Searcharxiv⌕ Search

arXiv subjects

Jun Yu

Publications and source records attributed to Jun Yu.

At least 19 recordsLinked to original sources
arXiv2026

Radical subgroups of finite reductive groups

Radical subgroups play an important role in both finite group theory and representation theory. This is the first of a series of papers of ours in classifying radical $p$-subgroups of finite reductive groups and in verifying the inductive blockwise Alperin weight condition for them, contributing to the program of proving the Alperin weight conjecture by verifying its inductive condition for finite simple groups. In this paper we present a uniform method for classifying radical subgroups of finite reductive groups. As applications, we investigate the inductive blockwise Alperin weight condition for classical groups, as well as groups of type $F_4$.

math.RT↗
arXiv2026

Searching with Sparse Histories: Cross-Task Architecture Search from Incomplete Search Logs and Zero-Cost Proxies

Neural architecture search (NAS) is expensive, but cross-task transfer can reduce costs by reusing knowledge from previous tasks. However, existing methods struggle with sparse search histories and task-specific transfer failures. We propose a method, a risk-robust strategy that combines sparse source logs with zero-cost proxies through champion-based initialization and joint pairwise-ranker pretraining. On TransNAS-Bench-101, the proposal mitigates severe worst-case failures, reducing worst-10% ranks from 290 to 43 for Micro autoencoding and from 465 to 45 for Macro room layout. Across 13 cell-space targets, it outperforms a proxy-only ranker on 10 targets in both mean and worst-case performance, while maintaining competitive average performance.

cs.LG↗
arXiv2026

Testing for an Explosive Bubble using High-Frequency Volatility

Based on a continuous-time stochastic volatility model with a linear drift, we develop a test for explosive behavior in financial asset prices at a low frequency when prices are sampled at a higher frequency. The test exploits the volatility information in the high-frequency data. The method consists of devolatizing log-asset price increments with realized volatility measures and performing a supremum-type recursive Dickey-Fuller test on the devolatized sample. The proposed test has a nuisance-parameter-free asymptotic distribution and is easy to implement. We study the size and power properties of the test in Monte Carlo simulations. A real-time date-stamping strategy based on the devolatized sample is proposed for the origination and conclusion dates of the explosive regime. Conditions under which the real-time date-stamping strategy is consistent are established. The test and the date-stamping strategy are applied to study explosive behavior in cryptocurrency and stock markets.

econ.EM↗
arXiv2026

Regularity and the Gelfand Property for Complex Symmetric Pairs

We prove that every symmetric pair of a connected complex reductive group is regular in the sense of Aizenbud--Gourevitch. This settles the Aizenbud--Gourevitch regularity conjecture over the complex numbers. Generalized Harish--Chandra descent then makes the canonical central cover of every complex symmetric pair a Gelfand--Kazhdan pair. An anti-automorphism arising from a compatible Chevalley involution upgrades the resulting GP2 bound to GP1 on the cover, and finite central descent transfers GP1 to the original pair. In particular, van Dijk's conjecture on complex symmetric pairs follows. Rubio reduced the unresolved irreducible regularity problem to four families: the DIII family $(D_r,A_{r-1}+\mathbb{C})$, the balanced CII family $(C_{2r},C_r+C_r)$, some remaining Spin block pairs, and the EVII pair $(E_7,E_6+\mathbb{C})$. We treat these cases by four different mechanisms. For DIII we construct a sign-equivariant Schwartz distribution on the regular set and extend it across a common orbit boundary by the Chen--Sun theorem. For balanced CII we combine homogeneity, distinguished nilpotent orbits, and a stable-density theorem for the centralizer representation. For Spin blocks we prove pleasantness for unequal odd--odd blocks, use Przebinda's orthogonal-distribution theorem in odd smaller rank, and construct a finite orbit closure with automatic extension in even smaller rank. For EVII we compute the graded-$\mathfrak{sl}_2$ data for all twenty-two nilpotent orbits and use central-torus characters to eliminate the remaining resonances, including the two residual triple-centralizer cases. A finite-component assembly theorem then handles arbitrary connected central quotients and diagonal couplings among simple factors.

math.RT↗
arXiv2026

Robust Vision-Language Navigation with Awareness of Instruction-Execution Progress

In Vision-Language Navigation (VLN) tasks involving multi-step instructions as guidance, an agent must associate its observations with the relevant instruction content at each decision point. Most graph-based methods commonly match candidate trajectories with the entire instruction, allowing already executed content to interfere with subsequent decisions. Although instruction-execution progress provides a direct cue for separating executed and pending content, existing progress estimation approaches often rely on explicit progress supervision, sophisticated progress-learning stages, or costly online reasoning with LLMs. In this paper, we propose a robust VLN framework with Awareness of Instruction-Execution Progress (VLN-AIEP), which couples progress estimation with navigation decision-making. VLN-AIEP associates the visited trajectory with the instruction to suppress embeddings corresponding to executed content, and then matches candidate locations with the pending content. This coupling makes correct navigation decisions depend on accurate progress estimation, allowing the progress estimator to learn directly from navigation decision labels without explicit progress annotations. To support progress estimation, we construct a landmark-anchored trajectory representation that encodes the semantic labels and viewpoint-relative geometry of instruction-referred landmarks. These landmarks serve as spatial references for associating the visited trajectory with executed instruction content. Extensive experiments on R2R, RxR, CVDN, and R2R-CE demonstrate that the proposed method achieves favorable performance across multiple metrics.

cs.CV↗
arXiv2026

Lifelong small-object navigation in changing object layouts: a benchmark and method

Household robots need to continually navigate to different objects in the same environment, many of which are small and portable, such as tools and toys. Their small visual footprint and frequent occlusion make reliable observation difficult, and they may be moved by people without the robot observing the changes. We formulate this challenging task as Lifelong Small-object Navigation in Changing Object Layouts (LiSoNav-COL). Agents must seek suitable viewpoints for reliable observation, accumulate and reuse scene knowledge to efficiently locate subsequent targets, and update outdated memory after object relocation. To eliminate the need for prior scene scanning, we also require agents to start navigation with empty scene memory. Although practical, this task still lacks benchmarks designed around its defining assumptions. To bridge this gap, we introduce LiSoNav-Eval, a dedicated benchmark spanning 28 indoor scenes with 45 small-object categories. Its lifelong navigation sequences include both unchanged and relocated targets to evaluate memory reuse and adaptation to object relocation. To address this challenging task, we propose a navigation method based on multi-view Inspection with Viewpoint-Anchored Memory, dubbed IVAM-Nav. IVAM-Nav actively observes supporting surfaces from complementary viewpoints for reliable small-object perception and anchors the resulting memory to their observation viewpoints, supporting relational memory reuse and revalidation under similar viewing conditions. Extensive experiments on LiSoNav-Eval demonstrate favorable performance of IVAM-Nav against representative methods. Benchmark analyses also show that smaller objects, larger environments, and longer relocation distances pose greater challenges. The dataset and code are available here.

cs.RO↗
arXiv2026

Design-Driven Inference for Online Experiments: Jointly Optimal Evidence Collection, Testing, and Estimation

Online multifactorial experiments requires early stopping for practically negligible effects while controlling false stopping under continuous monitoring and retaining precise treatment-effect estimation. We develop a design-driven framework that chooses allocation and the evidence rule jointly, treating allocation as evidence collection and testing and estimation as complementary uses of the same evidence. For multifactor experiments with nuisance block effects and treatment-by-block interactions, we show that block-orthogonal designs remove nuisance contamination from the treatment score and form a complete class for worst-case evidence growth. Under a fixed information budget, an isotropic allocation and an explicit radial e-value jointly attain the minimax rate optimal for directionally unknown alternatives. The same allocation is universally optimal for maximum likelihood estimation of treatment effects, simultaneously achieving A-, D-, and E-optimality and establishing double optimality for testing and estimation. Batchwise replication yields an anytime-valid e-process, retains minimax rate optimality for cumulative conditional worst-case evidence growth, and preserves universal estimation optimality for the prespecified complete experiment. Simulations and a large language model prompt experiment illustrate gains in stopping efficiency and estimation precision.

stat.ME↗
arXiv2026

Can Agent Harnesses and Inference Engines Hear Each Other? The HEAR Protocol for Agentic LLM Serving

LLM agents increasingly execute complex workflows involving multi-turn reasoning, tool use, and parallel agents. Efficient serving requires decisions that span two layers with complementary information: the agent harness understands workflow dependencies, context lifecycles, and execution objectives, whereas the inference engine observes request queues, KV-cache state, resource pressure, and execution capabilities. Existing interfaces do not systematically connect these views, limiting workflow-aware execution. HEAR, a bidirectional Harness--Engine Pairing protocol for agentic LLM serving. HEAR standardizes how the harness communicates workflow intent and execution requirements and how the engine returns runtime state, capabilities, and outcomes. By separating protocol semantics from optimization policies, HEAR supports diverse coordination strategies without changing workflow or model semantics. We instantiate HEAR for online cache-aware runtime coordination and workload-aware execution-mode selection for agent roles. Across four conversational and research-agent benchmarks under memory-constrained, concurrent serving, HEAR achieves a $1.61\times$ batch speedup and reduces median time-to-first-token by $2.23\times$ on SCBench. Mooncake shows that workflow intent and live engine state provide complementary benefits across load regimes. On BrowseComp-Plus and DeepResearchBench, workload-specific configurations yield $1.23\times$ and $2.45\times$ end-to-end speedups, respectively, without observed task-quality degradation. These results establish HEAR as a reusable coordination substrate for efficient agentic LLM serving.

cs.AI↗
arXiv2026

Beyond Refusal Patterns: Safe-Role Internalization for Robust and Generalizable LLM Safety Alignment

Large Language Models (LLMs) have achieved remarkable capabilities but remain vulnerable to jailbreak attacks that elicit harmful or unsafe outputs. Existing safety alignment approaches, including Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), often require substantial attack-specific supervision and computational resources, while remaining susceptible to shallow safety alignment and over-refusal. To address these challenges, we introduce SSRFT(Supervised Safe-Role Fine-Tuning), the first framework that reformulates safety alignment as the internalization of a predefined safe role. SSRFT constructs a Safe-Role Question-Answer (SRQA) dataset from psychometric questions, limited jailbreak prompts, and a safe-role description. Role-consistent responses are synthesized, validated, and expanded into diverse scenarios, enabling models to internalize safety-oriented values and principles rather than explicit refusal patterns. Experiments across multiple Base and Instruct models show that SSRFT achieves more robust and generalizable safety alignment than standard SFT. SSRFT shows substantially greater robustness to prefilling attacks and better generalization to unseen jailbreak domains, while reducing over-refusal on benign queries and preserving the model's general capabilities. These results establish safe-role internalization as an effective alternative to refusal-centric safety alignment. Warning: This paper contains examples of harmful and toxic language.

cs.AI↗
arXiv2026

ResidualKV: Residual-Based KV Cache Compression for Efficient Long-Context Inference

Efficient long-context inference faces two coupled bottlenecks: KV-cache memory grows linearly with context length, while attention computation grows quadratically. Existing approaches typically address one at the expense of irreversible token eviction, full-cache retention, or full-history reconstruction, limiting their effectiveness for multi-turn interaction and long-form reasoning. Motivated by two empirical properties, Long-Range Inter-Token Similarity and Smooth Residual Distribution, we propose ResidualKV, which factorizes the KV cache into a sparse set of globally retrieved references and compact, quantized residual codes for the remaining tokens. This representation preserves token-specific information without permanent eviction and, when combined with sparse attention, reconstructs only the selected states on demand. Dynamic-stride scheduling further reduces reference growth from linear to approximately logarithmic at ultra-long contexts. Across Llama, Qwen, LLaVA-OV, and Qwen3-VL backbones, ResidualKV maintains near-full-cache performance using only 13%-16% KV storage and 30% attention computation on LongBench, and 8%-10% storage and 10% computation in matched-budget multimodal evaluation. It also accelerates decoding by up to $1.5\times$ with KV-cache quantization and $3.4\times$ without it. These results show that global cross-token redundancy supports accurate, memory-efficient, and computation-efficient long-context inference. The source code is available at https://github.com/CURRENTF/ResidualKV.

cs.CL↗
arXiv2026

ShallowStream: Index Shallow then Answer Deep for Streaming Video Understanding

Streaming video understanding is a critical capability for real-world applications, including embodied intelligence, autonomous driving, industrial monitoring, surveillance and early warning, and wearable assistants. However, processing continuous video streams with multimodal large language models (MLLMs) is computationally expensive. Existing efforts have explored reducing streaming overhead through visual token pruning, token merging, quantization, on-demand frame retrieval, and context offloading. However, most existing methods overlook the dimension of model depth. Repeatedly executing full-depth MLLM prefill over incoming frames is prohibitively expensive, incurring substantial computational overhead and causing the KV cache to grow at a rate directly proportional to the prefill depth. To address these challenges, we propose ShallowStream, a novel framework that leverages the shallow layers of an MLLM to simultaneously perform frame encoding and retrieval index building. During stream processing, ShallowStream maintains an always-on lightweight index using the KV cache of shallow layers. During query-time answering, we leverage the attention scores generated by the shallow layers to score context frames and employ a diversity-aware selection strategy to retrieve precise and comprehensive evidence. ShallowStream achieves performance on par with the strongest existing streaming methods, while reducing per-frame prefill latency and 10-second end-to-end latency by up to 52.1x and 11.9x, respectively. Our code is available at https://github.com/CURRENTF/ShallowStream.

cs.CV↗
arXiv2026

SparseEngine: Sparse-First Inference Engine

Long-context LLM agents accumulate interaction histories that strain KV-cache memory and attention computation. Although sparse attention reduces these costs, heterogeneous cache representations and workflows hinder integration with existing inference engines, while prior sparse-serving abstractions support only specific layouts or workflows. We present SparseEngine, a ground-up, sparse-first inference engine whose shared lifecycle contract lets each method control its KV representation and computation while coordinating state transitions with common serving infrastructure. SparseEngine supports 15 methods across four categories and enables cross-request state management through Chain Cache, which resumes KV-eviction methods from retained history, and controllable Prefix-Cache Pruning, which removes KV from selected history regions while preserving logical-prefix matching. While maintaining method quality, SparseEngine delivers over 10x higher throughput with KV eviction, over 2.5x faster decoding at matched concurrency than vLLM, and over 2x end-to-end speedup on agent benchmarks. The code is available at https://github.com/CURRENTF/SparseEngine.

cs.LG↗
arXiv2026

MoRA: MoE Pruning via Router Bias Learning and Expert Approximation

Mixture-of-Experts (MoE) models enable parameter scaling with limited per-token computation by activating only a small subset of experts for each token, but deploying them still requires loading the complete expert pool into memory. Structured expert pruning can effectively reduce the memory usage by removing experts. However, existing pruning methods either use expert ranking criteria that are not well aligned with model performance or rely on effective expert subset searching that is computationally expensive. Moreover, these methods typically overlook the routing-behavior redundancy among the retained experts. In this paper, we propose MoE Pruning via Router Bias Learning and Expert Approximation (MoRA), a framework for structured MoE expert pruning. We introduce a learnable router bias for each expert and optimize these biases by minimizing the language-modeling loss and a routing-diversity regularizer. The learned router biases sharpen the routing probability distributions to identify experts critical to model performance while encouraging the selection of experts with diverse routing preferences. In addition, we introduce an expert approximation mechanism as a post-pruning enhancement. It leverages the remaining experts to approximate the outputs of pruned experts by affine transformation, further improving the performance of the pruned model. We evaluate MoRA on Qwen3-30B-A3B, DeepSeek-V2-Lite, and Moonlight-16B-A3B, removing 25\% and 50\% of the routed experts in each MoE layer. Extensive experiments on nine zero-shot benchmarks show that MoRA outperforms state-of-the-art pruning algorithms. Our code will be released.

cs.LG↗
arXiv2026

IDRBench: Benchmarking the Interactive Capabilities of Deep Research Agents

Large Language Model (LLM)-based deep research agents perform multi-step reasoning, web exploration, and long-form report generation. In these long-horizon workflows, early deviations from user intent can misdirect research and propagate through planning, search, and synthesis, making timely interaction essential. However, existing benchmarks primarily treat deep research as a static input-output task, overlooking agents' ability to elicit and use user feedback. We introduce IDRBench, a benchmark for evaluating interactive deep research with controlled opportunities for clarification. Within a common workflow and stage-wise interaction budget, IDRBench compares autonomous and interactive trajectories, measuring interaction benefit through changes in task-specific report alignment and interaction cost through turns and tokens. Comprehensive experiments on 100 tasks with seven proprietary and open-weight LLMs show that interaction improves all five alignment measures for every model, yielding an average gain of 6.39 points, while revealing distinct trade-offs among autonomous performance, alignment gain, and communication cost. At the task level, interaction improves performance in 74.4% of cases but degrades it in 19.9%, demonstrating that access to clarification alone does not guarantee better outcomes: success depends on what agents ask and how effectively they incorporate the resulting feedback.

cs.CL↗
arXiv2026

The "Rough" HAR Model

This paper proposes discrete-time approximations to rough continuous-time models of realized variance (RV). The leading rough models can be viewed as autoregressive processes driven by fractional Gaussian noise. We show that the Wold representation of this noise concentrates its dependence at the first lag when the Hurst parameter is below one half. Augmenting the autoregressive (AR) and heterogeneous autoregressive (HAR) models with a first-order moving-average (MA(1)) component therefore approximates the roughness, and the MA coefficient maps almost linearly into the Hurst parameter. We refer to these extensions as the "rough" AR and "rough" HAR models. Estimating them on the log RV of ten ETFs, we find negative MA coefficients for every asset, and the implied Hurst parameters align closely with the estimates from the continuous-time models. In the HAR literature, the negative MA(1) component is a significant feature that has been largely overlooked. In out-of-sample comparisons, the "rough" models outperform their classical counterparts for nearly every asset and horizon, with the largest gains at short horizons, and their accuracy is comparable to that of the rough continuous-time models but much easier to estimate by standard off-the-shelf software.

econ.EM↗
arXiv2026

Reconstructing Is Not Acting: Action-Centric Latent Dynamics Modeling

Latent action models (LAMs) learn action representations from unlabeled videos by inferring latent actions from visual transitions and reconstructing future states. However, we identify a fundamental $\textbf{reconstruction-action mismatch}$: lower reconstruction error does not necessarily yield better latent dynamics or downstream performance. We attribute this mismatch to two underconstrained aspects of reconstruction-based latent dynamics modeling: (i) the inverse dynamics model (IDM) is not explicitly encouraged to distinguish action-related transitions from nuisance appearance, and (ii) the forward dynamics model (FDM) can underutilize the inferred latent action by exploiting predictive shortcuts from the current state. To address both limitations, we propose $\textbf{ACT-LAM}$, a lightweight action-centric framework that strengthens both action extraction and action utilization. Specifically, its Action Query IDM (AQ-IDM) employs learnable action queries and gated aggregation to selectively extract rich action-related transition cues without strong information bottlenecks. And its Action Token FDM (AT-FDM) projects latent actions into action tokens that progressively interact with evolving state representations, enabling continuous state-aware action conditioning. ACT-LAM further streamlines feature processing to concentrate model capacity on latent dynamics modeling. Extensive experiments on several robotic datasets and the VP$^2$ benchmark demonstrate stronger latent action consistency, forward dynamics, and downstream visual planning performance with fewer trainable parameters and lower computational overhead. In particular, ACT-LAM surpasses the previous state of the art by $\textbf{7.6%}$ on the aggregated VP$^2$ success rate. Codes at $\href{https://github.com/DingjieFu/ACT-LAM}{url}$.

cs.CV↗
arXiv2026

Temporal Fourier Likelihoods with Spatial Hilbert-Space Gaussian Process Approximations

Reconstructing stationary space-time Gaussian processes at unobserved locations is costly when many sites share regular temporal records. We develop a spectral likelihood combining a temporal discrete Fourier transform (DFT) with a Hilbert-space Gaussian process (HSGP) representation of frequency-specific spatial covariance. We derive the exact covariance of the finite-record DFT coefficients and use a Whittle likelihood that approximates distinct frequencies as independent spatial problems. At each temporal frequency, HSGP approximates spatial covariance by evaluating the sampled spectral multiplier at retained Laplacian eigenfrequencies. For models specified by a joint spectral density whose half-spectrum lacks a convenient closed form, this construction avoids repeated Fourier inversion. The fixed spatial basis also permits cached feature projections to be reused in likelihood fitting and held-site reconstruction, with approximation accuracy depending on domain extension and basis size. In a simulation study of such a model, HSGP achieved reconstruction accuracy comparable to a high-accuracy quadrature reference while reducing mean fitting time by 69%. Additional applications to wind-field reconstruction tasks examine the effects of basis rank, temporal record length, and separability, and demonstrate accurate inference on held-out test sites when sufficiently rich bases are used. Taken together, the results indicate that computational savings are attainable when the spectral multiplier can be evaluated directly, numerical spatial inversion is costly, and an adequate basis has rank lower than the number of fitting sites.

stat.ME↗
arXiv2026

SoftRerank: Hierarchical Soft Fusion with Candidate-Label Reranking for Long-Tailed Micro-Action Recognition

Micro-actions are subtle, low-intensity non-verbal behaviors that provide cues to fine-grained human states, including emotions and intentions. Recognizing them remains difficult because they are brief, contain weak visual changes, and often exhibit similar motion patterns across categories. This paper addresses these challenges with a fine-grained micro-action recognition method that combines full fine-tuning of InternVideo2.5, hierarchical soft fusion, and a lightweight candidate-label reranker. For the long-tailed label distribution in MA-52, we use class-balanced sampling and inverse-frequency reweighting to reduce the effect of frequent classes during training. We fine-tune InternVideo2.5 end to end and attach coarse and group-conditional fine-grained classification heads to the shared video representation, improving the consistency between coarse and fine predictions. For ambiguous samples, the candidate-label reranker uses hard samples and video-label matching to focus on easily confused fine-grained actions. Experiments validate the proposed method, which achieves a 79.99% F1-mean on MA-52 and ranks first in the 3rd Micro-Action Analysis Grand Challenge at ACM Multimedia 2026.

cs.CV↗