×

Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

作者: Ao ×
换一批
01.
arXiv (CS.CL) 2026-06-16

Retrievable Gradients: Continual Post-Training Without Cumulative Weight Drift

Continual post-training enables models to absorb emerging knowledge after deployment, but repeatedly updating shared parameters can accumulate weight drift, potentially causing catastrophic forgetting and degrading general capabilities. Retrieval-augmented generation avoids such parameter drift, yet often lacks the depth of parametric knowledge integration. In this paper, we propose ReGrad (Retrievable Gradients), a new paradigm that treats gradients as retrievable units of knowledge. ReGrad pre-computes document-specific gradients offline, stores them in an indexed Gradient Bank, and retrieves only query-relevant gradients at inference time for temporary weight adaptation. However, raw language-modeling gradients are optimized for token-level document reconstruction rather than for query-driven knowledge use. We therefore introduce a bi-level meta-learning objective that reshapes document-derived gradients into generalizable adaptation signals for downstream tasks. Experiments across general and domain-specific settings show that \textsc{ReGrad} outperforms CPT and RAG baselines, enabling scalable and reversible parametric knowledge injection without accumulating weight drift.

02.
arXiv (CS.CV) 2026-06-24

MM-TRELLIS: Point-Cloud Guided Multi-Modal 3D Vehicle Generation in Autonomous Driving

Recovering realistic 3D vehicle models from autonomous driving scenes is crucial for synthesizing training data and building simulation environment. However, most existing vehicle generation methods fail to fully exploit multimodal sensors i.e. multi-view images and LiDAR point clouds) and rely on neural rendering based reconstruction, leading to low-quality mesh. Recently, native 3D generative models have made significant progress, yet they are not built for arbitrary multi-view inputs and often struggle with in-the-wild driving images. In this work, we present MM-TRELLIS, a multi-modal version of TRELLIS for in-the-wild 3D vehicle generation that integrates LiDAR and image sensors from autonomous driving datasets into native 3D generative models. Specifically, multi-view images are cycled as conditioning inputs, while LiDAR point clouds provide test-time guidance to ensure geometric accuracy and cross-view consistency. During denoising, we first align the guidance point cloud with the model priors, then enforce consistency between the generated geometry and the guidance point cloud. Finally, we introduce a voxel filtering strategy based on the opacity of 3D Gaussian Splatting to suppress floaters and produce clean meshes. Comprehensive experiments on Waymo dataset demonstrate our method outperforms existing methods in high-fidelity 3D vehicle generation. Code is available at https://github.com/HongliXiao/MM-TRELLIS.

03.
arXiv (CS.CV) 2026-06-12

Skill-3D: Evolving Scene-Aware Skills for Agentic 3D Spatial Reasoning

This paper explores agentic 3D spatial understanding, i.e., MLLM agents performing 3D reasoning through tool use. Existing methods often misuse tools and exhibit biased tool preferences under 3D scenarios, leaving the agentic paradigm with only marginal gains over non-agentic strategies. We reveal that 3D spatial reasoning tasks are heterogeneous across scenes, while these agents apply a uniform tool-use strategy to all scenes rather than selecting tools according to the specific scene and task. To address this, we propose Skill-3D, a framework that learns self-evolving scene-aware skills. Specifically, Skill-3D identifies the task scene and records the agent's tool-use trajectory into a Scene Memory, where successful trajectories from similar scenes are aggregated and distilled into a reusable scene-aware skill, with failed ones attached to the skill as lessons. During training, once a similar scene recurs, the corresponding skill is injected to guide the agent, producing new trajectories whose successes and failures further refine the skill, forming a loop in which the memory and the skill library co-evolve. Experiments show that Skill-3D substantially improves tool utilization in 3D spatial reasoning (from 39% to 78% on VSI-Bench), driving the agent toward correct and sufficient tool use. For instance, it improves Gemini-3-Flash by 67% on MMSI-Bench. Furthermore, we conduct agentic post-training over skill-guided trajectories, which boosts Qwen3-VL-8B by 60% on VSI-Bench.

04.
arXiv (CS.AI) 2026-06-11

SPEAR: A System for Post-Quantization Error-Adaptive Recovery Enabling Efficient Low-Bit LLM Serving

arXiv:2606.11244v1 Announce Type: cross Abstract: Efficient large language model (LLM) serving is increasingly constrained by deployment cost. Quantization is a key technique for reducing serving cost, yet even state-of-the-art 4-bit quantizers exhibit a noticeable quality gap from FP16, particularly for smaller models where low-bit serving is most beneficial. We identify a fundamental cause of this gap: quantization error is highly input-dependent and varies substantially across tokens, while existing post-quantization compensation methods are static and apply identical corrections to all inputs. As a result, easy tokens are over-corrected while hard tokens remain under-corrected. We present SPEAR, a system for post-quantization error-adaptive recovery that improves low-bit LLM serving. SPEAR introduces lightweight Error Compensators (ECs) modulated by per-token gates and places them only at the most error-sensitive layers identified through a CKA-guided entropy-aware diagnostic. This focuses a small parameter budget where it is most effective. Efficient deployment of ECs presents several systems challenges, including additional computation, tensor-parallel synchronization caused by input-dependent gating, and latency instability across configurations. SPEAR addresses these issues through adaptive kernel-fusion dispatch, combining an epilogue-integrated peer-reduction kernel with P2P dual-write to fuse the post-EC computation into low-bit GEMMs, and an SLO-constrained EC-aware scheduler for predictable serving performance. Across challenging per-channel quantization settings, SPEAR recovers 56-75% of the perplexity gap between W4 and FP16 while adding less than 1% model memory overhead and maintaining latency comparable to a widely used 4-bit serving deployment.

05.
arXiv (CS.AI) 2026-06-16

TimeVista: Exploring and Exploiting Vision-Language Models as Judges for Time Series Forecasting

arXiv:2606.16173v1 Announce Type: new Abstract: High-quality time series forecasting is pivotal for real-world decision-making. However, traditional point-wise metrics often fail to reveal complex temporal patterns and align poorly with human intuitive preferences. While the ''LLM-as-a-Judge'' paradigm has revolutionized text evaluation by providing flexible, human-aligned judgment, its application to time series remains largely unexplored. In this paper, we leverage Vision-Language Models (VLMs) as judges for time series forecasting, harnessing their ability to comprehend time series plots grounded in textual information. Specifically, we propose a novel framework integrating micro- and macro-level judgments informed by contextual information to evaluate time series forecasting. To this end, we introduce TimeVista, a comprehensive VLM-as-a-Judge benchmark comprising 5563 time series samples paired with detailed evaluation rubrics. Extensive meta-evaluations demonstrate that VLMs are highly reliable judges, achieving significantly higher consistency with human preferences than conventional metrics. Building upon our benchmark, we comprehensively assess recent Time Series Foundation Models (TSFMs) under the VLM-as-a-Judge paradigm. Our results demonstrate that VLMs serve as robust and interpretable judges, providing a comprehensive, human-aligned standard for evaluating time series models.

06.
arXiv (CS.CL) 2026-06-12

Multi-Bitwidth Quantization for LLMs Using Additive Codebooks

As large language models (LLMs) are increasingly deployed across heterogeneous hardware with varying resource constraints, the ability to adaptively manage the trade-off between performance and efficiency without retraining is critical. We propose Drop-by-Drop, a novel multi-bitwidth post-training quantization framework that enables inference-time precision control over LLM weights from a single trained model. Our method is theoretically grounded in information theory and successive refinement. We establish that LLM weights, which commonly follow a Gaussian distribution, can be optimally reconstructed with increasing fidelity as additional bits are incorporated, under a weighted mean squared error distortion motivated by LLM loss functions. To realize this in practice, Drop-by-Drop incorporates Matryoshka-style supervision into the loss function, exploiting the structure of additive codebooks. Drop-by-Drop produces a single model where ordered subsets of codebooks yield accurate partial reconstructions at each precision level. This approach significantly reduces storage and memory overhead by allowing a single checkpoint to serve multiple bitwidths, while maintaining competitive perplexity and accuracy across major architectures, such as Qwen, LLaMA, Gemma, and Mistral.

07.
arXiv (math.PR) 2026-06-11

Continuous stochastic flows driven by white noise and their duals

作者:

arXiv:2606.12143v1 Announce Type: new Abstract: We study a class of continuous stochastic flows driven by a space-time white noise and characterize their dual flows by explicit stochastic differential equations. A key ingredient of the proof is the convergence of solutions under coefficient approximations. As an application, we derive the dual flows in two illustrative examples, the squared Bessel flow and the Jacobi flow. We also introduce a new model of polynomially self-repelling (PSR) flow and show that it enjoys a self-duality property.

08.
arXiv (math.PR) 2026-06-16

Exponential Convengence of DLRA for SDEs

arXiv:2606.15843v1 Announce Type: new Abstract: We study dynamical orthogonal (DO) approximations of stochastic differential equations and investigate their long-time behaviour. The DO formulation represents the solution by a low-rank decomposition and leads to a coupled system consisting of an evolution equation on the Stiefel manifold and a reduced stochastic process. We establish the well-posedness of the strong DO system and derive quantitative error estimates between the original stochastic differential equation and its low-rank approximation in the Wasserstein distance. Our main contribution is the analysis of invariant probability measures for the DO dynamics. Under suitable dissipativity, Lipschitz continuity, and non-degeneracy assumptions on the coefficients, we prove the existence of an invariant probability measure for the strong DO system. The proof combines uniform moment estimates, a Krylov–Bogoliubov argument for an associated frozen system, and a Kakutani-Fan-Glicksberg fixed-point theorem to recover the self-consistent dynamics. We further show that the induced low-rank process admits an invariant probability measure and discuss the structure of invariant measures through several illustrative examples. These results provide a rigorous foundation for the use of dynamical low-rank approximations in the approximation of long-time statistical properties of stochastic dynamical systems.

09.
arXiv (CS.CL) 2026-06-12

Uncertainty-Aware Hybrid Retrieval for Long-Document RAG

Retrieval augmented generation (RAG) depends critically on the quality and granularity of retrieved evidence. Large retrieval units preserve context but often introduce irrelevant content, which can dilute answer bearing evidence and worsen long context utilization. Fine-grained units are more compact, but they may be difficult to retrieve reliably because short chunks can lack semantic, lexical, or bridging cues needed to match the query. We propose Uncertainty-aware Multi-Granularity RAG (UMG-RAG), a training-free hybrid retrieval framework that treats chunk granularity as query-specific reliability estimation. Instead of training a new retriever or modifying the generator, UMG-RAG uses existing dense and sparse retrievers as complementary experts across multiple chunk granularities. For each query, it converts each expert-granularity score list into an evidence distribution, estimates reliability from distribution entropy, and fuses candidates according to query-specific semantic, lexical, and granularity confidence. We further introduce UMGP-RAG, a parent promotion variant that uses fine-grained hits to locate relevant evidence while returning broader non-redundant parent chunks for local coherence. Experiments on question answering benchmarks show that uncertainty-aware fusion and parent promotion improve generation quality while maintaining a lightweight, plug-and-play retrieval pipeline.

10.
arXiv (CS.AI) 2026-06-17

LongWebBench: Evaluating Structural and Functional Webpage Generation in Long-Horizon Settings

arXiv:2606.17727v1 Announce Type: new Abstract: Recent vision-language models (VLMs) have shown promising progress in generating webpages from visual inputs, yet existing evaluations mainly focus on short, single-screen, and largely static webpages. We introduce LongWebBench, a benchmark for evaluating long-horizon webpage generation from both structural and functional perspectives. LongWebBench contains 490 real-world long webpages for structural fidelity evaluation and 507 goal-oriented interaction tasks over 129 webpages for functional evaluation. It employs two complementary protocols: a multi-dimensional VLM-based metric for assessing long-range structural coherence, and a DOM-augmented agent-based pipeline for end-to-end functional verification. We further examine the automatic evaluation protocols through human agreement analysis. Experiments with state-of-the-art open-source and proprietary VLMs under single-image and multi-image settings reveal that structural fidelity degrades as webpage length increases, while visually plausible generations often fail to support executable multi-step interactions. These results highlight the need to evaluate long webpage generation beyond visual similarity, with executable interaction as a core criterion. Our code and data are available at https://github.com/zheny2751-dotcom/LongWebBench.

11.
arXiv (CS.LG) 2026-06-15

Diffusion Policy Optimization without Drifting Apart

arXiv:2606.13795v1 Announce Type: new Abstract: RL post-training has become increasingly pivotal for improving diffusion policies, but existing diffusion policy-gradient methods are often unstable and cannot achieve reliable policy improvement. We identify the cause as the double-drift phenomenon: optimizing a variational surrogate can let the ELBO separate from the true log-likelihood, which then makes the resulting proxy policy gradient misaligned with the true policy gradient of expected return. We propose DiPOD, a diffusion policy optimization framework that maintains tight-bound behavior throughout training by interleaving self-distillation with policy-improving gradient updates. This leads to a simple and practical algorithm: augmenting each diffusion policy-gradient update with an on-policy ELBO regularizer. Across diffusion language model post-training and continuous-control diffusion policies, DiPOD substantially stabilizes training and reaches higher rewards than previous methods.

12.
arXiv (CS.AI) 2026-06-19

Co-policy: Responsive Human-Robot Co-Creation for Musical Performances

arXiv:2606.19914v1 Announce Type: cross Abstract: Art has long stood as a pivotal expression of human creativity. Embodied artificial intelligence offers a route for generative models to participate in that creativity through physical action rather than disembodied digital content. In robotic music co-creation, it is challenging to connect semantic musical understanding with real-time and physically executable performance. We present Co-policy, a framework for human-robot musical co-creation that separates semantic intent grounding, constrained musical variation, and visuomotor execution. To ground musical semantics, Co-policy uses pre-inference semantic anchors and a fine-tuned Qwen-vl planner (F-Qwen) to transform speech, live musical seeds, and visual observations into structured co-creation plans. To support low-latency execution, Co-policy introduces a Gaussian-Mixture Visuomotor Policy (GMP), implemented as a conditional mixture-density policy that maps target notes and visual context to multimodal robot actions in a single forward pass. Unlike robotic playback systems that merely reproduce user-specified notes, Co-policy generates complementary musical responses under both musical and physical constraints. Real-robot chime experiments, ablations, and expert evaluation show improved intent alignment, execution accuracy, and response frequency over diffusion-policy and ablated baselines, supporting physically grounded action generation as a key requirement for embodied human-AI co-creation.

13.
arXiv (quant-ph) 2026-06-16

Efficient Implementation of a Single-Qutrit Gate Set via Coherent Control

arXiv:2507.06860v2 Announce Type: replace Abstract: Qutrits offer the potential for enhanced quantum computation by exploiting an enlarged Hilbert space. However, the synthesis of high-fidelity and fast qutrit gates, particularly for single qutrits, remains an ongoing challenge, as it involves overcoming intrinsic constraints in quantum platforms. Here, we develop a novel framework for the efficient implementation of a single-qutrit gate set via coherent control, leveraging SU(3) dynamics while obviating platform-specific constraints such as those arising from the selection rule. As a proof-of-principle demonstration, we realize 35-ns qutrit Hadamard and X gates using a superconducting transmon, achieving an average fidelity of 99.5\%, as verified by randomized benchmarking. We further demonstrate two paradigmatic quantum circuits, which can be naturally extended to scalable qudit algorithms for phase estimation and parity check. In addition, we propose an SU(3)-based decomposition strategy for an arbitrary single-qutrit gate and numerically demonstrate its substantial efficiency improvement over conventional SU(2)-based protocols. By addressing the challenge of efficiently implementing single-qutrit gates, our protocol paves the way for realizing high-performance qutrit processors in diverse quantum platforms.

14.
arXiv (CS.CL) 2026-06-25

SARA: Unlocking Multilingual Knowledge in Mixture-of-Experts via Semantically Anchored Routing Alignment

Sparse Mixture-of-Experts (MoE) architectures have emerged as an increasingly influential paradigm as they offer a strategic balance between parameter scalability and computational efficiency. However, low-resource languages, which suffer from a scarcity of high-quality training data, often have their tokens routed to different experts than those predominantly activated by high-resource inputs, which limits cross-lingual expert sharing. This cross-lingual routing divergence consequently hinders their efficacy in multilingual contexts. To address this issue, we propose SARA (Semantically Anchored Routing Alignment), a framework designed to transfer specialized capabilities from high-resource languages as anchors to low-resource languages. SARA explicitly aligns the routing distribution of multilingual inputs with high-resource semantic anchors using a symmetric Jensen-Shannon (JS) divergence constraint. Unlike traditional distillation methods that operate on output logits, SARA directly aligns the internal routing distributions of MoE layers, encouraging mechanistic consistency in expert selection across languages. We conduct experiments on 2 LLMs across 5 low-resource languages and 3 benchmarks. Experiment results demonstrate that SARA outperforms standard instruction tuning, e.g., +0.8% on Qwen3-30B-A3B and +1.2% on Phi-3.5-MoE-instruct on Global-MMLU. Further analyses show that SARA effectively addresses performance bottlenecks in low-resource languages, providing a scalable pathway to enhance multilingual capabilities in sparse architectures.

15.
arXiv (CS.LG) 2026-06-12

SMGFM: Spectral Multimodal Graph Pretraining for Multimodal-Attributed Graphs

arXiv:2606.12867v1 Announce Type: new Abstract: Multimodal-attributed graphs (MAGs) couple graph topology with node semantics from text, images, and other modalities. Traditional graph learning contextualizes node semantics by coupling topology with node features. However, this coupling design becomes troublesome in MAGs, where structure-induced and modality-intrinsic semantics may contribute differently to downstream tasks. Structure-induced semantics promote relational consistency through smooth topological variation, whereas modality-intrinsic semantics often encode local, fine-grained distinctions that should not be uniformly smoothed or aligned. Therefore, the key challenge is to identify semantic roles before cross-modal fusion. To this end, we leverage graph-frequency variation as a prior, where low-frequency components capture topology-consistent semantics and high-frequency components preserve modality-specific semantics. Based on this intuition, we propose SMGFM, a spectral multimodal graph pretraining framework that decomposes each modality-specific node signal into graph-frequency bands and assigns band-level semantic roles before cross-modal interaction. Concretely, SMGFM constructs frequency-resolved modality tokens with scalable Chebyshev filters, estimates their coupling reliability through topology-conditioned routing, and performs band-modality interaction before fusion. Its frequency-routed objectives align smooth consensus routes while preserving modality-specific routes, mitigating spatial-domain entanglement and uniform cross-modal alignment. Extensive experiments conducted on the MAG datasets demonstrate that SMGFM achieves state-of-the-art performance across graph-level and modality-level tasks.

16.
arXiv (CS.CV) 2026-06-17

Recover Semantics First, Generate Better: Improved Latent Modeling for 3D MRI Reconstruction and Cross-Contrast Synthesis

Multi-contrast magnetic resonance imaging (MRI) provides complementary information for clinical diagnosis. However, acquiring all MRI sequences is often time-consuming and costly. Recent generative models perform cross-contrast synthesis to address this issue by inferring absent contrasts from the available ones. Nevertheless, synthesizing 3D MRI presents significant challenges. Due to the massive volume sizes, operating directly in the pixel space is computationally prohibitive; therefore, a common approach is to first compress the 3D volumes into a latent space and subsequently train generative models in that space. We observe that existing compression architectures face several critical issues: they under-preserve long-range anatomical coherence, discard clinically meaningful semantics, and rely on optimization objectives that lead to over-smoothed reconstructions. Ultimately, these shortcomings compromise the performance of subsequent generative models. In this work, we propose a semantics-first latent modeling framework for 3D MRI reconstruction and cross-contrast synthesis. Specifically, we introduce a Latent Harmonization Encoder (LHE) to capture global anatomical dependencies, ensuring coherent volumetric representations. To mitigate semantic degradation during latent compression, we further design a Semantic Recovery Block (SRB) that injects high-level priors from a self-supervised semantic teacher, enhancing contrast-aware separability in the latent space. Additionally, we propose an Anatomy-aware Frequency Loss (AFL) to adaptively preserve diagnostically relevant high-frequency structures. Extensive experiments on two public multi-contrast MRI datasets demonstrate consistent improvements in reconstruction fidelity and cross-contrast synthesis quality. Our code is available at https://github.com/script-Yang/RSF.

17.
arXiv (CS.CV) 2026-06-11

SpecLoR: Spectral Lookahead Rectification for Motion-Coherent Text-to-Video Generation

Flow Matching has enabled robust text-to-video generation via latent ODE sampling. However, velocity approximation and numerical discretization errors inevitably accumulate, causing sampling trajectories to drift. Consequently, generated videos often suffer from severe spatiotemporal inconsistencies. Nevertheless, directly correcting these drifted, noisy latents is challenging: (i) timestep-dependent noise obscures reliable structural cues; (ii) spatial interventions risk disrupting intricate local geometry while incurring heavy computational costs. To address this, we propose Spectral Lookahead Rectification (SpecLoR), a plug-and-play inference method that bypasses noise via lookahead prediction, and circumvents spatiotemporal entanglement by shifting corrections to the frequency domain, where universal statistical priors of natural videos are readily available. First, during early sampling stages, SpecLoR looks ahead to estimate the clean latent $z_{t,0}$ and computes its 3D spatiotemporal spectrum. Next, SpecLoR rectifies the amplitude spectrum to match the prior, leaving the phase intact. Finally, the corrected state is re-noised to resume ODE integration. Experiments on Wan2.2 demonstrate that SpecLoR significantly reduces physical artifacts and enhances motion coherence across multiple benchmarks with minimal computational overhead (4 additional NFEs).

18.
arXiv (CS.CL) 2026-06-16

ChildGuard: A Specialized Dataset for Combatting Child-Targeted Hate Speech

Mental health industry faces growing concerns regarding hate speech directed at children's on social media, as exposure to such content can contribute to adverse psychological outcomes during critical stages of development. Current hate speech datasets and detection systems provide limited support for child-focused applications because they are primarily designed for adults and lack dedicated representations of age-specific characteristics associated with hate speech directed at children's. To address this gap, we introduce ChildGuard, a large-scale English dataset for child-targeted hate speech containing 351,877 annotated instances collected from X (formerly Twitter), Reddit, and YouTube. The dataset covers three age groups such as younger children's (under 11), pre-teens (11-12), and teens (13-17). ChildGuard contains two subsets such as a contextual subset (157K) and a lexical subset (194K). Evaluation using recent transformer-based models and LLMs achieves a best Macro-F1 of 82.07%, decreasing to 79.41%, 79.24%, 76.04%, and 74.88% on younger children's, contextual, implicit hate, and cross-subset settings, respectively.

19.
arXiv (CS.CL) 2026-06-11

Neuron-based Personality Trait Induction in Large Language Models

Large language models (LLMs) have become increasingly proficient at simulating various personality traits, an important capability for supporting related applications (e.g., role-playing). To further improve this capacity, in this paper, we present a neuron-based approach for personality trait induction in LLMs, with three major technical contributions. First, we construct PersonalityBench, a large-scale dataset for identifying and evaluating personality traits in LLMs. This dataset is grounded in the Big Five personality traits from psychology and is designed to assess the generative capabilities of LLMs towards specific personality traits. Second, by leveraging PersonalityBench, we propose an efficient method for identifying personality-related neurons within LLMs by examining the opposite aspects of a given trait. Third, we develop a simple yet effective induction method that manipulates the values of these identified personality-related neurons. This method enables fine-grained control over the traits exhibited by LLMs without training and modifying model parameters. Extensive experiments validate the efficacy of our neuron identification and trait induction methods. Notably, our approach achieves comparable performance as fine-tuned models, offering a more efficient and flexible solution for personality trait induction in LLMs. We provide access to all the mentioned resources at https://github.com/RUCAIBox/NPTI.

20.
arXiv (CS.LG) 2026-06-18

Gaussian Mixture Attention: Linear-Time Sequence Mixing via Probabilistic Latent Routing

arXiv:2606.18283v1 Announce Type: new Abstract: The dense token-to-token interaction pattern of standard dot-product attention remains a central bottleneck in scaling Transformer architectures to long contexts. We introduce Gaussian Mixture Attention (GMA), a probabilistic attention-style sequence mixer that replaces explicit pairwise query–key comparison with routing through $K$ learned Gaussian mixture components. Queries and keys are mapped to posterior responsibility vectors over a shared latent routing space; their overlap defines an implicit responsibility-space affinity, while values are written into and read from a $K$-slot latent memory. By exploiting the associativity of matrix multiplication, GMA avoids materializing the induced $N\times N$ affinity matrix and instead uses two responsibility matrices whose dominant activation storage scales as $\mathcal{O}(NK)$ rather than $\mathcal{O}(N^2)$ for fixed $K$. We formulate bidirectional and causal variants of GMA, provide an end-to-end differentiable parameterization of the Gaussian mixture components, and analyze its responsibility-modulated gradient structure, constrained non-negative low-rank affinity interpretation, and local routing stability. Empirically, GMA exhibits the intended fixed-$K$ linear memory scaling and is competitive with attention-style baselines on long-context classification, while causal GMA improves over tested linear/random-feature attention variants on WikiText-103 but remains behind optimized causal SDPA and Mamba in the current implementation. Analysis of learned responsibilities further shows broad component usage and moderate alignment with surface-form token categories, supporting GMA as a probabilistic, interpretable, fixed-$K$ linear-time attention-style alternative rather than a universal replacement for optimized softmax attention or state-space models.

21.
arXiv (CS.AI) 2026-06-24

CompressKV: Semantic-Retrieval-Guided KV-Cache Compression for Resource-Efficient Long-Context LLM Inference

arXiv:2606.24467v1 Announce Type: new Abstract: Long-context large language model (LLM) inference is increasingly constrained by the memory footprint and decoding cost of key-value (KV) caches, limiting sustainable deployment on resource-constrained hardware. Existing KV cache eviction methods typically apply heuristic token scoring over all heads in GQA-based LLMs. These methods ignore the different functionalities of attention heads, leading to the eviction of critical tokens and thus degrading the performance of LLMs. To address this issue, we propose CompressKV, a resource-efficient KV-cache compression framework for GQA-based LLMs. Instead of aggregating attention scores from all heads, CompressKV identifies Semantic Retrieval Heads (SRHs) that capture both the initial and final tokens of a prompt and semantically important mid-context evidence, and uses them to select tokens whose KV pairs should be retained. Furthermore, CompressKV allocates cache budgets across layers according to offline estimates of layer-wise eviction error. Experiments on LongBench and Needle-in-a-Haystack show that CompressKV consistently outperforms existing KV-cache eviction methods across memory budgets. Notably, it preserves over 97\% of full-cache performance using only 3\% of the KV cache on LongBench question-answering tasks and achieves 90\% accuracy with just 0.7\% KV storage on Needle-in-a-Haystack. These results demonstrate an improved resource–performance trade-off for long-context LLM inference. Our code is publicly available at: https://github.com/TUDa-HWAI/CompressKV

22.
arXiv (CS.AI) 2026-06-17

MODE: Modality-Decomposed Expert-Level Mixed-Precision Quantization for MoE Multimodal LLMs

arXiv:2606.17118v1 Announce Type: cross Abstract: Mixture-of-Experts Multimodal Large Language Models (MoE-MLLMs) offer remarkable performance but incur prohibitive GPU memory costs, making compression essential. Among PTQ methods, expert-level mixed-precision quantization has proven effective for MoE-LLMs, yet suffers notable degradation on MoE-MLLMs due to two overlooked biases in expert importance estimation. (1) At the cross-modal level, the numerical dominance of vision tokens causes expert selection frequency to be dominated by vision tokens, masking experts that are critical to the text modality; (2) at the intra-vision level, the large proportion of redundant vision tokens further skew frequency statistics, obscuring experts critical for informative visual content. To bridge gaps, we propose MODE, a modality-decomposed expert-level mixed-precision quantization framework for MoE-MLLMs that decomposes expert selection frequency by modality, filters redundant vision tokens to obtain denoised visual frequency, and further evaluates quantization sensitivity per modality as a complementary signal to frequency-based estimation. These signals are integrated into an Integer Linear Programming formulation to assign per-expert bit-widths under a given budget. Extensive experiments show that MODE is particularly well-suited for MoE-MLLMs, limiting average performance loss to within 2.9% at W3A16, with larger gains at the extreme 2-bit setting.

23.
arXiv (CS.CL) 2026-06-16

Beyond Monolingual Deep Research: Evaluating Agents and Retrievers with Cross-Lingual BrowseComp-Plus

Deep research agents are increasingly evaluated on their ability to search for evidence, reason over retrieved sources, and produce grounded answers. Existing browsing benchmarks, however, largely assume that the user's query and the supporting evidence are written in the same language, leaving open whether agentic search systems can operate when relevant evidence appears in another language. We introduce XBCP (Cross-lingual BrowseComp-Plus), a controlled benchmark that preserves the English question-and-answer space of BrowseComp-Plus but varies the languages of the supporting documents. XBCP instantiates two complementary settings: in the cross-lingual setting, each query is paired with evidence in a single assigned language. In the multilingual setting, the full evidence corpus is distributed equally and randomly across 12 languages spanning high-resource and low-resource regimes. We evaluate four deep research agents using sparse and dense multilingual retrievers, measuring answer accuracy, evidence recall, search behavior, calibration, citation fidelity, and oracle retrieval. Results reveal substantial degradation when evidence is translated. Even strong, dense retrievers lose evidence recall, and agents become less calibrated and cite evidence less reliably. Notably, accuracy remains lower even when all gold evidence is supplied directly. These findings suggest that cross-lingual deep research exposes both retrieval failures and an independent, agent-side difficulty in integrating language-mismatched evidence.

24.
arXiv (CS.CV) 2026-06-16

MoECa: Aligning Feature Reuse with Expert Decomposition in Diffusion Transformers

Diffusion Transformers with Mixture-of-Experts (DiT-MoE) improve model capacity under sparse activation, but diffusion inference is still bottlenecked by redundant computation across timesteps. Existing caching methods mainly operate at the token level, which becomes suboptimal in DiT-MoE because each token update is internally decomposed into multiple routed expert branches. Our analysis shows that cross-timestep redundancy in DiT-MoE is better characterized at the expert-branch level than at the whole-token level. Based on this observation, we propose MoECa, a fine-grained caching framework that performs branch-level feature reuse across timesteps. MoECa further introduces expert-aware adaptive control and synchronized cache updates across MoE and attention paths to maintain stable intermediate states. Experiments on multiple DiT-MoE models show that MoECa consistently achieves a better speed-quality trade-off than prior caching methods, with up to 2.83$\times$ inference speedup and minimal quality degradation.

25.
arXiv (CS.CV) 2026-06-12

CineDance: Towards Next-Generation Multi-Shot Long-Form Cinematic Audio-Video Generation

The fidelity and structural diversity of training datasets fundamentally determine the capabilities of video generation models. While commercial systems showremarkableabilitytogeneratecinematicnarratives, the progress of open-source models remains limited by the scarcity of high-quality training data. To bridge this gap, we introduce CineDance-1M, a large-scale, open research Text-to-Audio-Video (T2AV) dataset designed specifically for multi-shot, long-form joint audio-video generation. Averaging 92.8 seconds and 24.2 continuous shots per video, it provides configurable, structured annotations for both audio and video modalities. This exceptional quality is achieved through a rigorous three-stage curation pipeline: i) diverse sourcing and comprehensive cleansing, ii) film-theory-inspired narrative parsing, and iii) hierarchical dual-modal captioning. For a comprehensive assessment, we propose CineBench, featuring a diverse prompt suite and a six-dimensional, human-aligned metric system tailored for complex narrative audio-video evaluation. Furthermore, we adapt LTX-2.3 into CineDance, which demonstrates exceptional single-modality quality alongside precise audio-video alignment and robust subject and environment consistency, effectively validating our curation strategy and the high quality of CineDance-1M. We anticipate that this work will serve as a solid foundation for accelerating future research in multi-shot, long-form joint audio-video generation. Our project page is available at https://aliothchen.github.io/projects/CineDance/.