Near-Oracle KV Selection via Pre-hoc Sparsity for Long-Context Inference

A core bottleneck in large language model (LLM) inference is the cost of attending over the ever-growing key-value (KV) cache. Although near-oracle top-k KV selection can preserve the quality of dense attention while sharply reducing computation and bandwidth, existing sparse methods generally rely on posterior heuristics, i.e., selectors conditioned on observed attention or proxy scores. Such conditioning introduces posterior bias: it tends to distort true token importance and miss salient tokens, thereby impairing long-range reasoning. To tackle this problem, we propose Pre-hoc Sparsity (PrHS), which selects KV entries before attention scoring and provides explicit accuracy control. Let the attention mass of discarded entries be delta (the dropped mass). Through a marginal-to-mutual-information analysis, we derive an upper bound on the mutual-information loss that depends only on the dropped mass. This relation explains failure modes of posterior heuristics and enables verifiable guarantees by controlling the dropped mass in advance. Within PrHS, we instantiate three orthogonal pre-hoc selectors along the axes of time, depth, and layer. Extensive experiments on LLaMA and Mistral families validate PrHS. Across GSM8K and CoQA, PrHS reduces retrieval overhead by over 90%, achieving 3x higher retrieval sparsity than HShare at matched or better accuracy. It incurs under 1% average degradation on LongBench, lowers attention FLOPs by about 15% versus prior sparse baselines, and yields a 9.9x speedup in attention-operator latency and 2.8x higher throughput on NVIDIA A100-80GB GPUs than the dense baseline.

Document Reconstruction Unlocks Scalable Long-Context RLVR

Reinforcement Learning with Verifiable Rewards~(RLVR) has become a prominent paradigm to enhance the capabilities (i.e.\ long-context) of Large Language Models~(LLMs). However, it often relies on gold-standard answers or explicit evaluation rubrics provided by powerful teacher models or human experts, which are costly and time-consuming. In this work, we investigate unsupervised approaches to enhance the long-context capabilities of LLMs, eliminating the need for heavy human annotations or teacher models' supervision. Specifically, we first replace a few paragraphs with special placeholders in a long document. LLMs are trained through reinforcement learning to reconstruct the document by correctly identifying and sequencing missing paragraphs from a set of candidate options. This training paradigm enables the model to capture global narrative coherence, significantly boosting long-context performance. We validate the effectiveness of our method on two widely used benchmarks, RULER and LongBench~v2. While acquiring noticeable gains on RULER, it can also achieve a reasonable improvement on LongBench~v2 without any manually curated long-context QA data. Furthermore, we conduct extensive ablation studies to analyze the impact of reward design, data curation strategies, training schemes, and data scaling effects on model performance. We publicly release our code, data, and models.

Internalizing LLM Reasoning via Discovery and Replay of Latent Actions

The internalization of chain-of-thought processes into hidden states has emerged as a highly efficient paradigm for scaling test-time compute. However, existing activation steering methods rely on static control vectors that fail to adapt to the non-stationary evolution of complex reasoning tasks. To address this limitation, we propose STIR (Self-Distilled Tools for Internal Reasoning), a framework that reformulates reasoning enhancement as a dynamic latent trajectory control problem. STIR introduces a synergistic three-stage pipeline: (1) differential intrinsic action induction harvests latent reasoning successes to crystallize steering primitives; (2) sparse control basis construction curates a compact, geometrically diverse tool library; and (3) value-modulated trajectory intervention dynamically injects context-specific impulses via anchor-based gating. Extensive experiments on six arithmetic and logical benchmarks across four representative models demonstrate that STIR improves average accuracy by 1.9% to 7.5% while reducing average token consumption by up to 35% compared to vanilla decoding. These findings demonstrate that the benefits of explicit chain-of-thought can be realized through dynamic latent trajectory control, internalizing the reasoning process to bypass the explicit generation while achieving superior fidelity. Our code is available at https://github.com/sznnzs/LLM-Latent-Action.

GraDE: A Graph Diffusion Estimator for Frequent Subgraph Discovery in Neural Architectures

Finding frequently occurring subgraph patterns or network motifs in neural architectures is crucial for optimizing efficiency, accelerating design, and uncovering structural insights. However, as the subgraph size increases, enumeration-based methods are perfectly accurate but computationally prohibitive, while sampling-based methods are computationally tractable but suffer from a severe decline in discovery capability. To address these challenges, this paper proposes GraDE, a diffusion-guided search framework that ensures both computational feasibility and discovery capability. The key innovation is the Graph Diffusion Estimator (GraDE), which is the first to introduce graph diffusion models to identify frequent subgraphs by scoring their typicality within the learned distribution. Comprehensive experiments demonstrate that the estimator achieves superior ranking accuracy, with up to 114\% improvement compared to sampling-based baselines. Benefiting from this, the proposed framework successfully discovers large-scale frequent patterns, achieving up to 30$\times$ higher median frequency than sampling-based methods.

FluxNet: Learning Capacity-Constrained Local Transport Operators for Conservative and Bounded PDE Surrogates

Autoregressive learning of time-stepping operators offers an effective approach to data-driven PDE simulation on grids. For conservation laws, however, long-horizon rollouts are often destabilized when learned updates violate global conservation and, in many applications, additional state bounds such as nonnegative mass and densities or concentrations constrained to [0,1]. Enforcing these coupled constraints via direct next-state regression remains difficult. We introduce a framework for learning conservative transport operators on regular grids, inspired by lattice Boltzmann-style discrete-velocity transport representations. Instead of predicting the next state, the model outputs local transport operators that update cells through neighborhood exchanges, guaranteeing discrete conservation by construction. For bounded quantities, we parameterize transport within a capacity-constrained feasible set, enforcing bounds structurally rather than by post-hoc clipping. We validate FluxNet on 1D convection-diffusion, 2D shallow water equations, 1D traffic flow, and 2D spinodal decomposition. Experiments on shallow-water equations and traffic flow show improved rollout stability and physical consistency over strong baselines. On phase-field spinodal decomposition, the method enables large time-steps with long-range transport, accelerating simulation while preserving microstructure evolution in both pointwise and statistical measures.

Self-Rewarding Sequential Monte Carlo for Masked Diffusion Language Models

This work presents self-rewarding sequential Monte Carlo (SMC), an inference-time scaling algorithm enabling effective sampling of masked diffusion language models (MDLMs). Our algorithm stems from the observation that most existing MDLMs rely on a confidence-based sampling strategy, where only tokens with the highest prediction confidence are preserved at each step. This restricts the generation to a noise-sensitive, greedy decoding paradigm, resulting in an inevitable collapse in the diversity of possible paths. We address this problem by launching multiple interacting diffusion processes in parallel, referred to as particles, for trajectory exploration. Importantly, we introduce the trajectory-level confidence as a self-rewarding signal for assigning particle importance weights. During sampling, particles are iteratively weighted and resampled to systematically steer generation towards globally confident, high-quality samples. Our self-rewarding SMC is verified on various masked diffusion language models and benchmarks, achieving significant improvement without extra training or reward guidance, while effectively converting parallel inference capacity into improved sampling quality. Our code is available at https://github.com/Algolzw/self-rewarding-smc.

LLM-AutoDP: Automatic Data Processing via LLM Agents for Model Fine-tuning

Large Language Models (LLMs) can be fine-tuned on domain-specific data to enhance their performance in specialized fields. However, such data often contains numerous low-quality samples, necessitating effective data processing (DP). In practice, DP strategies are typically developed through iterative manual analysis and trial-and-error adjustment. These processes inevitably incur high labor costs and may lead to privacy issues in high-privacy domains like healthcare due to direct human access to sensitive data. Thus, achieving automated data processing without exposing the raw data has become a critical challenge. To address this challenge, we propose LLM-AutoDP, a novel framework that leverages LLMs as agents to automatically generate and optimize data processing strategies. Our method generates multiple candidate strategies and iteratively refines them using feedback signals and comparative evaluations. This iterative in-context learning mechanism enables the agent to converge toward high-quality processing pipelines without requiring direct human intervention or access to the underlying data. To further accelerate strategy search, we introduce three key techniques: Distribution Preserving Sampling, which reduces data volume while maintaining distributional integrity; Processing Target Selection, which uses a binary classifier to identify low-quality samples for focused processing; Cache-and-Reuse Mechanism}, which minimizes redundant computations by reusing prior processing results. Results show that models trained on data processed by our framework achieve over 80% win rates against models trained on unprocessed data. Compared to AutoML baselines based on LLM agents, LLM-AutoDP achieves approximately a 65% win rate. Moreover, our acceleration techniques reduce the total searching time by up to 10 times, demonstrating both effectiveness and efficiency.

DeepResearchEval: An Automated Framework for Deep Research Task Construction and Agentic Evaluation

Deep research systems are widely used for multi-step web research, analysis, and cross-source synthesis, yet their evaluation remains challenging. Existing benchmarks often require annotation-intensive task construction, rely on static evaluation dimensions, or fail to reliably verify facts when citations are missing. To bridge these gaps, we introduce DeepResearchEval, an automated framework for deep research task construction and agentic evaluation. For task construction, we propose a persona-driven pipeline generating realistic, complex research tasks anchored in diverse user profiles, applying a two-stage filter Task Qualification and Search Necessity to retain only tasks requiring multi-source evidence integration and external retrieval. For evaluation, we propose an agentic pipeline with two components: an Adaptive Point-wise Quality Evaluation that dynamically derives task-specific evaluation dimensions, criteria, and weights conditioned on each generated task, and an Active Fact-Checking that autonomously extracts and verifies report statements via web search, even when citations are missing.

How Order-Sensitive Are LLMs? OrderProbe for Deterministic Structural Reconstruction

Large language models (LLMs) excel at semantic understanding, yet their ability to reconstruct internal structure from scrambled inputs remains underexplored. Sentence-level restoration is ill-posed for automated evaluation because multiple valid word orders often exist. We introduce OrderProbe, a deterministic benchmark for structural reconstruction using fixed four-character expressions in Chinese, Japanese, and Korean, which have a unique canonical order and thus support exact-match scoring. We further propose a diagnostic framework that evaluates models beyond recovery accuracy, including semantic fidelity, logical validity, consistency, robustness sensitivity, and information density. Experiments on twelve widely used LLMs show that structural reconstruction remains difficult even for frontier systems: zero-shot recovery frequently falls below 35%. We also observe a consistent dissociation between semantic recall and structural planning, suggesting that structural robustness is not an automatic byproduct of semantic competence.

RePose: A Real-Time 3D Human Pose Estimation and Biomechanical Analysis Framework for Rehabilitation

We propose a real-time 3D human pose estimation and motion analysis method termed RePose for rehabilitation training. It is capable of real-time monitoring and evaluation of patients'motion during rehabilitation, providing immediate feedback and guidance to assist patients in executing rehabilitation exercises correctly. Firstly, we introduce a unified pipeline for end-to-end real-time human pose estimation and motion analysis using RGB video input from multiple cameras which can be applied to the field of rehabilitation training. The pipeline can help to monitor and correct patients'actions, thus aiding them in regaining muscle strength and motor functions. Secondly, we propose a fast tracking method for medical rehabilitation scenarios with multiple-person interference, which requires less than 1ms for tracking for a single frame. Additionally, we modify SmoothNet for real-time posture estimation, effectively reducing pose estimation errors and restoring the patient's true motion state, making it visually smoother. Finally, we use Unity platform for real-time monitoring and evaluation of patients' motion during rehabilitation, and to display the muscle stress conditions to assist patients with their rehabilitation training.