arXiv subjects

Junda Chen

Publications and source records attributed to Junda Chen.

At least 19 recordsLinked to original sources

arXiv2026

AutoLab: Can Frontier Models Solve Long-Horizon Auto Research and Engineering Tasks?

Scientific and engineering progress is fundamentally a long-horizon iterative process: proposing changes, running experiments, measuring outcomes, and continuously refining artifacts. Yet existing benchmarks for frontier models primarily evaluate either single-turn responses or short-horizon agent trajectories, failing to capture the challenges of sustained iterative improvement over extended time horizons. To address this gap, we introduce AutoLab, a new benchmark for ultra long-horizon closed-loop optimization. AutoLab consists of 36 realistic, expert-curated tasks spanning four diverse domains: system optimization, puzzle & challenge, model development, and CUDA kernel optimization. Each task begins with a correct but deliberately suboptimal baseline and challenges agents to improve it within a strict wall-clock budget. Evaluating 17 state-of-the-art models reveals the dominant predictor of success is not the quality of an agent's initial attempt, but its persistence in repeatedly benchmarking, editing, and incorporating empirical feedback. While claude-opus-4.6 exhibits strong long-horizon optimization capabilities, most frontier models, including several proprietary ones, either terminate prematurely or exhaust their budgets with minimal progress. These results underscore the importance of time awareness and persistent iteration in autonomous agents. We open-source the full benchmark, evaluation harness, and task artifacts, to accelerate research toward truly capable long-horizon agents.

cs.AI↗

arXiv2026

From Abstraction to Instantiation: Learning Behavioral Representation for Vision-Language-Action Model

Vision-Language-Action (VLA) models often suffer from performance degradation under distribution shifts, as they struggle to learn generalized behavior representations across varying environments. While existing approaches attempt to construct behavior representations through action-centric latent variables, they are often limited by short-horizon temporal fragmentation and static execution-alignment, leading to inconsistent behaviors in complex scenarios. To address these limitations, we propose \textbf{BehaviorVLA}, a framework that facilitates robust manipulation through the learning of a temporally coherent behavioral representations. Our approach features two symmetric components: (1) the \textbf{Visuomotor Behavior Encoder (VBE)}, which utilizes a causal Mamba-based architecture to aggregate long-horizon trajectory information into a unified behavior representation; and (2) the \textbf{Phase-conditioned Behavior Decoder (PBD)}, which decodes this representation into precise actions by dynamically aligning task-level priors with real-time execution progress. Experiments on RoboTwin 2.0, LIBERO, and CALVIN demonstrate state-of-the-art success rates of 58\%, 98\%, and 4.36 (Avg.Len), respectively. Notably, in real-world sim-to-real transfer, BehaviorVLA matches the performance of OpenVLA-OFT using only 50\% of the demonstration data, showcasing its superior data efficiency and generalization.

cs.CV↗

arXiv2025

VideoAgentTrek: Computer Use Pretraining from Unlabeled Videos

Training computer-use agents requires massive amounts of GUI interaction data, but manually annotating action trajectories at scale is prohibitively expensive. We present VideoAgentTrek, a scalable pipeline that automatically mines training data from publicly available screen-recorded videos at web scale, eliminating the need for manual annotation. Our approach addresses a key challenge: raw videos contain implicit demonstrations but lack explicit action labels. To solve this, we develop Video2Action, an inverse dynamics module (IDM) with two components: (1) a video grounding model that detects and localizes GUI actions with precise temporal boundaries and context, and (2) an action-content recognizer that extracts structured parameters like click coordinates and typed text with high fidelity. Applied to 39,000 YouTube tutorial videos, our pipeline generates 1.52 million interaction steps automatically. We leverage this data through continued pretraining followed by supervised fine-tuning. On OSWorld-Verified, our approach improves task success rates from 9.3% (SFT-only baseline) to 15.8%, a 70% relative improvement. On AgentNetBench, step accuracy increases from 64.1% to 69.3%. Our results demonstrate that passive internet videos can be transformed into high-quality supervision for computer-use agents, providing a scalable alternative to expensive manual annotation.

cs.CL↗

arXiv2025

Efficient Long-context Language Model Training by Core Attention Disaggregation

We present core attention disaggregation (CAD), a technique that improves long-context large language model training by decoupling the core attention computation, softmax(QK^T)V, from the rest of the model and executing it on a separate pool of devices. In existing systems, core attention is colocated with other layers; at long context lengths, its quadratic compute growth compared to the near-linear growth of other components causes load imbalance and stragglers across data and pipeline parallel groups. CAD is enabled by two observations. First, core attention is stateless: it has no trainable parameters and only minimal transient data, so balancing reduces to scheduling compute-bound tasks. Second, it is composable: modern attention kernels retain high efficiency when processing fused batches of token-level shards with arbitrary lengths. CAD partitions core attention into token-level tasks and dispatches them to dedicated attention servers, which dynamically rebatch tasks to equalize compute without sacrificing kernel efficiency. We implement CAD in a system called DistCA, which uses a ping-pong execution scheme to fully overlap communication with computation and in-place execution on attention servers to reduce memory use. On 512 H200 GPUs and context lengths up to 512k tokens, DistCA improves end-to-end training throughput by up to 1.35x, eliminates data and pipeline parallel stragglers, and achieves near-perfect compute and memory balance.

cs.LG↗

arXiv2025

Human vs. AI Safety Perception? Decoding Human Safety Perception with Eye-Tracking Systems, Street View Images, and Explainable AI

The way residents perceive safety plays an important role in how they use public spaces. Studies have combined large-scale street view images and advanced computer vision techniques to measure the perception of safety of urban environments. Despite their success, such studies have often overlooked the specific environmental visual factors that draw human attention and trigger people's feelings of safety perceptions. In this study, we introduce a computational framework that enriches the existing body of literature on place perception by using eye-tracking systems with street view images and deep learning approaches. Eye-tracking systems quantify not only what users are looking at but also how long they engage with specific environmental elements. This allows us to explore the nuance of which visual environmental factors influence human safety perceptions. We conducted our research in Helsingborg, Sweden, where we recruited volunteers outfitted with eye-tracking systems. They were asked to indicate which of the two street view images appeared safer. By examining participants' focus on specific features using Mean Object Ratio in Highlighted Regions (MoRH) and Mean Object Hue (MoH), we identified key visual elements that attract human attention when perceiving safe environments. For instance, certain urban infrastructure and public space features draw more human attention while the sky is less relevant in influencing safety perceptions. These insights offer a more human-centered understanding of which urban features influence human safety perceptions. Furthermore, we compared the real human attention from eye-tracking systems with attention maps obtained from eXplainable Artificial Intelligence (XAI) results. Several XAI models were tested, and we observed that XGradCAM and EigenCAM most closely align with human safety perceptual patterns.

cs.HC↗

arXiv2025

OpenCUA: Open Foundations for Computer-Use Agents

Vision-language models have demonstrated impressive capabilities as computer-use agents (CUAs) capable of automating diverse computer tasks. As their commercial potential grows, critical details of the most capable CUA systems remain closed. As these agents will increasingly mediate digital interactions and execute consequential decisions on our behalf, the research community needs access to open CUA frameworks to study their capabilities, limitations, and risks. To bridge this gap, we propose OpenCUA, a comprehensive open-source framework for scaling CUA data and foundation models. Our framework consists of: (1) an annotation infrastructure that seamlessly captures human computer-use demonstrations; (2) AgentNet, the first large-scale computer-use task dataset spanning 3 operating systems and 200+ applications and websites; (3) a scalable pipeline that transforms demonstrations into state-action pairs with reflective long Chain-of-Thought reasoning that sustain robust performance gains as data scales. Our end-to-end agent models demonstrate strong performance across CUA benchmarks. In particular, OpenCUA-72B achieves an average success rate of 45.0% on OSWorld-Verified, establishing a new state-of-the-art (SOTA) among open-source models. Further analysis confirms that our approach generalizes well across domains and benefits significantly from increased test-time computation. We release our annotation tool, datasets, code, and models to build open foundations for further CUA research.

cs.AI↗

arXiv2025

Pushing DSP-Free Coherent Interconnect to the Last Inch by Optically Analog Signal Processing

To support the boosting interconnect capacity of the AI-related data centers, novel techniques enabled high-speed and low-cost optics are continuously emerging. When the baud rate approaches 200 GBaud per lane, the bottle-neck of traditional intensity modulation direct detection (IM-DD) architectures becomes increasingly evident. The simplified coherent solutions are widely discussed and considered as one of the most promising candidates. In this paper, a novel coherent architecture based on self-homodyne coherent detection and optically analog signal processing (OASP) is demonstrated. Proved by experiment, the first DSP-free baud-rate sampled 64-GBaud QPSK/16-QAM receptions are achieved, with BERs of 1e-6 and 2e-2, respectively. Even with 1-km fiber link propagation, the BER for QPSK reception remains at 3.6e-6. When an ultra-simple 1-sps SISO filter is utilized, the performance degradation of the proposed scheme is less than 1 dB compared to legacy DSP-based coherent reception. The proposed results pave the way for the ultra-high-speed coherent optical interconnections, offering high power and cost efficiency.

physics.optics↗

arXiv2025

Quantum-Secured DSP-Lite Data Transmission Architectures for AI-Driven Data Centres

Artificial intelligence-driven (AI-driven) data centres, which require high-performance, scalable, energy-efficient, and secure infrastructure, have led to unprecedented data traffic demands. These demands involve low latency, high bandwidth connections, low power consumption, and data confidentiality. However, conventional optical interconnect solutions, such as intensity-modulated direct detection and traditional coherent systems, cannot address these requirements simultaneously. In particular, conventional encryption protocols that rely on complex algorithms are increasingly vulnerable to the rapid advancement of quantum computing. Here, we propose and demonstrate a quantum-secured digital signal processing-lite (DSP-Lite) data transmission architecture that meets all the stringent requirements for AI-driven data centre optical interconnects (AI-DCIs) scenarios. By integrating a self-homodyne coherent (SHC) system and quantum key distribution (QKD) through the multicore-fibre-based space division multiplexing (SDM) technology, our scheme enables secure, high-capacity, and energy-efficient data transmission while ensuring resilience against quantum computing threats. In our demonstration, we achieved an expandable transmission capacity of 2 Tbit per second (Tb/s) and a quantum secret key rate (SKR) of 229.2 kb/s, with a quantum bit error rate (QBER) of approximately 1.27% and with ultralow power consumption. Our work paves the way for constructing secure, scalable, and cost-efficient data transmission frameworks, thus enabling the next generation of intelligent, leak-proof optical interconnects for data centres.

quant-ph↗

arXiv2024

Efficiently Scaling LLM Reasoning with Certaindex

Test-time reasoning algorithms such as chain-of-thought, self-consistency, and MCTS enhance LLM problem-solving but can wastefully generate many tokens without improving accuracy. At the same time, we observe that these algorithms exhibit answer stabilization: their intermediate solutions often cease to change after a certain point, and further investment of compute does not change their final answer. To quantify this phenomenon, we introduce Certaindex, an algorithm-agnostic metric measuring this evolving stability, signaling when further computation is unlikely to alter the final result. Certaindex is lightweight, can accelerate reasoning program inference via early exit, and further enables dynamic token allocation, gang scheduling, and many opportunities when integrated with real-world LLM serving systems. To quantify real-world benefits, we built Certaindex as a scheduler into Dynasor, our reasoning-aware LLM serving system, and demonstrate up to 50% compute savings and 3.3x higher throughput in real workloads with no accuracy drop. Our code is available at https://github.com/hao-ai-lab/Dynasor.git

cs.LG↗

arXiv2024

No Request Left Behind: Tackling Heterogeneity in Long-Context LLM Inference with Medha

Deploying million-token Large Language Models (LLMs) is challenging because production workloads are highly heterogeneous, mixing short queries and long documents. This heterogeneity, combined with the quadratic complexity of attention, creates severe convoy effects where long-running requests stall short, interactive ones, degrading system responsiveness. We present Medha, a serving system that eliminates these convoys by introducing fine-grained, preemptive scheduling to LLM inference. Medha makes preemption practical with a co-designed set of mechanisms -- including Adaptive Chunking and Stream Pipeline Parallel that overcome the perceived inefficiencies and scaling challenges of chunking. Additionally, we present a new parallelism strategy KV-Cache Parallelism to reduce the decode latency and afford interactivity despite very long context. These mechanisms are orchestrated by a Length-Aware Relative Slack (LARS) scheduler, a deadline and heterogeneity-aware scheduling policy that prevents both the convoy effect and the starvation that plagues simpler policies. Under a heterogeneous workload, Medha improves throughput by 5.7x while reducing median and 99th percentile latency by 30x and 174x, respectively, compared to state-of-the-art non-preemptive systems.

cs.LG↗

arXiv2024

DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving

DistServe improves the performance of large language models (LLMs) serving by disaggregating the prefill and decoding computation. Existing LLM serving systems colocate the two phases and batch the computation of prefill and decoding across all users and requests. We find that this strategy not only leads to strong prefill-decoding interferences but also couples the resource allocation and parallelism plans for both phases. LLM applications often emphasize individual latency for each phase: time to first token (TTFT) for the prefill phase and time per output token (TPOT) of each request for the decoding phase. In the presence of stringent latency requirements, existing systems have to prioritize one latency over the other, or over-provision compute resources to meet both. DistServe assigns prefill and decoding computation to different GPUs, hence eliminating prefill-decoding interferences. Given the application's TTFT and TPOT requirements, DistServe co-optimizes the resource allocation and parallelism strategy tailored for each phase. DistServe also places the two phases according to the serving cluster's bandwidth to minimize the communication caused by disaggregation. As a result, DistServe significantly improves LLM serving performance in terms of the maximum rate that can be served within both TTFT and TPOT constraints on each GPU. Our evaluations show that on various popular LLMs, applications, and latency requirements, DistServe can serve 7.4x more requests or 12.6x tighter SLO, compared to state-of-the-art systems, while staying within latency constraints for > 90% of requests.

cs.DC↗

arXiv2023

Understanding Place Identity with Generative AI

Researchers are constantly leveraging new forms of data with the goal of understanding how people perceive the built environment and build the collective place identity of cities. Latest advancements in generative artificial intelligence (AI) models have enabled the production of realistic representations learned from vast amounts of data. In this study, we aim to test the potential of generative AI as the source of textual and visual information in capturing the place identity of cities assessed by filtered descriptions and images. We asked questions on the place identity of a set of 31 global cities to two generative AI models, ChatGPT and DALL-E2. Since generative AI has raised ethical concerns regarding its trustworthiness, we performed cross-validation to examine whether the results show similar patterns to real urban settings. In particular, we compared the outputs with Wikipedia data for text and images searched from Google for image. Our results indicate that generative AI models have the potential to capture the collective image of cities that can make them distinguishable. This study is among the first attempts to explore the capabilities of generative AI in understanding human perceptions of the built environment. It contributes to urban design literature by discussing future research opportunities and potential limitations.

cs.LG↗

arXiv2023

Cooperative guidance of multiple missiles: a hybrid co-evolutionary approach

Cooperative guidance of multiple missiles is a challenging task with rigorous constraints of time and space consensus, especially when attacking dynamic targets. In this paper, the cooperative guidance task is described as a distributed multi-objective cooperative optimization problem. To address the issues of non-stationarity and continuous control faced by cooperative guidance, the natural evolutionary strategy (NES) is improved along with an elitist adaptive learning technique to develop a novel natural co-evolutionary strategy (NCES). The gradients of original evolutionary strategy are rescaled to reduce the estimation bias caused by the interaction between the multiple missiles. Then, a hybrid co-evolutionary cooperative guidance law (HCCGL) is proposed by integrating the highly scalable co-evolutionary mechanism and the traditional guidance strategy. Finally, three simulations under different conditions demonstrate the effectiveness and superiority of this guidance law in solving cooperative guidance tasks with high accuracy. The proposed co-evolutionary approach has great prospects not only in cooperative guidance, but also in other application scenarios of multi-objective optimization, dynamic optimization and distributed control.

cs.NE↗

arXiv2022

Formation control with connectivity assurance for missile swarm: a natural co-evolutionary strategy approach

Formation control problem is one of the most concerned topics within the realm of swarm intelligence, which is usually solved by conventional mathematical approaches. In this paper, however, we presents a metaheuristic approach that leverages a natural co-evolutionary strategy to solve the formation control problem for a swarm of missiles. The missile swarm is modeled by a second-order system with heterogeneous reference target, and exponential error function is made to be the objective function such that the swarm converge to optimal equilibrium states satisfying certain formation requirements. Focusing on the issue of local optimum and unstable evolution, we incorporate a novel model-based policy constraint and a population adaptation strategies that greatly alleviates the performance degradation. With application of the Molloy-Reed criterion in the field of network communication, we developed an adaptive topology method that assure the connectivity under node failure and its effectiveness are validated both theoretically and experimentally. Experimental results valid the effectiveness of the proposed formation control approach. More significantly, we showed that it is feasible to treat generic formation control problem as Markov Decision Process(MDP) and solve it through iterative learning.

eess.SY↗

arXiv2022

Zenix: Efficient Execution of Bulky Serverless Applications

Serverless computing, commonly offered as Function-as-a-Service, was initially designed for small, lean applications. However, there has been an increasing desire to run larger, more complex applications (what we call bulky applications) in a serverless manner. Existing strategies for enabling such applications are to either increase function sizes or to rewrite applications as DAGs of functions. These approaches cause significant resource wastage, manual efforts, and/or performance overhead. We argue that the root cause of these issues is today's function-centric serverless model, where a function is the resource allocation and scaling unit. We propose a new, resource-centric serverless-computing model for executing bulky applications in a resource- and performance-efficient way, and we build the Zenix serverless platform following this model. Our results show that Zenix reduces resource consumption by up to 90% compared to today's function-centric serverless systems, while improving performance by up to 64%.

cs.DC↗

arXiv2021

Orbit configuration space of standard action and cellular methods of poset

We extend the existing idea of "cellular poset", introduce a collection of "cellular methods" for the computation of homology of intersection lattice of a complicated subspace arrangement, and for the computation of multiplicative structure induced by intersection. As an application, we give a presentation of cohomology ring of orbit configuration space of standard action by cellular methods and a spectral sequence associated with Grothendieck fibration of poset.

math.AT↗

arXiv2021

Cohomology ring of manifold arrangements

We study the cohomology ring of the complement $\mathcal{M}(\mathcal{A})$ of a manifold arrangement $\mathcal{A}$ in a smooth manifold $M$ without boundary. We first give the concept of monoidal cosheaf on a locally geometric poset $\mathfrak{L}$, and then define the generalized Orlik--Solomon algebra $A^*(\mathfrak{L}, \mathcal{C})$ over a commutative ring with unit, which is built by the classical Orlik--Solomon algebra and a monoidal cosheaf $\mathcal{C}$ as coefficients. Furthermore, we construct a monoidal cosheaf $\hat{\mathcal{C}}(\mathcal{A})$ associated with $\mathcal{A}$, so that the generalized Orlik--Solomon algebra $A^*(\mathfrak{L}, \hat{\mathcal{C}}(\mathcal{A}))$ becomes a double complex with suitable multiplication structure and the associated total complex $Tot(A^*(\mathfrak{L}, \hat{\mathcal{C}}(\mathcal{A})))$ is a differential algebra. Our main result is that $H^*(Tot(A^*(\mathfrak{L}, \hat{\mathcal{C}}(\mathcal{A}))))$ is isomorphic to $H^*(\mathcal{M}(\mathcal{A}))$ as algebras. Our argument is of topological with the use of a spectral sequence induced by a geometric filtration associated with $\mathcal{A}$. In particular, we also discuss the mixed Hodge complex structure on our model if $M$ and all elements in $\mathcal{A}$ are complex smooth varieties, and show that it induces the canonical mixed Hodge structure of $\mathcal{M}(\mathcal{A})$. As an application, we calculate the cohomology of chromatic configuration spaces, which agrees with many known results in some special cases. In addition, some explicit formulas with respect to Poincaré polynomial and chromatic polynomial are also given.

math.AT↗

arXiv2018

Statistical tracing of magnetic fields: comparing and improving the techniques

Magnetohydrodynamic(MHD) turbulence displays velocity anisotropies which reflect the direction of the magnetic field. This anisotropy has led to the development of a number of statistical techniques for studying magnetic fields in the interstellar medium. In this paper, we review and compare three techniques that use radio position-position-velocity data for determining magnetic field strength and morphology : the correlation function anisotropy (CFA), Principal Component Analysis of Anisotropies (PCAA), and the more recent Velocity Gradient Technique (VGT). We compare these three techniques and suggest improvements to the CFA and PCAA techniques to increase their accuracy and versatility. In particular, we suggest and successfully implement a much faster way of calculating non-periodic correlation functions for the CFA. We discuss possible improvements to the current implementation of the PCAA. We show the advantages of the VGT in terms of magnetic field tracing and stress the complementary nature with the other two techniques.

astro-ph.GA↗