and mice, and single-task, multitask and cross-species decoding settings. BCIJelly establishes a unified and extensible infrastructure that bridges decoder development and hardware-aware deployment for BCI research.

cs.HC↗
arXiv2026

Quantum Limits to Ground-State Cooling of Traveling Hypersound Phonons

The steady final phonon occupation in waveguide optomechanical systems based on backward stimulated Brillouin-Mandelstam scattering has not been established in the strong-coupling regime. In this work, the displacement spectra of anti-Stokes optical modes and acoustic modes in tapered chalcogenide photonic crystal fiber are derived from the Lindblad (or Gorini-Kossakowski-Sudarshan-Lindblad) master equation. By analyzing the full spectral response, we indicate that the system can enter the strong-coupling regime through the emergence of normal-mode splitting and avoided crossings. Within a non-Hermitian framework, the threshold for strong coupling is identified, showing that it can be achieved at relatively low pump power even at room temperature. Furthermore, we derive a unified analytical expression for the final phonon occupation, revealing that quantum backaction and zero-point fluctuations impose additional fundamental limits that hinder the achievement of ground-state cooling. These results redefine the quantum limits of steady-state cooling in continuum optomechanics, motivating the search for new strategies to access the quantum ground-state of macroscopic phonons.

physics.optics↗
arXiv2026

WarpagePINN: Thermal Warpage Prediction in Advanced Packaging via a Two-Stage Physics-Informed Neural Networks

Thermal warpage has become a critical issue in advanced packaging, primarily caused by the mismatch in coefficients of thermal expansion (CTE) among heterogeneously integrated materials. However, only a limited number of studies have focused on developing computational methods for coupled thermal-warpage prediction in the chiplet. This paper proposes a two-stage physics-informed neural network (WarpagePINN) framework to compute both temperature profile and warpage deformation of chiplets. The neural networks are trained without relying on labeled datasets generated by conventional simulators. In the first stage, the temperature field is modeled using a Fourier series representation that inherently satisfies boundary conditions, and the network is trained solely through a loss function derived from the governing equation. In the second stage, a multilayer perceptron (MLP) is employed for warpage prediction, utilizing a novel hybrid supervisory strategy to optimize the energy-based loss function instead of residual loss. A parametric WarpagePINN is also developed to quantify uncertainties associated with the CTE. Numerical results show that the proposed WarpagePINN framework achieves excellent agreement with conventional finite element methods, with a mean absolute error (MAE) of 0.2 μm, while achieving a speedup of approximately 1000 {\times} in CTE parameterization studies.

math.NA↗
arXiv2026

Energy-Dependent Shifts of Medium-Scale Anisotropies in Very-High-Energy Cosmic Rays Observed by LHAASO-KM2A

Small deviations from isotropy in the arrival directions of Galactic cosmic rays serve as a unique probe of the local magnetic environment. In this Letter, we report observations of medium-scale anisotropies (MSA) at energies above 10 TeV using the LHAASO-KM2A array. Our analysis identifies four regions of excess and four regions of deficit, each spanning angular scales of approximately ten degrees. Crucially, we detect significant energy-dependent shifts in the centroids of two excess regions: Region B and the newly identified Region $\mathrm{\widetilde{D}}$. We also characterize the energy evolution of the fractional relative intensity across both excess and deficit regions. These findings imply that the observed anisotropies are shaped by the specific realization of the local turbulent magnetic field within the cosmic ray scattering length. Such energy-dependent behaviors impose strict constraints on local turbulence models and cosmic ray propagation theories.

astro-ph.HE↗
arXiv2026

Extreme PeV accelerator associated with GRS 1915+105

Microquasars, binary systems featuring relativistic jets, have emerged as sources for particle acceleration beyond PeV energies. We present a study of the broadband $γ$-ray emission from one of the most prominent Galactic microquasars GRS 1915+105 based on data accumulated by LHAASO and Fermi-LAT over 4 and 17 years, respectively. A joint analysis of LHAASO-WCDA and LHAASO-KM2A data reveals extended $γ$-ray emission whose centroid appears significantly shifted, by ~ 0.13°, from the binary system and its jets. The spectral energy distribution is well described by a curved spectrum with progressive steepening that can be described by a log-parabola function with no evidence for a sharp cutoff, consistent with parent particles reaching multi-PeV energies and an extreme acceleration efficiency approaching the limit set by the available potential drop across the source. Several features, most notably the shift of the emission and single-power-law spectrum down to GeV band, favor radiation by cosmic rays accelerated in the source interacting with the dense ambient medium. Our spectral modeling implies that at least a few percent of the jet mechanical power is transferred to protons, whose maximum energy reaches beyond 5 PeV. These results strengthen the case for microquasars as exceptionally efficient accelerators in our Galaxy.

astro-ph.HE↗
arXiv2026

Finite-Time Electrometry with a Quantum-Regime Single-Ion Phonon Laser

The phonon laser realized in a trapped ion, i.e., a self-sustained mechanical oscillator, has demonstrated the unique characteristics in practically detecting externally applied electric signals without the prerequisite of sideband cooling. Entering the quantum regime via sideband cooling is expected to further improve its sensing performance. Here we report the first experimental realization of a quantum-regime single-ion phonon laser ($\bar{n}<10$) using a trapped $^{40}\mathrm{Ca}^+$ ion and demonstrate electrometry based on its phase-space symmetry-breaking response to weak resonant electric fields. By tuning the phonon-laser parameters, we reveal that the sensing performance is fundamentally governed by the finite-time relaxation dynamics of the underlying open quantum system. We find that a slow Liouvillian relaxation, correlated with the finite experimental interaction window, effectively enhances the dynamic susceptibility while maintaining the structural robustness of the limit cycle. This regime, when applied to the detection of electric fields, produces a shot-noise-limited peak sensitivity of $14.15 \pm 0.77~μ\mathrm{V/m}/\sqrt{\mathrm{Hz}}$ and a minimum detectable field variation of $δE_{\mathrm{min}} \approx 1.83~μ\mathrm{V/m}$. Our results establish quantum phonon lasers as a practical platform for advanced sensing and highlight the central role of Liouvillian dynamics in non-equilibrium electrometry.

quant-ph↗
arXiv2026

Simulated Customers Never Walk Away: Decision Fidelity of LLM User Simulators Measured Against Real Purchase Outcomes

LLM-as-user-simulation has become core infrastructure for conversational AI: agent benchmarks (tau-bench), training pipelines, and a growing body of fidelity studies all rely on LLMs role-playing the human side of dialogue. Existing frameworks measure communicative fidelity -- whether simulators talk like humans -- against ground truth from paid participants role-playing assigned goals. We argue this has a structural blind spot: when the goal is assigned, the user's willingness is exogenous, so no framework can test whether simulators make decisions like real users whose motivation is endogenous, latent, and decaying. We introduce decision fidelity -- whether a simulated population reproduces the decision-state dynamics of real users facing real, consequential choices -- and measure it on a unique testbed: 2,790 production conversations between an LLM sales agent and real customers, including 793 with verified payment outcomes. Using a teacher-forced probe protocol that holds context and instrument fixed, we find a systematic, outcome-correlated failure we call the disengagement deficit: simulators reproduce eventual buyers almost exactly (depth bias +0.09) but inflate eventual non-buyers toward the purchase frame (depth bias +0.40; d=0.38, p<0.001), halving expressed resistance (25.1% to 13.5%) and nearly doubling deliberation (21.9% to 40.1%) while fabricating no purchases. The deficit replicates across model families (DeepSeek: d=0.41, p=0.002) and resists the obvious fix: instructing the simulator that it may disengage cuts marginal bias five-fold but barely moves the outcome-conditioned contrast (d=0.34, p=0.008). Real non-buyers say "not now" and stop; simulated non-buyers ask about price. Evaluating or training sales and persuasion agents against such simulators overstates funnel progress exactly where it matters most -- the customers who walk away.

cs.AI↗
arXiv2026

AgentCanary: A Security Evaluation Framework for Autonomous AI Agents in Real Executable Environments

Autonomous AI agents have driven the transition from conversation to task execution, shifting security failures from textual deception to system compromise. Although security evaluation is crucial for proactive risk prevention, prior work is constrained by fundamental bottlenecks, including fragmented risk coverage, static or low-fidelity execution environments, and single-dimensional and coarse-grained assessment metrics. To address these challenges, we propose AgentCanary, a comprehensive security evaluation framework for autonomous AI agents. AgentCanary provides a systematic solution along three contributions. First, comprehensive risk coverage: we introduce an orthogonal Entry $\times$ Impact risk taxonomy that decouples how adversarial influence enters the agent from what harm it ultimately causes, and instantiate it as a scenario-aligned task suite spanning realistic deployment workflows. Second, a high-fidelity real executable environment: rather than static Q&A or mocked tool responses, agents interact with real tools against dynamically provisioned task artifacts, with persistent state across multi-step interactions that naturally supports long-horizon attack evaluation. Third, trajectory-grounded multi-dimensional evaluation: evaluation consumes the full agent trajectory rather than the reply text or a single tool call, enabling decomposed scoring along three orthogonal dimensions, Outcome Safety, Security Awareness, and Task Utility. We evaluate a broad set of frontier models on AgentCanary against multiple established adversarial attack methods across three agent frameworks. The results reveal that current agents often fail to recognize the attacks they face, particularly under compromised skills, persistent state, and long-horizon execution attacks, and provide a systematic baseline for developing more reliable and secure agent systems.

cs.CR↗
arXiv2026

Physics-Driven Semantic Scattering Structure Understanding of Aircraft Target in SAR Images

Synthetic aperture radar (SAR) has become indispensable for target interpretation owing to its all-day and all-weather observation capability. In SAR target interpretation, electromagnetic scattering information provides a physically grounded cue beyond visual texture and has been widely exploited for target interpretation. However, existing methods remain dominated by local scattering center representations. Such unordered and component-agnostic representations are highly unstable for aircraft targets. As a result, physically existing components with weak scattering responses are often missed, resulting in the incomplete reconstructed topology structure. To address this limitation, we establish Semantic Scattering Structure Understanding as a new paradigm for SAR aircraft interpretation. Semantic scattering keypoints are defined to associate local electromagnetic responses with physically meaningful aircraft components, while visibility-aware attributes are introduced to retain weakly observable yet physically existed components. The keypoints are further organized into a stable semantic scattering structure. Build upon this, we propose S3U-SAR, a physics-driven framework to localize semantic scattering keypoints and construct the complete representation constrained by multi-dimensional physical priors containing scattering heterogeneity, rigid-body topology, speckle uncertainty. A confidence-gated joint supervision strategy is further introduced to alleviate optimization conflicts. We construct KP-SAR-Aircraft-1.0, the first fine-grained benchmark for semantic scattering structure understanding. Extensive experiments demonstrate that S3U-SAR achieves the best performance compared with baselines. Cross-category and cross-dataset evaluations further verify its robustness and transferability.

eess.IV↗
arXiv2026

Beyond Two-Stage Training: Cooperative SFT and RL for LLM Reasoning

Supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR) are two widely used post-training paradigms for improving the reasoning ability of large language models (LLMs). Recent methods attempt to integrate SFT and RLVR in a single stage by reweighting or scheduling their objectives. However, such coupling can be counterproductive because supervised updates are not uniformly beneficial for reward optimization. To address this, we propose BRIDGE, a scalable framework in which SFT learns to supervise RL by selectively transferring knowledge that improves reward optimization. Specifically, BRIDGE alternates two updates at each meta-training step: a base-model update that fuses the SFT and RL gradients, and an update to a lightweight low-rank adapter (LoRA) that coordinates the two objectives by maximizing a cooperative-gain signal, defined as the reward of joint SFT-RL training over an RL-only baseline. Across five mathematical reasoning benchmarks, BRIDGE consistently outperforms two-stage cold start, naive mixing, and representative single-stage integration baselines, yielding over three points average absolute improvement and more stable training dynamics. We further show that BRIDGE extends to logical reasoning and generalizes out-of-distribution to code and science without additional training, while staying robust under noisy rewards.

cs.CL↗
arXiv2026

Looking Beyond Text: Reducing Language bias in Large Vision-Language Models via Multimodal Dual-Attention and Soft-Image Guidance

Large vision-language models (LVLMs) have achieved impressive results in various vision-language tasks. However, despite showing promising performance, LVLMs suffer from hallucinations caused by language bias, leading to diminished focus on images and ineffective visual comprehension. We identify two primary reasons for this bias: 1. Different scales of training data between the pretraining stage of LLM and multimodal alignment stage. 2. The learned inference bias due to short-term dependency of text data. Therefore, we propose LACING, a systemic framework designed to address the language bias of LVLMs with muLtimodal duAl-attention meChanIsm (MDA) aNd soft-image Guidance (IFG). Specifically, MDA introduces a parallel dual-attention mechanism that enhances the integration of visual inputs across the model. IFG introduces a learnable soft visual prompt during training and inference to replace visual inputs, designed to compel LVLMs to prioritize text inputs. Then, IFG further proposes a novel decoding strategy using the soft visual prompt to mitigate the model's over-reliance on adjacent text inputs. Comprehensive experiments demonstrate that our method effectively debiases LVLMs from their language bias, enhancing visual comprehension and reducing hallucinations without requiring additional training resources or data. The code and model are available at [lacing-lvlm.github.io](https://lacing-lvlm.github.io).

cs.CV↗
arXiv2026

MENTOR: Efficient Multimodal-Conditioned Tuning for Autoregressive Vision Generation Models

Recent text-to-image models produce high-quality results but still struggle with precise visual control, balancing multimodal inputs, and requiring extensive training for complex multimodal image generation. To address these limitations, we propose MENTOR, a novel autoregressive (AR) framework for efficient Multimodal-conditioned Tuning for Autoregressive multimodal image generation. MENTOR combines an AR image generator with a two-stage training paradigm, enabling fine-grained, token-level alignment between multimodal inputs and image outputs without relying on auxiliary adapters or cross-attention modules. The two-stage training consists of: (1) a multimodal alignment stage that establishes robust pixel- and semantic-level alignment, followed by (2) a multimodal instruction tuning stage that balances the integration of multimodal inputs and enhances generation controllability. Despite modest model size, suboptimal base components, and limited training resources, MENTOR achieves strong performance on the DreamBench++ benchmark, outperforming competitive baselines in concept preservation and prompt following. Additionally, our method delivers superior image reconstruction fidelity, broad task adaptability, and improved training efficiency compared to diffusion-based methods. Dataset, code, and models are available at: https://github.com/HaozheZhao/MENTOR

cs.CV↗
arXiv2026

Crafter: A Multi-Agent Harness for Editable Scientific Figure Generation from Diverse Inputs

Scientific figures are among the most effective means of communicating complex research ideas, yet producing publication-quality illustrations remains one of the most labor-intensive parts of paper preparation. Existing automated systems each target a single figure type under text-only input, leaving the diversity of types and conditions researchers actually use unaddressed; their raster outputs further cannot be locally revised. Because scientific figures are structured compositions of discrete semantic components, the localized errors generators produce on such layouts demand not a stronger backbone but a harness. We instantiate this harness in two complementary systems: Crafter, a multi-agent harness for figure generation that generalizes across figure types and input conditions without architectural changes, and CraftEditor, which applies the same pattern to convert raster outputs into editable SVGs. Moreover, we introduce CraftBench, a benchmark spanning three figure types and four input conditions with human quality annotation. Experiments show that Crafter substantially outperforms both standalone generators and the agentic baseline on PaperBanana-Bench and CraftBench, with ablations confirming each component's independent contribution; CraftEditor faithfully converts outputs into editable SVGs that surpass all baselines. Our code and benchmark are available at https://github.com/HaozheZhao/Crafter.

cs.CV↗
arXiv2026

Revisiting Graph Autoencoders as Implicit Contrastive Learners

Graph autoencoders (GAEs) and graph contrastive learning (GCL) are two major paradigms for self-supervised representation learning on graphs, yet they are often studied in isolation and treated as fundamentally different approaches. In this work, we revisit GAEs through the lens of contrastive learning and show that both structure-based and feature-based GAEs can be conceptualized as implicitly graph contrastive learners. This perspective reveals that many existing GAEs differ primarily in how contrastive views are constructed, rather than in their learning objectives or architectures. Building on this insight, we introduce a unified formulation that highlights contrastive view design as a central and previously less explored dimension in GAEs. In particular, we identify asymmetric contrastive views, arising from mismatches in subgraph views, as an important yet underexplored design axis in prior GAE research. We formalize this insight within a unified framework and conduct systematic experiments on representative graph learning tasks to examine its impact on performance and efficiency. Our results show that interpreting GAEs as implicit contrastive learners offers a clearer understanding of existing models and provides practical guidance for designing effective and scalable graph autoencoders.

cs.LG↗
arXiv2026

SaaS-Bench: Can Computer-Use Agents Leverage Real-World SaaS to Solve Professional Workflows?

Computer-Using Agents (CUAs) are rapidly extending large language models (LLMs) beyond text-based reasoning toward action execution in more complex environments, such as web browsers and graphical user interfaces (GUIs). However, existing web and GUI agent benchmarks often rely on simplified settings, isolated tasks, or short-horizon interactions, making it difficult to assess capabilities of agents in realistic professional workflows. Software-as-a-Service (SaaS) environments are a natural choice for CUA evaluation, as they host a large share of modern digital work and naturally involve dynamic system states, cross-application coordination, domain-specific knowledge, and long-horizon dependencies. To this end, we introduce SaaS-Bench, a benchmark built on 23 deployable SaaS systems across six professional domains, containing 106 tasks grounded in realistic work scenarios. These tasks require long-horizon execution, cover both text-only and multimodal settings, and are evaluated with weighted verification checkpoints that measure strict task completion and partial progress. Experiments show that representative LLM-based agents struggle on SaaS-Bench, with even the strongest model completing fewer than 4% of tasks end-to-end, exposing limitations in planning, state tracking, cross-application context maintenance, and error recovery. Code are available at https://github.com/UniPat-AI/SaaS-Bench for reproduction.

cs.AI↗
arXiv2026

EvoCode-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions

Coding agents are increasingly used as iterative development partners, but most benchmarks still evaluate one specification followed by one final assessment. This leaves out a basic question: can an agent keep its own codebase working as requirements change? We introduce EvoCode-Bench, a benchmark of 26 stateful coding tasks and 227 evaluated rounds. Each task preserves the agent's workspace for 5-15 rounds, states requirements through observable behavior, and uses cumulative executable tests to check new requirements and still-active prior ones. We evaluate 13 coding agents with two metrics: MT@4, a four-attempt fail-stop multi-round score, and SR, a single-round score from a reference-completed prior state. For most agents, SR exceeds MT@4 by 22-40 points. The gap also changes rankings: the highest-SR agent (78.9) ranks only third in persistent execution (44.0 MT@4). Even the strongest agents achieve only about 50% success on multi-turn metrics, and aggregate pass rate drops below half of round-1 performance by round 5. Failure analysis shows tier-dependent behavior: weaker agents fail early, while stronger agents survive long enough to expose specification-tracking and regression failures. We release the benchmark data and Harbor multi-turn infrastructure.

cs.AI↗
arXiv2026

An efficient second-order cone programming approach for dynamic optimal transport on staggered grid discretization

This paper proposes an efficient numerical method based on second-order cone programming (SOCP) to solve dynamic optimal transport (DOT) problems with quadratic cost on staggered grid discretization. By properly reformulating discretized DOT problems into a linear SOCP, the proposed method eliminates the interpolation matrices and thus avoids solving a series of cubic equations and linear systems induced by interpolation. Then, by taking advantage of the SOCP reformulation, we can solve them efficiently by a computationally highly economical implementation of an inexact decomposition-based proximal augmented Lagrangian method. Moreover, we have made the proposed approach an open-source software package. Numerical experiments on various DOT problems suggest that the proposed approach performs significantly more efficiently than state-of-the-art software packages. In addition, it exhibits prominent robustness to problems with non-negative measures.

math.OC↗