arXiv subjects

Hao Kang

Publications and source records attributed to Hao Kang.

At least 19 recordsLinked to original sources

arXiv2026

Embedding a de Rham Complex Structure into Eig-PIELM for Maxwell Eigenvalue Problems

The perfect electric conductor (PEC) boundary condition alone does not eliminate the gradient nullspace of the Maxwell curl-curl operator. Consequently, a boundary-only Maxwell implementation of Eig-PIELM, an eigenvalue extension of the physics-informed extreme learning machine (PIELM), may retain represented gradient modes in its reduced trial space, producing nonphysical, gradient-contaminated modes. We develop dR-Eig-PIELM, a finite-dimensional structure-preserving extension that retains the linear-coefficient, boundary-null-space, and one-shot multimode architecture of Eig-PIELM. It combines compatible tensor-product grad--curl trial spaces with a constrained null-space construction to restrict the sampled PEC space to the mass-orthogonal complement of the represented gradient image before the eigensolve. Symmetric weak-form Gram assembly yields a symmetric-definite reduced eigenproblem without an eigenvalue cutoff under the stated sampling and material-positivity assumptions. For rectangles and cuboids, we establish fixed-degree exactness at the PEC vector-field space and derive rank and dimension identities. Numerical experiments show that dR-Eig-PIELM excludes the spurious eigenmodes with significant gradient components observed in the boundary-only baseline. The method exhibits rapid convergence with increasing polynomial degree and competitive work--precision performance in the tested configurations. A covariantly mapped waveguide example illustrates the potential of dR-Eig-PIELM for integration into design optimization workflows through a constrained local geometry update driven by computed cutoff frequencies and their sensitivities. This work provides a mathematical and computational framework for separating represented gradient modes from physical Maxwell eigenmodes within Eig-PIELM and supports its use in waveguide spectral analysis and geometry design.

math.NA↗

arXiv2026

A Clustered Equivalent-Source Method with Local Response Compression and Group-Symmetry Reduction for Antenna-Array Analysis

Full-wave analysis of finite antenna arrays is computationally demanding because detailed element responses must be resolved and coupled across the entire array. To reduce this cost, this paper presents a clustered equivalent-source method that combines local response compression with group-symmetry reduction. Local responses are represented in compressed equivalent-surface bases or coordinates inherited from child clusters. Symmetry constraints associated with the cluster partition and excitation reduce the coupled system, while the corresponding group transformations recover the complete response in the retained coordinates. To assess overall efficiency, we evaluate the costs of local construction, propagation, global solution, and recovery. Numerical examples involving layered and finite annular patch arrays show close agreement between the computed and reference complex far fields and shorter complete-run times than those of the corresponding symmetry-reduced direct solvers. Among the three equivalent-source implementations for the layered array, the one with the smallest coupled system has the longest runtime because local construction and propagation dominate the cost. These findings highlight the need to consider local construction and coupling costs when assessing the benefit of reducing the global system dimension.

math.NA↗

arXiv2026

Jet-QSparse: Accelerating Agentic LLM Inference with 4-Bit Sparse Attention

Long-horizon reasoning and agentic workloads strain KV-cache capacity and decode attention throughput. We present Jet-QSparse, a training-free method that combines compact NVFP4 cache storage with native FP4 query-key computation on Blackwell tensor cores. A dense scan scores the full key history and selects 128-token blocks for sparse FP16 value aggregation with FP32 accumulation, preserving higher precision in the output-forming step while enabling contiguous tile loads. Calibrated key centering and at most two BF16 pages per sequence and layer address quantization error; FP8 prefill uses a temporary arena while the persistent cache is packed directly from BF16 activations. Across four hybrid LLMs, a budget of 128 blocks per KV head shows no detected accuracy loss on the evaluated mathematical reasoning, coding, and agentic benchmarks, although some smaller budgets incur losses. With 64 blocks, Qwen3.8-27B achieves a $1.94\times$ end-to-end decode speedup over dense BF16 at 128K context and batch size 16 on one GB300, and the measured Qwen KV pools provide $2.58$-$2.76\times$ BF16's token capacity.

cs.LG↗

arXiv2026

Delta-Matching: Closing the Final Gap of Native 8-bit Training for LLMs

Reliable FP8 attention remains a barrier to fully native 8-bit large language model training. We derive how forward-backward inconsistencies produce stale delta and empirically show how it distorts training dynamics. Our stale-delta hybrid runs show a modest loss gap at 569M parameters but substantial loss increases and downstream degradation at 1.67B and 5.29B. QK normalization, NoPE (no positional encoding), and lower-learning-rate context extension mitigate or delay degradation without eliminating it. This pattern suggests accumulated optimization error that smaller models and short runs can conceal. We propose Delta-Matching, proving that it restores the softmax gradient's zero-row-sum invariant under the stated numerical assumptions. It enables native block-scaled FP8 in every forward and backward attention-core matmul without architectural changes, smaller global batches, or auxiliary forward outputs. Across tested architectures, scales, and training stages, Delta-Matching matches BF16/FP32 mixed-precision training loss and overall downstream performance. We will release our implementation, trained models, and data recipes (https://jet-ai-projects.github.io/delta-matching).

cs.LG↗

arXiv2026

Routing Policy Optimization for Online Diffusion Reinforcement Learning

Reinforcement learning can align visual generators with reward signals beyond their pretraining objectives, but existing score-function methods typically introduce stochasticity by perturbing the reverse denoising trajectory, leading to costly rollout-based optimization. We instead ask whether stochastic exploration can be introduced inside the generator while preserving its original sampler. Routing Policy Optimization (RPO) answers this question by turning attention routing into a stochastic exploration action. At a few selected denoising levels, each query block keeps its own block and samples further key blocks in proportion to the generator's native attention mass, or keeps dense attention. The sampled route perturbs the velocity of that transition, and a Gaussian likelihood around the dense velocity scores the perturbation with a score-function objective. RPO further uses a controlled two-rollout estimator: two rollouts share the same prompt, initial latent, routed denoising levels, and dense anchor layers, differing only in their sampled attention routes. Their reward difference provides an advantage signal for the routed transitions. Because the likelihood differentiates the dense velocity itself, the learned updates remain effective after routing is removed. At inference, RPO restores dense attention and retains the original deterministic sampler. Experiments on text-to-image and text-to-video generation show that RPO improves generation quality while achieving nearly 4x higher training efficiency than Flow-GRPO-style methods. Please see our project page at https://velox-world.com/rpo for visual results.

cs.CV↗

arXiv2026

Long-time behavior of reducible stochastic linear reaction networks: a spectral and structural classification

Exponential ergodicity of stochastic reaction networks has attracted considerable attention in recent years [SIAM J. Appl. Dyn. Syst. 24, 1668-1710 (2025)]. Here, we provide a structural classification of the long-time behavior of reducible stochastic linear reaction networks under the $L^1$-Wasserstein and total variation distances. The classification is determined by the maximal eigenvalue $λ_{\max}$ of the first-order influx matrix $A$, the position of the zero-order influx vector $b$ relative to the left nullspace of $A$, and the conservation-law structure of the network restricted to the persistent species. We first prove that every stochastic linear reaction network is non-explosive. In the stable regime $λ_{\max}<0$, the process converges exponentially fast to a unique stationary distribution. In the critical regime $λ_{\max}=0$, when the zero eigenvalue of $A$ is semisimple and $b$ is orthogonal to the left nullspace of $A$, exponential convergence occurs if and only if a regularity condition on the restricted network is satisfied. If these two spectral conditions hold but the regularity condition fails, convergence occurs only in total variation and is non-exponential. In the divergent regimes, no closed irreducible positive recurrent class can contain an interior state, and any stationary distribution, if it exists, must be supported on the boundary. Moreover, by employing a coupling method, we obtain the optimal convergence rate in the exponentially convergent cases.

math.PR↗

arXiv2026

ScribbleEdit: A Benchmark for Scribble-Only Image Editing

Scribble-based interaction provides a lightweight and intuitive way for users to specify image editing intents in interactive editing tools. However, current image editing models based on VLMs or LLMs struggle to understand and execute edits based solely on scribble inputs. To systematically study this problem, we construct a new benchmark, ScribbleEdit, that evaluates the ability of image editing models to perform image editing conditioned on scribbles. This task requires both a deep understanding of the intention of the scribble and an accurate interpretation of its spatial information. In ScribbleEdit, we design an automated data construction pipeline and introduce a dedicated evaluation protocol that explicitly measures intention alignment. Our analysis reveals that existing VLM/LLM-based editing models fail to accurately capture scribble intentions. To guide future progress on scribble-only image editing, we propose a simple yet effective soft-token baseline, which enhances the model's understanding of scribble semantics and outperforms standard image editing models on our benchmark. Our evaluation and baseline together provide a concrete foundation for assessing and improving the scribble-driven image editing.

cs.GR↗

arXiv2026

Evaluating Test-Time Scaling of General LLM Agents

LLM agents are increasingly expected to operate as general-purpose systems that resolve real-world user requests, yet their dynamic scaling behavior in realistic environments remains poorly understood. In this paper, we systematically investigate two principal test-time scaling axes of LLM agents: sequential scaling through extended interaction and parallel scaling through trajectory sampling. We first introduce a realistic benchmark that provides one unified framework for evaluating LLM agents across search, coding, reasoning, and tool-use domains, more faithfully reflecting the heterogeneity of real-world deployments. Evaluating ten leading LLM agents reveals substantial performance degradation when transitioning from domain-specific evaluations to this realistic setting. Building on this foundation, we progressively scale test-time compute along fine-grained increments to characterize the performance upper bound. We find that neither scaling axis can consistently yield meaningful gains from additional test-time compute in realistic environments, a phenomenon we attribute to two fundamental limitations: the scaling plateau that bottlenecks sequential scaling and the verification gap that undermines parallel scaling. Code is publicly available at https://github.com/cxcscmu/General-AgentBench.

cs.AI↗

arXiv2026

Bistable wavefronts in the Gurtin-MacCamy population model

In this paper we prove the existence of monotone traveling wave solutions for an age-structured population model with a monotone and bistable nonlinearity. We formulate the problem in terms of a discrete-time recurrence $u_{n+1} = Q[u_n]$, with operator $Q$ encoding the role of the speed into its definition. For each speed, we prove the existence of a traveling wave by the general theory of monotone evolution systems, and from appropriate estimates on the speed, we deduce the existence of a traveling wave of the original problem. Also, the sign of the speed is established under an invasion condition, and the asymptotic behavior of the profile is determined by some Tauberian results.

math.AP↗

arXiv2026

ThunderAgent: A Simple, Fast and Program-Aware Agentic Inference System

Large language models(LLMs) are now used to power complex multi-turn agentic workflows. Existing systems run agentic inference by loosely assembling isolated components: an LLM inference engine (e.g., vLLM) and a tool orchestrator (e.g., Kubernetes). Although agentic workflows involve multiple LLM and tool requests, these systems schedule and allocate resources separately on a per-request basis, without end-to-end knowledge of the workflow. This leads to sub-optimal management of KV cache and tool execution environments. To address the challenges, we propose ThunderAgent, a fast, simple, and program-aware agentic inference system. We first abstract agentic workflows as LLM Programs, enabling a unified view of heterogeneous resources, including KV caches, system states, and external tool assets such as disk memory and network ports. Built upon this abstraction, ThunderAgent introduces a program-aware scheduler and a tool resource manager designed to maximize KV cache hit rates, mitigate memory imbalances, and enable asynchronous environment preparation. Evaluations across coding, routing, and scientific discovery agents demonstrate that ThunderAgent achieves 1.5-3.6x throughput improvements in serving, 1.8-3.9x in RL rollout, and up to 4.2x disk memory savings compared to state-of-the-art inference systems. To facilitate reproducibility and support future development, we open-source the system implementations of the whole ThunderAgent at: https://github.com/Agentic-Kinetics/ThunderAgent.

cs.OS↗

arXiv2026

ActWorld: From Explorable to Interactive World Model via Action-Aware Memory

Interactive world models aim to simulate environment dynamics under real-time user actions. However, their action vocabulary is largely confined to navigation: most actions correspond to motion (e.g., walk, turn, look around), while interaction with objects in the scene (e.g., pick up plates, open doors, or trigger physical responses) is either absent, restricted to game domains, or relegated to prompt-to-full-video scenarios. The resulting worlds are visually explorable but not truly actionable. In this work, we present ActWorld, an interactive world model that extends prior navigation-centric generators to support mid-rollout object interaction within a chunk-autoregressive framework. We argue that the navigation-interaction gap stems from two bottlenecks. First, a data bottleneck: the lack of human-object interaction data with accurate, dense labels. Second, a memory bottleneck: recency-biased history compression in existing world models discards the event-transition frames that causally determine subsequent object states, leading to an action-forgetting pathology. On the data side, we construct a 100K interaction video dataset, each annotated with per-chunk captions via chain-of-thought reasoning. On the model side, we introduce a hierarchical action-aware memory design that routes history compression by interaction importance, complemented by a persistent memory bank that maintains event-update and object-identity tokens across long rollouts. Experiments show that ActWorld supports both flexible navigation and rich object interaction within a single model, substantially improving interaction fidelity over navigation-only baselines without sacrificing viewpoint control. Project page is available at https://interactwm.github.io/ActWorld.

cs.CV↗

arXiv2026

OckBench: Measuring the Efficiency of LLM Reasoning

Large language models (LLMs) such as GPT-5 and Gemini 3 have pushed the frontier of automated reasoning and code generation. Yet current benchmarks emphasize accuracy and output quality, neglecting a critical dimension: efficiency of token usage. The token efficiency is highly variable in practical. Models solving the same problem with similar accuracy can exhibit up to a \textbf{5.0$\times$} difference in token length, leading to massive gap of model reasoning ability. Such variance exposes significant redundancy, highlighting the critical need for a standardized benchmark to quantify the gap of token efficiency. Thus, we introduce OckBench, the first benchmark that jointly measures accuracy and token efficiency across reasoning and coding tasks. Our evaluation reveals that token efficiency remains largely unoptimized across current models, significantly inflating serving costs and latency. These findings provide a concrete roadmap for the community to optimize the latent reasoning ability, token efficiency. Ultimately, we argue for an evaluation paradigm shift: tokens must not be multiplied beyond necessity. Our benchmarks are available at https://ockbench.github.io/.

cs.CL↗

arXiv2026

Idleness is Relative: Exploiting Tool-Call Idle Windows for Offloading in Agentic Systems with MORI

Modern LLM serving systems increasingly host agentic workloads, whose sessions issue tens of model invocations interleaved with tool calls, accumulating KV cache that can be reused across steps. As requests' total KV cache size easily exceeds GPU HBM capacity, researchers offload them to CPU DRAM. However, tool-call durations span orders of magnitude, and the cost of transferring KV cache between tiers makes it impractical to re-place entries on every call. We observe that agentic programs exhibit a two-phase structure: busy phases of rapid short tool calls and idle phases dominated by long-running calls. Current eviction policies such as LRU fail to capture this property. A binary busy/idle label also falls short because the ratio of busy to idle programs may not match the hardware's GPU-to-CPU capacity ratio. When it does not, one tier sits underutilized while the other is oversubscribed, wasting memory or forcing unnecessary evictions. We present MORI, an agent serving system that solves the above problem. Our key insight is that idleness is a continuous, relative spectrum. MORI ranks all active programs by idleness, assigns the busiest to GPU HBM and the most idle to CPU DRAM, dynamically shifts the partition boundary to match hardware capacity, and enforces admission control at each memory tier. Evaluated on real coding agent workloads collected from Claude Code across four GPU and model pairs, MORI delivers 20--71% higher throughput and 18--43% lower TTFT than the best baseline with offloading.

cs.OS↗

arXiv2026

PithTrain: A Compact and Agent-Native MoE Training System

Mixture-of-Experts (MoE) has become the dominant architecture for frontier language models. To meet this demand, production frameworks have built optimized MoE training stacks over years of engineering effort. Yet evolving these stacks for new architectures and system optimizations remains expensive. With the rise of AI coding agents, they could automate parts of training-framework development and accelerate this evolution. But applying them to these existing frameworks carries hidden costs, invisible to today's throughput-only evaluations. We name this missing dimension agent-task efficiency (ATE): the cost of using coding agents to understand, operate, and extend a framework. Grounded in four agent-native design principles, we build PithTrain, a compact, agent-native MoE training framework. We further introduce ATE-Bench, covering real-world training-framework tasks. Our evaluation shows PithTrain matches the throughput of production frameworks, and on ATE-Bench, PithTrain enables higher agent-task efficiency, with up to 62% fewer Agent Turns and 64% less Active GPU Time.

cs.LG↗

arXiv2026

Auto Research with Specialist Agents Develops Effective and Non-Trivial Training Recipes

We study auto research as a closed empirical loop driven by external measurement. Each submitted trial carries a hypothesis, an executable code edit, an evaluator-owned outcome, and feedback that shapes the next proposal. The output is not a generated paper or a single model checkpoint, but an auditable trajectory of proposals, code diffs, experiments, scores, and failure labels. We instantiate this loop with specialist agents that partition recipe surfaces and share measured lineage across trials. The central empirical finding is that lineage feedback lets agents turn evaluator outcomes, including crashes, budget overruns, size failures, and accuracy-gate misses, into later program-level recipe edits rather than one-shot suggestions. Across 1,197 headline-run trials plus 600 Parameter Golf control trials after one-time setup and launch, humans did not choose proposals, edit recipes, override scores, or repair failed trials during the search. In the three headline runs, the same submitted-trial loop reduces Parameter Golf validation bpb by $0.81\%$, raises NanoChat-D12 CORE by $38.7\%$, and reduces CIFAR-10 Airbench96 wallclock by $4.59\%$, with each task measured by its own external evaluator and legality checks. The trace includes a strict architecture-domain audit of 157 headline-run submissions and program rewrites such as a NanoChat attention-kernel path change. Within this scope the loop autonomously writes code, submits experiments, absorbs feedback, applies and combines known techniques inside each environment, and improves public starting recipes.

cs.MA↗

arXiv2026

InstructMoLE: Instruction-Guided Mixture of Low-rank Experts for Multi-Conditional Image Generation

Parameter-Efficient Fine-Tuning of Diffusion Transformers (DiTs) for diverse, multi-conditional tasks often suffers from task interference when using monolithic adapters like LoRA. The Mixture of Low-rank Experts (MoLE) architecture offers a modular solution, but its potential is usually limited by routing policies that operate at a token level. Such local routing can conflict with the global nature of user instructions, leading to artifacts like spatial fragmentation and semantic drift in complex image generation tasks. To address these limitations, we introduce InstructMoLE, a novel framework that employs an Instruction-Guided Mixture of Low-Rank Experts. Instead of per-token routing, InstructMoLE utilizes a global routing signal, Instruction-Guided Routing (IGR), derived from the user's comprehensive instruction. This ensures that a single, coherently chosen expert council is applied uniformly across all input tokens, preserving the global semantics and structural integrity of the generation process. To complement this, we introduce an output-space orthogonality loss, which promotes expert functional diversity and mitigates representational collapse. Extensive experiments demonstrate that InstructMoLE significantly outperforms existing LoRA adapters and MoLE variants across challenging multi-conditional generation benchmarks. Our work presents a robust and generalizable framework for instruction-driven fine-tuning of generative models, enabling superior compositional control and fidelity to user intent.

cs.CV↗

arXiv2026

Privatar: Scalable Privacy-preserving Multi-user VR via Secure Offloading

Multi-user virtual reality enables immersive interaction. However, rendering avatars for numerous participants on each headset incurs prohibitive computational overhead, limiting scalability. We introduce a framework, Privatar, to offload avatar reconstruction from headset to untrusted devices within the same local network while safeguarding attacks against adversaries capable of intercepting offloaded data. Privatar's key insight is that domain-specific knowledge of avatar reconstruction enables provably private offloading at minimal cost. (1) System level. We observe avatar reconstruction is frequency-domain decomposable via BDCT with negligible quality drop, and propose Horizontal Partitioning (HP) to keep high-energy frequency components on-device and offloads only low-energy components. HP offloads local computation while reducing information leakage to low-energy subsets only. (2) Privacy level. For individually offloaded, multi-dimensional signals without aggregation, worst-case local Differential Privacy requires prohibitive noise, ruining utility. We observe users' expression statistical distribution are slowly changing over time and trackable online, and hence propose Distribution-Aware Minimal Perturbation. DAMP minimizes noise based on each user's expression distribution to significantly reduce its effects on utility, retaining formal privacy guarantee. Combined, HP provides empirical privacy against expression identification attacks. DAMP further augments it to offer a formal guarantee against arbitrary adversaries. On a Meta Quest Pro, Privatar supports 2.37x more concurrent users at 6.5% higher reconstruction loss and 9% energy overhead, providing a better throughout-loss Pareto frontier over quantization, sparsity and local construction baselines. Privatar provides both provable privacy guarantee and stays robust against both empirical and NN-based attacks.

cs.CR↗

arXiv2026

NTIRE 2026 Challenge on Short-form UGC Video Restoration in the Wild with Generative Models: Datasets, Methods and Results

This paper presents an overview of the NTIRE 2026 Challenge on Short-form UGC Video Restoration in the Wild with Generative Models. This challenge utilizes a new short-form UGC (S-UGC) video restoration benchmark, termed KwaiVIR, which is contributed by USTC and Kuaishou Technology. It contains both synthetically distorted videos and real-world short-form UGC videos in the wild. For this edition, the released data include 200 synthetic training videos, 48 wild training videos, 11 validation videos, and 20 testing videos. The primary goal of this challenge is to establish a strong and practical benchmark for restoring short-form UGC videos under complex real-world degradations, especially in the emerging paradigm of generative-model-based S-UGC video restoration. This challenge has two tracks: (i) the primary track is a subjective track, where the evaluation is based on a user study; (ii) the second track is an objective track. These two tracks enable a comprehensive assessment of restoration quality. In total, 95 teams have registered for this competition. And 12 teams submitted valid final solutions and fact sheets for the testing phase. The submitted methods achieved strong performance on the KwaiVIR benchmark, demonstrating encouraging progress in short-form UGC video restoration in the wild.

cs.CV↗