arXiv subjects

Cailian Chen

Publications and source records attributed to Cailian Chen.

At least 19 recordsLinked to original sources

arXiv2026

Sample Complexity of Policy Gradient for Log-Growth Control

We study the sample complexity of policy gradient for log-growth control -- the problem of learning, from observed state transitions, a feedback gain that optimally stabilizes a scalar linear system driven through a multiplicative-noise actuation channel. The objective $J(K) = \mathbb{E}[\log|1+BK|]$ is the top Lyapunov exponent of the closed loop. This problem carries a structural difficulty we call the cusp obstruction: the optimal gain $K^*$ always places the noise singularity $b_{\rm sing}(K) = -1/K$ in the interior of the support. At this singular optimum the policy gradient exists only as a Cauchy principal value, not as a Lebesgue integral, and the natural single-sample gradient estimator has infinite variance. Standard first-order stochastic-optimization analysis is thus inapplicable at the optimum, and merely smoothing the objective does not resolve the difficulty. The obstruction, however, has an exploitable symmetry: the Cauchy kernel is an odd function of the displacement from the moving pole, so pairing each observation with its reflection through the pole cancels the divergent part. This one cancellation simultaneously controls the population curvature, the gradient-estimator variance, and the bias incurred when the noise density is estimated. Combining these bounds with a closed-form single-transition gradient oracle, we prove that projected mini-batch policy gradient, initialized in any compact subset of the stabilizing region, attains total sample complexity $\tilde{O}(1/\eta)$ when the noise density is known and $\tilde{O}(\eta^{-(2s+1)/(2s)})$ when it must be estimated, for $C^s$ noise densities with $s \geq 2$.

eess.SY↗

arXiv2026

GTA-2: Benchmarking General Tool Agents from Atomic Tool-Use to Open-Ended Workflows

The development of general-purpose agents requires a shift from executing simple instructions to completing complex, real-world productivity workflows. However, current tool-use benchmarks remain misaligned with real-world requirements, relying on AI-generated queries, dummy tools, and limited system-level coordination. To address this, we propose GTA-2, a hierarchical benchmark for General Tool Agents (GTA) spanning atomic tool use and open-ended workflows. Built on real-world authenticity, it leverages real user queries, deployed tools, and multimodal contexts. (i) GTA-Atomic, inherited from our prior GTA benchmark, evaluates short-horizon, closed-ended tool-use precision. (ii) GTA-Workflow introduces long-horizon, open-ended tasks for realistic end-to-end completion. To evaluate open-ended deliverables, we propose a recursive checkpoint-based evaluation mechanism that decomposes objectives into verifiable sub-goals, enabling unified evaluation of both model capabilities and agent execution frameworks (i.e., execution harnesses). Experiments reveal a pronounced capability cliff: while frontier models already struggle on atomic tasks (below 50%), they largely fail on workflows, with top models achieving only 14.39% success. Further analysis shows that checkpoint-guided feedback improves performance, while advanced frameworks such as Manus and OpenClaw substantially enhance workflow completion, highlighting the importance of execution harness design beyond the underlying model capacity. These findings provide guidance for developing reliable personal and professional assistants. Dataset and code will be available at https://github.com/open-compass/GTA.

cs.CL↗

arXiv2026

AgentConductor: Topology Evolution for Multi-Agent Competition-Level Code Generation

Large language model(LLM)-driven multi-agent systems(MAS) coordinate specialized agents through predefined interaction topologies and have shown promise for complex tasks such as competition-level code generation. Recent studies demonstrate that carefully designed multi-agent workflows and communication graphs can significantly improve code generation performance by leveraging collaborative reasoning. However, existing methods neither adapt topology density to task difficulty nor iteratively refine the topology within an instance using execution feedback, which leads to redundant communication and performance bottlenecks. To address these issues, we propose AgentConductor: a reinforcement learning-optimized MAS with an LLM-based orchestrator agent as its core, which enables end-to-end feedback-driven dynamic generation of interaction topologies. For each query, AgentConductor infers agent roles and task difficulty, then constructs a task-adapted, density-aware layered directed acyclic graph (DAG) topology, underpinned by two key innovations. First, we design a novel topological density function that captures communication-aware mathematical characterizations of multi-agent interactions. Second, we adopt difficulty interval partitioning to avoid excessive pruning for precise topological density upper bound measurement per difficulty level and finer-grained control. Empirically, across three competition-level and two foundational code datasets, AgentConductor achieves state-of-the-art accuracy, outperforming the strongest baseline by up to 14.6% in pass@1 accuracy, 13% in density reduction, and 68% in token cost reduction.

cs.MA↗

arXiv2026

RouteMoA: Dynamic Routing without Pre-Inference Boosts Efficient Mixture-of-Agents

Mixture-of-Agents (MoA) improves LLM performance through layered collaboration, but its dense topology raises costs and latency. Existing methods employ LLM judges to filter responses, yet still require all models to perform inference before judging, failing to cut costs effectively. They also lack model selection criteria and struggle with large model pools, where full inference is costly and can exceed context limits. To address this, we propose RouteMoA, an efficient mixture-of-agents framework with dynamic routing. It employs a lightweight scorer to perform initial screening by predicting coarse-grained performance from the query, narrowing candidates to a high-potential subset without inference. A mixture of judges then refines these scores through lightweight self- and cross-assessment based on existing model outputs, providing posterior correction without additional inference. Finally, a model ranking mechanism selects models by balancing performance, cost, and latency. RouteMoA outperforms MoA across varying tasks and model pool sizes, reducing cost by 89.8% and latency by 63.6% in the large-scale model pool.

cs.AI↗

arXiv2025

A First Look at Inter-Cell Interference in the Wild

In cellular networks, inter-cell interference management has been studied for decades, yet its real-world effectiveness remains under-explored. To bridge this gap, we conduct a first measurement study of inter-cell interference for operational 4G/5G networks. Our findings reveal the prevalence of inter-cell interference and a surprising absence of interference coordination among operational base stations. As a result, user equipments experience unnecessary interference, which causes significant signal quality degradation, especially under frequency-selective channel fading. We examine the inter-cell interference issues from four major perspectives: network deployment, channel assignment, time-frequency resource allocation, and network configuration. In none of these dimensions is inter-cell interference effectively managed. Notably, even when spectrum resources are underutilized and simple strategies could effectively mitigate inter-cell interference, base stations consistently prioritize using the same set of time-frequency resources, causing interference across cells. Our measurements reveal substantial opportunities for improving signal quality by inter-cell interference management.

cs.NI↗

arXiv2025

Distributed Constraint-coupled Resource Allocation: Anytime Feasibility and Violation Robustness

This paper considers distributed resource allocation problems (DRAPs) with a coupled constraint for real-time systems. Based on primal-dual methods, we adopt a control perspective for optimization algorithm design by synthesizing a safe feedback controller using control barrier functions to enforce constraint satisfaction. On this basis, a distributed anytime-feasible resource allocation (DanyRA) algorithm is proposed. It is shown that DanyRA algorithm converges to the exact optimal solution of DRAPs while ensuring feasibility of the coupled inequality constraint at all time steps. Considering constraint violation arises from potential external interferences, a virtual queue with minimum buffer is incorporated to restore the constraint satisfaction before the pre-defined deadlines. We characterize the trade-off between convergence accuracy and violation robustness for maintaining or recovering feasibility. DanyRA algorithm is further extended to address DRAPs with a coupled equality constraint, and its linear convergence rate is theoretically established. Finally, a numerical example is provided for verification.

math.OC↗

arXiv2025

NTIRE 2025 Image Shadow Removal Challenge Report

This work examines the findings of the NTIRE 2025 Shadow Removal Challenge. A total of 306 participants have registered, with 17 teams successfully submitting their solutions during the final evaluation phase. Following the last two editions, this challenge had two evaluation tracks: one focusing on reconstruction fidelity and the other on visual perception through a user study. Both tracks were evaluated with images from the WSRD+ dataset, simulating interactions between self- and cast-shadows with a large number of diverse objects, textures, and materials.

cs.CV↗

arXiv2024

CAD-GPT: Synthesising CAD Construction Sequence with Spatial Reasoning-Enhanced Multimodal LLMs

Computer-aided design (CAD) significantly enhances the efficiency, accuracy, and innovation of design processes by enabling precise 2D and 3D modeling, extensive analysis, and optimization. Existing methods for creating CAD models rely on latent vectors or point clouds, which are difficult to obtain, and storage costs are substantial. Recent advances in Multimodal Large Language Models (MLLMs) have inspired researchers to use natural language instructions and images for CAD model construction. However, these models still struggle with inferring accurate 3D spatial location and orientation, leading to inaccuracies in determining the spatial 3D starting points and extrusion directions for constructing geometries. This work introduces CAD-GPT, a CAD synthesis method with spatial reasoning-enhanced MLLM that takes either a single image or a textual description as input. To achieve precise spatial inference, our approach introduces a 3D Modeling Spatial Mechanism. This method maps 3D spatial positions and 3D sketch plane rotation angles into a 1D linguistic feature space using a specialized spatial unfolding mechanism, while discretizing 2D sketch coordinates into an appropriate planar space to enable precise determination of spatial starting position, sketch orientation, and 2D sketch coordinate translations. Extensive experiments demonstrate that CAD-GPT consistently outperforms existing state-of-the-art methods in CAD model synthesis, both quantitatively and qualitatively.

cs.CV↗

arXiv2024

OIDM: An Observability-based Intelligent Distributed Edge Sensing Method for Industrial Cyber-Physical Systems

Industrial cyber-physical systems (ICPS) integrate physical processes with computational and communication technologies in industrial settings. With the support of edge computing technology, it is feasible to schedule large-scale sensors for efficient distributed sensing. In the sensing process, observability is the key to obtaining complete system states, and stochastic scheduling is more suitable considering uncertain factors in wireless communication. However, existing works have limited research on observability in stochastic scheduling. Targeting this issue, we propose an observability-based intelligent distributed edge sensing method (OIDM). Deep reinforcement learning (DRL) methods are adopted to optimize sensing accuracy and power efficiency. Based on the system's ability to achieve observability, we establish a bridge between observability and the number of successful sensor transmissions. Novel linear approximations of observability criteria are provided, and probabilistic bounds on observability are derived. Furthermore, these bounds guide the design of action space to achieve a probabilistic observability guarantee in stochastic scheduling. Finally, our proposed method is applied to the estimation of slab temperature in industrial hot rolling process, and simulation results validate its effectiveness.

eess.SY↗

arXiv2024

Infinite-Horizon Optimal Wireless Control Over Shared State-Dependent Fading Channels for IIoT Systems

Heterogeneous systems consisting of a multiloop wireless control system (WCS) and a mobile agent system (MAS) are ubiquitous in Industrial Internet of Things systems. Within these systems, the positions of mobile agents may lead to shadow fading on the wireless channel that the WCS is controlled over and can significantly compromise its performance, requiring joint coordination between the WCS and MAS. Such coordination introduces different time steps and hybrid state spaces consisting of logical components and continuous components. This paper focuses on the infinite-horizon optimal control of MAS to ensure the performance of WCS while minimizing an average cost for the heterogeneous system subject to safety constraints. A state-dependent fading channel is modeled to capture interference among transmission links, as well as the effects of mobile agents' movements on successful wireless transmission. In order to address the heterogeneous system dynamics, the optimal control problem is formulated as the optimal constrained set stabilization of the MAS by establishing a necessary and sufficient condition for the Lyapunov-like performance of WCS with the expected decay rates. Using the semi-tensor product of matrices, a constrained optimal state transition graph is constructed to encode the constrained system dynamics as well as objective function, which further reduces the problem into a minimum-mean cycle problem for the graph. By studying the properties of the graph, the feasibility is proven, and an effective algorithm is proposed for the construction of optimal input sequences. An illustrative example is provided to demonstrate effectiveness of the proposed method.

eess.SY↗

arXiv2024

Power-Domain Interference Graph Estimation for Multi-hop BLE Networks

Traditional wisdom for network management allocates network resources separately for the measurement and communication tasks. Heavy measurement tasks may compete limited resources with communication tasks and significantly degrade overall network performance. It is therefore challenging for the interference graph, deemed as incurring heavy measurement overhead, to be used in practice in wireless networks. To address this challenge in wireless sensor networks, our core insight is to use power as a new dimension for interference graph estimation (IGE) such that IGE can be done simultaneously with the communication tasks using the same frequency-time resources. We propose to marry power-domain IGE with concurrent flooding to achieve simultaneous measurement and communication in BLE networks, where the power linearity prerequisite for power-domain IGE holds naturally true in concurrent flooding. With extensive experiments, we conclude the necessary conditions for the power linearity to hold and analyze several nonlinearity issues of power related to hardware imperfections. We design and implement network protocols and power control algorithms for IGE in multi-hop BLE networks and conduct experiments to show that the marriage is mutually beneficial for both IGE and concurrent flooding. Furthermore, we demonstrate the potential of IGE in improving channel map convergence and convergecast in BLE networks.

cs.NI↗

arXiv2024

GTA: A Benchmark for General Tool Agents

Significant focus has been placed on integrating large language models (LLMs) with various tools in developing general-purpose agents. This poses a challenge to LLMs' tool-use capabilities. However, there are evident gaps between existing tool-use evaluations and real-world scenarios. Current evaluations often use AI-generated queries, single-step tasks, dummy tools, and text-only interactions, failing to reveal the agents' real-world problem-solving abilities effectively. To address this, we propose GTA, a benchmark for General Tool Agents, featuring three main aspects: (i) Real user queries: human-written queries with simple real-world objectives but implicit tool-use, requiring the LLM to reason the suitable tools and plan the solution steps. (ii) Real deployed tools: an evaluation platform equipped with tools across perception, operation, logic, and creativity categories to evaluate the agents' actual task execution performance. (iii) Real multimodal inputs: authentic image files, such as spatial scenes, web page screenshots, tables, code snippets, and printed/handwritten materials, used as the query contexts to align with real-world scenarios closely. We design 229 real-world tasks and executable tool chains to evaluate mainstream LLMs. Our findings show that real-world user queries are challenging for existing LLMs, with GPT-4 completing less than 50% of the tasks and most LLMs achieving below 25%. This evaluation reveals the bottlenecks in the tool-use capabilities of current LLMs in real-world scenarios, which provides future direction for advancing general-purpose tool agents. The code and dataset are available at https://github.com/open-compass/GTA.

cs.CL↗

arXiv2024

Distributed Nonconvex Optimization: Gradient-free Iterations and $ε$-Globally Optimal Solution

Distributed optimization utilizes local computation and communication to realize a global aim of optimizing the sum of local objective functions. This article addresses a class of constrained distributed nonconvex optimization problems involving univariate objectives, aiming to achieve global optimization without requiring local evaluations of gradients at every iteration. We propose a novel algorithm named CPCA, exploiting the notion of combining Chebyshev polynomial approximation, average consensus, and polynomial optimization. The proposed algorithm is i) able to obtain $ε$-globally optimal solutions for any arbitrarily small given accuracy $ε$, ii) efficient in both zeroth-order queries (i.e., evaluations of function values) and inter-agent communication, and iii) distributed terminable when the specified precision requirement is met. The key insight is to use polynomial approximations to substitute for general local objectives, distribute these approximations via average consensus, and solve an easier approximate version of the original problem. Due to the nice analytic properties of polynomials, this approximation not only facilitates efficient global optimization, but also allows the design of gradient-free iterations to reduce cumulative costs of queries and achieve geometric convergence for solving nonconvex problems. We provide a comprehensive analysis of the accuracy and complexities of the proposed algorithm.

math.OC↗

arXiv2023

Private and Robust Distributed Nonconvex Optimization via Polynomial Approximation

There has been work that exploits polynomial approximation to solve distributed nonconvex optimization problems involving univariate objectives. This idea facilitates arbitrarily precise global optimization without requiring local evaluations of gradients at every iteration. Nonetheless, there remains a gap between existing guarantees and practical requirements, e.g., privacy preservation and robustness to network imperfections. To fill this gap and keep the above strengths, we propose a Private and Robust Chebyshev-Proxy-based distributed Optimization Algorithm (PR-CPOA). Specifically, to ensure both the accuracy of solutions and the privacy of local objectives, we design a new privacy-preserving mechanism. This mechanism leverages the randomness in blockwise insertions of perturbed vector states and hence provides a stronger privacy guarantee in the scope of ($α,β$)-data-privacy. Furthermore, to gain robustness against network imperfections, we use the push-sum consensus protocol as a backbone and discuss its specific enhancements. Thanks to the purely consensus-type iterations, we avoid the privacy-accuracy trade-off and the bother of selecting proper step sizes in different settings. We rigorously analyze the accuracy, privacy, and complexity of the proposed algorithm. We show that the advantages brought by introducing polynomial approximation are maintained when all the above requirements exist.

math.OC↗

arXiv2023

Bearing-Based Target Entrapping Control of Multiple Uncertain Agents With Arbitrary Maneuvers

This paper is concerned with bearing-based cooperative target entrapping control of multiple uncertain agents with arbitrary maneuvers including shape deformation, rotations, scalings, etc. A leader-follower structure is used, where the leaders move with the predesigned trajectories, and the followers are steered by an estimation-based control method, integrating a distance estimator using bearing measurements and a stress matrix-based formation controller. The signum functions are used to compensate for the uncertainties so that the agents' accelerations can be piecewise continuous and bounded to track the desired dynamics. With proper design of the leaders' trajectories and a geometric configuration, an affine matrix is determined so that the persistently exciting conditions of the inter-agent relative bearings can be satisfied since the bearing rates are related to different weighted combinations of the affine matrix vectors. The asymptotic convergence of the estimation error and control error is proved using Filipov properties and cascaded system theories. A sufficient condition for inter-agent collision avoidance is also proposed. Finally, simulation results are given to validate the effectiveness of the method in both 2D and 3D cases.

eess.SY↗

arXiv2023

Resilient Clock Synchronization Architecture for Industrial Time-Sensitive Networking

Time-Sensitive Networking (TSN) is a promising industrial Internet of Things technology. Clock synchronization provides unified time reference, which is critical to the deterministic communication of TSN. However, changes in internal network status and external work environments of devices both degrade practical synchronization performance. This paper proposes a temperature-resilient architecture considering delay asymmetry (TACD) to enhance the timing accuracy under the impacts of internal delay and external thermal changes. In TACD, an anti-delay-asymmetry method is developed, which employs a partial variational Bayesian algorithm to promote adaptability to non-stationary delay variation. An optimized skew estimator is further proposed, fusing the temperature skew model for ambiance perception with the traditional linear clock model to compensate for nonlinear error caused by temperature changes. Theoretical derivation of skew estimation lower bound proves the promotion of optimal accuracy after the fusion of clock models. Evaluations based on measured delay data demonstrate accuracy advantages regardless of internal or external influences.

eess.SY↗

arXiv2023

Scalable Scheduling for Industrial Time-Sensitive Networking: A Hyper-flow Graph Based Scheme

Industrial Time-Sensitive Networking (TSN) provides deterministic mechanisms for real-time and reliable flow transmission. Increasing attention has been paid to efficient scheduling for time-sensitive flows with stringent requirements such as ultra-low latency and jitter. In TSN, the fine-grained traffic shaping protocol, cyclic queuing and forwarding (CQF), eliminates uncertain delay and frame loss by cyclic traffic forwarding and queuing. However, it inevitably causes high scheduling complexity. Moreover, complexity is quite sensitive to flow attributes and network scale. The problem stems in part from the lack of an attribute mining mechanism in existing frame-based scheduling. For time-critical industrial networks with large-scale complex flows, a so-called hyper-flow graph based scheduling scheme is proposed to improve the scheduling scalability in terms of schedulability, scheduling efficiency and latency & jitter. The hyper-flow graph is built by aggregating similar flow sets as hyper-flow nodes and designing a hierarchical scheduling framework. The flow attribute-sensitive scheduling information is embedded into the condensed maximal cliques, and reverse maps them precisely to congestion flow portions for re-scheduling. Its parallel scheduling reduces network scale induced complexity. Further, this scheme is designed in its entirety as a comprehensive scheduling algorithm GH^2. It improves the three criteria of scalability along a Pareto front. Extensive simulation studies demonstrate its superiority. Notably, GH^2 is verified its scheduling stability with a runtime of less than 100 ms for 1000 flows and near 1/430 of the SOTA FITS method for 2000 flows.

cs.NI↗

arXiv2023

Hydrogen Supply Infrastructure Network Planning Approach towards Chicken-egg Conundrum

In the early commercialization stage of hydrogen fuel cell vehicles (HFCVs), reasonable hydrogen supply infrastructure (HSI) planning decisions is a premise for promoting the popularization of HFCVs. However, there is a strong causality between HFCVs and hydrogen refueling stations (HRSs): the planning decisions of HRSs could affect the hydrogen refueling demand of HFCVs, and the growth of demand would in turn stimulate the further investment in HRSs, which is also known as the ``chicken and egg'' conundrum. Meanwhile, the hydrogen demand is uncertain with insufficient prior knowledge, and thus there is a decision-dependent uncertainty (DDU) in the planning issue. This poses great challenges to solving the optimization problem. To this end, this work establishes a multi-network HSI planning model coordinating hydrogen, power, and transportation networks. Then, to reflect the causal relationship between HFCVs and HRSs effectively without sufficient historical data, a distributionally robust optimization framework with decision-dependent uncertainty is developed. The uncertainty of hydrogen demand is modeled as a Wasserstein ambiguity set with a decision-dependent empirical probability distribution. Subsequently, to reduce the computational complexity caused by the introduction of a large number of scenarios and high-dimensional nonlinear constraints, we developed an improved distribution shaping method and techniques of scenario and variable reduction to derive the solvable form with less computing burden. Finally, the simulation results demonstrate that this method can reduce costs by at least 10.4% compared with traditional methods and will be more effective in large-scale HSI planning issues. Further, we put forward effective suggestions for the policymakers and investors to formulate relevant policies and decisions.

eess.SY↗