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。

01.
arXiv (math.PR) 2026-06-15

Sharp Favard length of random Cantor sets

arXiv:2512.17753v2 Announce Type: replace-cross Abstract: We show that for a large class of planar $1$-dimensional random fractals $S$, the Favard length $\operatorname{Fav}(S(r))$ of the neighborhood $S(r)$ is comparable to $\log^{-1}(1/r)$, matching a universal lower bound; up to now, this was only known in expectation for a few concrete models. In particular, we show that there exist $1$-Ahlfors regular sets with the fastest possible Favard length decay. For a wide class of planar one-dimensional "grid random fractals", including fractal percolation and its Ahlfors-regular variants, we further show that $\operatorname{Fav}(S(r))/\log(1/r)$ converges almost surely, and we identify the limit explicitly. Furthermore, we prove that for some $1$-dimensional Ahlfors-regular random fractals $S$, the Favard length of $S(r)$ decays instead like $\log\log(1/r)/\log(1/r)$, showing that the $1/\log(1/r)$ decay is not universal among random fractals, as might be expected from previous results.

02.
arXiv (quant-ph) 2026-06-15

Dissipation-induced superradiance in matter coupled to a self-interacting cavity

arXiv:2606.14526v1 Announce Type: new Abstract: Light-matter interactions are often modeled via the Dicke model, namely, by two-level systems coupled to a cavity mode. Alas, the threshold for superradiance is often experimentally inaccessible or hindered by light's diamagnetic term. Here, within the Dicke setting, we consider self-interacting light in a cavity, modeled by a photonic Kerr nonlinearity. We show that negative Kerr nonlinearity gives rise to a low-threshold superradiant phase with spin inversion. While unstable in a closed system, cavity dissipation stabilizes this lit phase, opening avenues for lasing and bath-engineered phases.

03.
arXiv (CS.CL) 2026-06-16

Re-feeding Is Not Replaying: Measuring Replay Noise in Counterfactual Token-Credit Estimation

Per-token counterfactual credit estimation asks which token in a language-model rollout caused the final answer to be right or wrong: cut the transcript at a pivot, substitute an alternative token, replay continuations, and compare outcomes. Published methods re-feed the transcript prefix as a fresh prompt, assuming this reproduces the state the model passed through during generation. We measure what that assumption costs on a stock inference engine, with a three-pass design: continuations resumed from the verified decode-time KV state, an identical second exact pass (a replica noise floor), and a re-feed pass. Across six configurations and three models (including a GRPO-trained checkpoint), at low-margin decision tokens, re-feeding changes the credit estimate at rates 14-28 percentage points above the replica floor (7-21pp under a treatment-independent conditioning; problem-clustered t = 2.9-6.4). Most changes are zero-boundary crossings of the quantized estimator rather than polarity reversals, and the perturbation is consistent with mean-zero, so averaged quantities are largely safe; but selection is not: a critical-token set chosen by thresholding $|\hat{A}_t|$ under re-feed overlaps the exact-resume selection at Jaccard 0.34-0.90, versus a 0.63-0.96 replica ceiling. A causal confirmation closes the loop: under vLLM's batch-invariant kernels all three passes are identical on every measured channel, with both disagreement rates exactly zero. Replica passes themselves disagree on 9-23% of eligible estimates: single-sample credit measurements at decision tokens are unreliable under any replay. Settings were fixed in advance; exact-pass cache hits in the second campaign are instrumented (100% hit rate, 3,434 pivots); total compute was under 10 USD. We recommend that counterfactual credit studies resume decoder state or use batch-invariant kernels, and report a replica floor.

04.
arXiv (CS.CV) 2026-06-16

IGLU: The Integrated Gaussian Linear Unit Activation Function

Activation functions are fundamental to deep neural networks, governing gradient flow, optimization stability, and representational capacity. Within historic deep architectures, while ReLU has been the dominant choice for the activation function, modern transformer-based models increasingly are adopting smoother alternatives such as GELU and other self-gated alternatives. Despite their empirical success, the mathematical relationships among these functions and the principles underlying their effectiveness remains only partially understood. We introduce IGLU, a parametric activation function derived as a scale mixture of GELU gates under a half-normal mixing distribution. This derivation yields a closed-form expression whose gating component is exactly the Cauchy CDF, providing a principled one-parameter family that continuously interpolates between identity-like and ReLU-like behavior via a single sharpness parameter $\sigma$. Unlike GELU's Gaussian gate, IGLU's heavy-tailed Cauchy gate decays polynomially in the negative tail, guaranteeing non-zero gradients for all finite inputs and offering greater robustness to vanishing gradients. We further introduce IGLU-Approx, a computationally efficient rational approximation of IGLU expressed entirely in terms of ReLU operations that eliminates transcendental function evaluation. Through evaluations on CIFAR-10, CIFAR-100, and WikiText-103 across ResNet-20, ViT-Tiny, and GPT-2 Small, IGLU achieves competitive or superior performance on both vision and language datasets against ReLU and GELU baselines, with IGLU-Approx recovering this performance at substantially reduced computational cost. In particular, we show that employing a heavy-tailed gate leads to considerable performance gains in heavily imbalanced classification datasets.

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

Surpassing Scale by Efficiency: A Compact 135M Parameter Foundational LLM Natively Adapted for the Bangla Language

While the NLP landscape is dominated by multi-billion parameter architectures, their deployment in low-resource, non-Latin scripts remains computationally prohibitive for edge configurations, mobile systems, and decentralized local hardware. This paper presents bangla-smollm-135m, a highly compact 135-million parameter decoder-only foundational model engineered explicitly for high-efficiency language modeling in the Bangla script. By leveraging a deterministic intersect-and-append token merging strategy between TituLLMs and SmolLM2-135M, the model overcomes subword script fragmentation without destabilizing early pretrained parameter states. In zero-shot multi-task benchmark evaluations (PIQA_bn, OpenBookQA_bn, CommonsenseQA_bn, and Bangla_MMLU), bangla-smollm-135m matches or outperforms models twice its size (Gemma-3-270m) and achieves parity with models in the 1B parameter tier. The model is available at rnnandi/bangla-smollm-135m

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

Marginal Alignment Does Not Guarantee Joint-Distribution Fidelity: An Official-Reference Audit of Nemotron-Personas-Korea with Cross-Locale Replication

Synthetic persona datasets cite alignment with official demographics as a basis for trust, yet downstream users consume them as joint structures across age, sex, region, occupation, education, name, and institutional status. Marginal alignment does not imply that these joints are preserved. We propose the Independence-Assumption Footprint (IAF), an audit primitive that operates on the attribute combinations a dataset card itself documents as treated independently. For each such combination, IAF compares the synthetic joint against an external official or institutional reference, using direct joint tables where available and rule-implied checks otherwise. Applied to NVIDIA Nemotron-Personas-Korea (one million Korean synthetic personas), IAF finds that NPK aligns with KOSIS marginals while three joints fail. The major-by-occupation distribution against the KEIS graduate universe carries a large conditional mismatch. The age profile of military service is institutionally inconsistent. Female representation in male-dominated occupations is substantially over-flattened toward parity, with the strict screening verdict mapping-dependent and age-robust under direct standardisation. A transferability demonstration across six further NPK locales finds locale-dependent rather than universal diagnostics, with reference-taxonomy cardinality confounding cross-locale flag counts. For synthetic personas used as silicon samples, marginal claims must therefore be paired with disclosure-anchored joint audits before reuse. The released audit artefacts (reference manifests, occupational crosswalks, derived metrics, reproducibility scripts) instantiate this protocol on the NPK family and are released for retargeting at other synthetic persona resources.

07.
arXiv (CS.LG) 2026-06-11

NetBurst: Event-Centric Forecasting of Bursty, Intermittent Time Series

arXiv:2510.22397v2 Announce Type: replace-cross Abstract: Network operators monitor their infrastructure by collecting telemetry data such as packet counts, byte rates, or flow volumes, yet answering the questions that effective operations demand – forecasting future load, diagnosing and characterizing anomalies, and searching for and retrieving historical precedents – requires more than raw measurements. Bridging this gap calls for learned representations: compact per-entity summaries that capture temporal dynamics from each entity's univariate time series. Time-series foundation models are the natural starting point, but they are designed for dense, periodic benchmark datasets – the mild statistical regime. However, network telemetry data inhabits the wild regime: operationally relevant events are rare, separated by variable-length stretches of low or no activity (``ebbs''), with intermittent bursts of heavy-tailed extremes (``tides''). We present NetBurst, an event-centric pipeline that collapses ebbs, separates each time series into a stream of burst timings and a stream of burst magnitudes, and learns a single representation serving all three operational tasks. Compared to the strongest competitors among eight baselines – including Amazon's Chronos-2 and Datadog's Toto – and across nine production telemetry configurations, NetBurst reduces median forecasting error by $1.3$–$116\times$ on wild-regime data with a $1.0$–$7.5\times$ better match to the true burst distribution, and matches baselines on mild-regime benchmarks. For characterizing anomalies, NetBurst produces balanced, well-spread clusters that are $16\times$ more describable in operator-familiar terms under a novel interpretability score, and cluster-filtered search delivers $7.5\times$ faster end-to-end retrieval.

08.
arXiv (CS.LG) 2026-06-15

Ensembling Sparse Autoencoders

arXiv:2505.16077v2 Announce Type: replace Abstract: Sparse autoencoders (SAEs) are used to decompose neural network activations into human-interpretable features. Typically, features learned by a single SAE are used for downstream applications. However, it has recently been shown that a single SAE captures only a limited subset of features that can be extracted from the activation space. Motivated by this limitation, we introduce and formalize SAE ensembles. Furthermore, we propose to ensemble multiple SAEs through naive bagging and boosting. In naive bagging, SAEs trained with different weight initializations are ensembled, whereas in boosting SAEs sequentially trained to minimize the residual error are ensembled. Theoretically, naive bagging and boosting are justified as approaches to reduce reconstruction error. Empirically, we evaluate our ensemble approaches with three settings of language models and SAE architectures. Our empirical results demonstrate that, compared to an expanded SAE that matches the number of features in the ensemble, ensembling SAEs improves the reconstruction of language model activations along with SAE stability. Additionally, on downstream tasks such as concept detection and spurious correlation removal, SAE ensembles achieve better performance, showing improved practical utility.

09.
arXiv (CS.LG) 2026-06-19

How to sketch a learning algorithm

作者:

arXiv:2604.07328v3 Announce Type: replace Abstract: How does the choice of training data influence an AI model? This broad question is of central importance to interpretability, privacy, and basic science. At its technical core is the data deletion problem: after a reasonable amount of precomputation, quickly predict how the model would behave in a given situation if a given subset of training data had been excluded from the learning algorithm. We present a data deletion scheme capable of predicting model outputs with vanishing error $\varepsilon$ and failure probability $\delta$ in the deep learning setting. Our precomputation and prediction algorithms are only $\tilde{O}(\log(1/\delta)/\varepsilon^2)$ factors slower than regular training and inference, respectively. The storage requirements are those of $\tilde{O}(\log(1/\delta)/\varepsilon^2)$ models. Our proof is based on an assumption that we call stability. In contrast to the assumptions made by prior work, stability appears to be fully compatible with learning powerful AI models. In support of this, we show that stability is satisfied in a minimal set of experiments with microgpt. Our code is available at https://github.com/SamSpo1/microgpt-sketch. At a technical level, our work is based on a new method for locally sketching an arithmetic circuit by computing higher-order derivatives in random complex directions. Forward-mode automatic differentiation allows cheap computation of these derivatives.

11.
bioRxiv (Bioinfo) 2026-06-12

Systematic functional annotation of thousands of BAHD acyltransferases in plant genomes using Protein Language Model and phylogenomic tools

The functional annotation of plant genes lags significantly behind their genomic annotation. Closing this gap requires thorough cataloging of reported protein activities alongside predictive methods that scale beyond sequence-similarity inference. Focusing on the BAHD acyltransferase enzyme family as a model, we assembled FuncZymeDB-BAHD, a large database of 2,705 LLM-retrieved and curated enzyme-acceptor-donor activities covering 336 BAHDs from 156 plant species, a 2-to-6-fold expansion over Swiss-Prot and prior compilations. We further developed FuncPred-OG, which maps queries to orthologous groups and previously characterized enzymes in FuncZymeDB-BAHD, returning hits with high evidence provenance. FuncPred-OG enabled functional prediction of over half of BAHDs across 85 plant proteomes, of which five novel predictions were validated via in vitro assays and recent studies. For the remaining BAHDs without FuncPred-OG annotation, we developed FuncPred-AI, where logistic-regression classifiers trained on protein language model embeddings achieved high Area-Under-the-Precision-Recall-curve (AUPR) scores and correct-hit rates up to 93%. FuncPred-AI yielded >1 probable donor/acceptor annotation for 99.9% (8894/8897) of BAHDs in our pan-plant dataset. Finally, the FuncPred workflow and datasets were deployed on a web portal for broader utilization, potentially reducing experimentalist efforts for selecting candidates from days to minutes. Overall, this framework provides a generalizable template for functional annotation of entire enzyme families.

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

High-Fidelity Two-Step Image Generation via Teacher-Aligned End-to-End Distillation

Few-step diffusion distillation has become increasingly mature for 4-8-step generation, yet pushing further to 2 steps remains challenging. In this work, we introduce Z-Image Turbo++, a high-quality 2-step image generation model distilled from the 8-step Z-Image Turbo teacher. Our method addresses the central bottlenecks of increased task difficulty and limited model capacity in 2-step generation through three simple but effective design choices tailored to this regime. First, we propose Distribution-Aligned Adversarial Learning, which uses teacher-generated images rather than external real images as real samples for GAN training, providing a more attainable and informative adversarial target. Second, we adopt Step-Decoupled Parameterization, assigning independent model parameters to the two denoising steps to better match their distinct capacity demands. Third, we perform End-to-End Training with Iterative Regularization, allowing the first step to receive gradients from final image quality while preserving a meaningful intermediate generation through an explicit step-1 loss. Together, these designs substantially narrow the quality gap between 2-step and 8-step generation in both qualitative and quantitative evaluations, highlighting the potential of carefully tailored distillation strategies for improving the quality-efficiency trade-off in few-step generation.

13.
arXiv (CS.AI) 2026-06-19

UltraQuant: 4-bit KV Caching for Context-Heavy Agents

arXiv:2606.20474v1 Announce Type: cross Abstract: Context-heavy agents place unusual pressure on the key-value (KV) cache: long prefixes are reused across many short turns, while concurrency determines whether the serving system can keep GPUs utilized. We study 4-bit KV-cache compression for this setting, using TurboQuant-style rotation and codebook quantization as a quality anchor and vLLM FP8 KV caching as the deployment anchor. We report three contributions. First, we frame 4-bit KV caching around multi-round agent workloads where task quality, cache residency, and serving throughput must be measured jointly. Second, we describe the practical design choices needed to make the 4-bit path robust, including asymmetric K/V treatment, Walsh-Hadamard rotation, QJL removal, and block-scale variants. Third, we present serving optimizations on AMD GPUs, including optimized decode-attention kernels and UltraQuant, an FP4 approximation path that uses FP8 queries, FP4 KV tensors, UE8M0 group scales, and native scaled-MFMA support on CDNA4. On a long-context, multi-turn agentic workload, UltraQuant cuts P50 time-to-first-token by 3.47x in the cache-pressured late rounds (2.3x across all rounds) and raises output throughput by 1.63x over the FP8 KV baseline.

14.
arXiv (CS.CL) 2026-06-17

HistoRAG: Embedding Historical Methodology in Retrieval-Augmented Generation Through Critical Technical Practice

Retrieval-Augmented Generation (RAG) is the prevailing architecture for grounding language model outputs in external evidence, yet its dominant evaluation paradigms and default configurations remain oriented toward factual question-answering. For interpretive disciplines such as historical studies, RAG embeds assumptions that conflict with scholarly practice. We introduce HistoRAG, a framework that translates historiographical principles into concrete architectural interventions. Separated retrieval and generation decouples source discovery from interpretation, temporal windowing enforces balanced source representation across the research period as a methodological requirement of historical inquiry, and LLM-as-judge evaluation makes relevance judgments transparent and contestable. We evaluate these interventions using SPIEGELragged, applied to 102,189 articles from Der Spiegel (1950-1979). Each intervention addresses a measurable deficiency in standard RAG: era-specific vocabulary retrieves zero chunks from the 1950s when using 1970s terminology, evidence of the temporal skew that motivates windowing; vector similarity and LLM-assessed relevance correlate only weakly (Spearman rho = 0.275), motivating post-retrieval evaluation; and keyword-based and semantic retrieval surface largely disjoint source pools, motivating an architecture in which both operate as complementary retrieval layers under a shared LLM evaluation filter. We also introduce the concept of Zwischentexte (intermediate texts that function as interpretive proposals rather than findings) as a framework for responsible integration of LLM-generated text into scholarly practice. The architecture offers a model for how domain-specific epistemological commitments can be translated into RAG design decisions, and may transfer to other interpretive disciplines working with large corpora.

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

ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models

Multimodal large language models (MLLMs) may memorize sensitive cross-modal information during pretraining, making machine unlearning (MU) crucial. Existing methods typically evaluate unlearning effectiveness based on output deviations, while overlooking the generation quality after unlearning. This can easily lead to hallucinated or rigid responses, thereby affecting the usability and safety of the unlearned model. To address this issue, we propose ASRU, a controllable multimodal unlearning framework that incorporates generation quality as a core evaluation objective. ASRU first induces initial refusal behavior through activation redirection, and then optimizes fine-grained refusal boundaries using a customized reward function, thereby achieving a better trade-off between target knowledge unlearning and model utility. Experiments on Qwen3-VL show that ASRU significantly improves unlearning effectiveness (+24.6%) on average and generation quality (5.8X) on average while effectively preserving model utility, using only a small amount of retained supervision data.

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

Learning Upper Lower Value Envelopes to Shape Online RL: A Principled Approach

arXiv:2510.19528v2 Announce Type: replace-cross Abstract: We investigate the fundamental problem of leveraging offline data to accelerate online reinforcement learning - a direction with strong potential but limited theoretical grounding. Our study centers on how to learn and apply value envelopes within this context. To this end, we introduce a principled two-stage framework: the first stage uses offline data to derive upper and lower bounds on value functions, while the second incorporates these learned bounds into online algorithms. Our method extends prior work by decoupling the upper and lower bounds, enabling more flexible and tighter approximations. In contrast to approaches that rely on fixed shaping functions, our envelopes are data-driven and explicitly modeled as random variables, with a filtration argument ensuring independence across phases. The analysis establishes high-probability regret bounds determined by two interpretable quantities, thereby providing a formal bridge between offline pre-training and online fine-tuning. Empirical results on tabular MDPs demonstrate substantial regret reductions compared with both UCBVI and prior methods while remaining competitive with related approaches.

17.
PLOS Computational Biology 2026-06-02

PepAnno: A structure-aware deep learning framework for bioactive peptide prediction, structural visualization, and physicochemical profiling

作者:

by Enyan Liu, Yueming Hu, Liya Liu, Yifan Chen, Shilong Zhang, Sida Li, Haoyu Chao, Luyao Xie, Yi Shen, Liangwei Wu, Julio Raúl Fernández Massó, Ming Chen Peptides are gaining prominence as therapeutic candidates due to their diverse physiological functions and structural simplicity. Although multiple computational tools exist for bioactive peptide prediction, many suffer from limitations such as non-intuitive interfaces, sequence-only representations, insufficient structural awareness, restricted interpretability, or fragmented analysis workflows, leading to reduced research efficiency and higher costs. To address these challenges, we present PepAnno (https://bis.zju.edu.cn/pepanno/), a comprehensive and user-friendly web server for multi-functional peptide annotation. PepAnno is powered by a novel structure-aware, multi-view geometric deep learning framework that integrates pre-trained sequence embeddings with predicted 3D structural graphs through a dual-stream architecture combining a Transformer and a GATv2 network. A cross-modal attention mechanism is employed to effectively fuse semantic and geometric representations, enabling accurate multi-task prediction across 7 key bioactivities, including antimicrobial and anticancer properties. Comprehensive evaluation on seven curated bioactivity datasets demonstrates that PepAnno achieves robust and competitive predictive performance across tasks, consistently outperforming or matching existing methods in terms of discrimination and stability. Beyond functional prediction, PepAnno provides automated calculation of physicochemical properties, structure visualization, and access to an integrated repository of peptide-related databases and tools. By enabling one-click peptide annotation, PepAnno offers an efficient and interpretable solution for large-scale peptide analysis and facilitates downstream experimental design and peptide-based drug discovery.

18.
arXiv (CS.CV) 2026-06-18

From Bounding Boxes to Visual Reasoning: An On-Policy Data Annotation Tool for Vision-Language Models

Vision-language models (VLMs) are rapidly advancing toward sophisticated grounded structured visual reasoning. Training models for such advanced capabilities demands a new genre of data that seamlessly unifies spatial coordinates, open-vocabulary descriptions, structured attributes, and topological relationships into a singular representation. However, existing data annotation tools fundamentally fail to meet these intricate demands, suffering from three systematic bottlenecks: limited expressiveness, severe annotation-training decoupling, and poor data reusability. To bridge this infrastructure gap, we introduce an open-source annotation tool, ScreenAnnotator. First, we define a unified annotation atom schema that binds spatial, semantic, and structural primitives into a single unit. Second, we implement an on-policy annotation loop embedded with a Bayesian Annotation Verifier (BAV). Finally, we design a template-driven multi-task data synthesis process dynamically transforms static atoms into diverse multi-dimensional reasoning tasks, eliminating redundant re-annotation. The on-policy loop drives the annotation accept rate to nearly 100% on flowcharts and 77% on GUI screenshots, while steadily reducing per-image annotation time as labeled data accumulate. In the flowchart scenario, fine-tuning a VLM yields 76.1% average accuracy, which is a 35.1% point absolute gain. Our code is available at: https://github.com/WnQinm/Annotator.

19.
arXiv (quant-ph) 2026-06-16

Connecting entanglement growth with local integrals of motion in the disordered Fermi-Hubbard model

arXiv:2606.15481v1 Announce Type: new Abstract: Generically a quantum system initialized in an unentangled state will, under unitary dynamics, rapidly become entangled, a process closely related to information transport and to thermalization. Disorder can suppress the growth of entanglement and result in memory of initial conditions. In non-interacting systems this arises from localization of single-particle states, the occupancy of which is fixed by the initial condition. In interacting systems similar localized conserved quantities persist, but with the added feature that they are coupled, resulting in entanglement growth which is distinct from both non-interacting localized systems and from generic ergodic systems. The Fermi-Hubbard model has two degrees of freedom per site – charge and spin – and disorder may be present in both of these. We study the growth of entanglement in two scenarios – disorder in charge equal and unequal to that in spin, and determine the distinct contributions of charge and spin degrees of freedom by expanding the Hamiltonian in terms of a set of optimally localized conserved quantities with separate charge and spin character. We find that coupling between charge and spin is significantly weaker than charge-charge and spin-spin coupling. While this decoupling is present in all our results, it is only apparent when the strength of the disorder in the two sectors is different such that there is a separation between the characteristic timescales of the contributions to entanglement made by charge and by spin.

20.
arXiv (CS.CV) 2026-06-11

Natural-Language Temporal Grounding in Hour-Long Videos is a Search Problem: A Benchmark and Empirical Decomposition

Temporal grounding–returning the interval $[t_s, t_e]$ for a natural-language query over a video–is the language interface to long-form video, yet has been studied on short videos; the dynamics of hour-scale natural-language grounding remain underexplored. We take the position that at hour-scale, the binding constraint is search, not recognition: Video-LLMs are bottlenecked not by localizing a nearby event, but–given a natural-language query–by searching for the relevant region of a long video. To test this, we release ExtremeWhenBench, the first open hour-scale grounding benchmark (2,273 queries over 194 videos, mean 75.7 min, max 9 hr) with an open-form query distribution. Every open Video-LLM collapses while a frame-level retrieval baseline outperforms them; a failure taxonomy attributes 85% of failures to search; and a retrieve-then-ground hybrid recovers 6.7x over the monolithic Video-LLM–mirroring retrieve-then-read in open-domain QA.

21.
arXiv (CS.AI) 2026-06-12

Can I Buy Your KV Cache?

arXiv:2606.13361v1 Announce Type: new Abstract: Right now, across the world, AI agents are repeating the same absurd act: to read one document, they each recompute it from scratch. Every agent re-runs prefill, the most compute-intensive step a large model takes, over identical text, only to rebuild a key-value (KV) cache identical to the one the agent before it just built. The same answer, computed a million times. We make a proposal that is almost offensively simple: compute it once. Let a publisher precompute a document's KV cache, and let every other agent buy the right to load it and skip prefill. It works, and it is token-exact: loading a precomputed KV and continuing matches prefilling from scratch (24/24 greedy tokens, and at the logits level), with no accuracy cost. On Qwen3-4B, reuse is 9-50x cheaper in compute than prefill, and the gap widens with length (prefill's attention scales with L^2), so a single reuse already pays it back. Then the part that matters: where the KV lives. Shipping it fails, because KV is nearly incompressible, so per-load egress costs more than the prefill it saves. Hosting it provider-side, exactly as production prompt-caching works, removes egress entirely. The size of the prize is set by our measured compute saving: serving one hot 3774-token document to 80M agents costs ~$1.5M to re-prefill but only ~$0.03M of reuse compute (49.7x less). The 0.1x cache-read tariff APIs charge passes a 10x discount to users while sitting inside this measured envelope, so the 10x is a floor that the measured ~50x compute saving clears, and the gap to the physical ~50x is provider margin: millions of dollars per popular document. We frame the resulting agent-native prefill CDN and leave lossless KV compression and a cross-party payment layer as the open problems.

22.
arXiv (CS.AI) 2026-06-18

Sparsity Curse: Understanding RLVR Model Parameter Space from Model Merging

arXiv:2606.18521v1 Announce Type: cross Abstract: Reinforcement Learning with Verifiable Reward (RLVR) has emerged as a powerful post-training paradigm that surpasses Supervised Fine-Tuning (SFT) in eliciting reasoning intelligence and resisting catastrophic forgetting. Recent studies further reveal that RLVR induces highly sparse and off-principal parameter updates compared to SFT. This naturally raises the question: does such sparsity make RLVR models more amenable to model merging? If so, model merging would offer a scalable, training-free path to aggregate diverse reasoning capabilities from independently trained RLVR models. Surprisingly, we find the opposite, uncovering a sparsity curse: the sparse RLVR updates are spread farther apart in parameter space, forming near-orthogonal shortcuts that make aggregation inherently fragile. This is likely rooted in the stochasticity of RL optimization and the diversity of emergent reasoning patterns. Unlike SFT models that converge to shared, flat basins and merge naturally, RLVR models suffer severe degradation under standard merging methods. Through systematic empirical analysis of the update geometry, we characterize the mechanisms behind this failure and propose Sensitivity-aware Resolving Merging (SAR-Merging), a merging recipe tailored for the unique structure of RLVR parameter spaces. SAR-Merging resolves conflicts in overlapping update regions via Fisher Information-based sensitivity arbitration, followed by magnitude-aware sparsification and rescaling to preserve fragile reasoning pathways. Experiments on mathematical and coding benchmarks demonstrate that SAR-Merging substantially outperforms existing merging methods on RLVR models, enabling both single-task enhancement and multi-capability fusion.

23.
arXiv (CS.CV) 2026-06-15

Prompt2Effect: Training-Free Image-to-Video Model Specialization via LoRA Generation

Personalizing Image-to-Video (I2V) diffusion models with specific visual effects is increasingly demanded for high-end video generation. Current practice requires training a separate Low-Rank Adaptation (LoRA) module for each effect, incurring substantial data curation and iterative optimization costs that hinder interactive control. We present Prompt2Effect, a weight-driven hypernetwork that amortizes per-effect training by directly synthesizing effect-specific LoRA weights in a single forward pass. Unlike prior hypernetworks that regress adapter weights purely from semantics, Prompt2Effect is explicitly conditioned on the frozen base model weights, grounding weight prediction in the structural geometry of each layer. Furthermore, instead of predicting raw LoRA matrices, we introduce an SVD-canonicalized parameterization that resolves factorization ambiguity and stabilizes large-scale weight synthesis. Together, these design principles enable accurate and scalable LoRA prediction for high-dimensional I2V diffusion models. Extensive experiments demonstrate that Prompt2Effect achieves on-par or superior video quality and effect alignment compared to conventional LoRA fine-tuning, while reducing the computational cost from 56 GPU training hours to 3.3 seconds of hypernetwork inference. When used as initialization for subsequent fine-tuning, our predicted weights further improve final performance and accelerate optimization by approximately 10x.

24.
arXiv (CS.LG) 2026-06-18

Protein-Based Fish Species Identification: Dataset, Models, and Insights from Native Bangladeshi Fish

arXiv:2606.18302v1 Announce Type: cross Abstract: Correct identification of fish species is highly significant for food security, economic development, and climate resilience in Bangladesh. Protein sequences directly reflect functional and evolutionary constraints which are important for species authentication and biodiversity monitoring. Yet there exists no benchmark for native Bangladeshi fish species identification from protein sequence. In this study, we addressed this gap by introducing the first curated dataset for nine native Bangladeshi fish species of 2845 high quality protein sequences. We also established the first protein sequence classification baseline for this domain through a systematic benchmarking of seven architectural paradigms. Moreover, we propose a realistic deployable novel hybrid architecture of MotifCNN and Transformer with Terminal-Aware Positional-Encoding (MotifCNN-Transformer+TA-PE). Our novel architecture achieves 79.80% accuracy with macro-F1 of 0.80. The highest 83.04% accuracy is achieved by finetuned protein language model ProtBERT that has 420M parameters and requires dual 16GB GPUs for inference. According to McNemar's test, ProtBERT's 3.24% accuracy gain over our MotifCNN-Transformer+TA-PE is statistically insignificant (p = 0.1120). Our novel architecture beats it among six of the nine classes in per class identification. Also our MotifCNN-Transformer+TA-PE is approximately 5x faster, 42x smaller, and supports 16x larger batch size than ProtBERT and has GPU free inference, making it more practical for deployment in resources constrained areas such as rural Bangladesh. Beyond this, our foundational work shows effects of phylogenetic relationships on sequence similarity and establishes pathways for fisheries management, food authentication and biodiversity conservation in South Asia's protein dependent economy.

25.
arXiv (CS.AI) 2026-06-11

WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces

arXiv:2606.09426v2 Announce Type: replace Abstract: Computer-use agents (CUAs) increasingly operate in runtimes that combine visual desktop control, command-line execution, code editing, browsers, and external tools. Existing benchmarks, however, often evaluate these interfaces as separable capabilities, leaving long-horizon cross-interface orchestration under-tested. Thus, we introduce WeaveBench, a long-horizon hybrid-interface benchmark with 114 tasks across 8 real-world work domains, grounded in real user requests and publicly verifiable artifacts. Each task requires agents to combine GUI observations/actions with CLI/code operations within a single trajectory. We evaluate these tasks on a real Ubuntu desktop inside deployed CLI-agent runtimes, augmented with a minimal desktop-control plugin. We also propose a companion trajectory-aware judge that inspects deliverables, files, screenshots, logs, and action traces, while detecting shortcut behaviors such as fabricated visual evidence or hard-coded metrics. Across frontier model-runtime pairings, the best PassRate reaches only 41.2%, showing the benchmark remains far from saturated. The trajectory-aware judge further reveals that outcome-only grading substantially overestimates agent performance. Overall, WeaveBench exposes a critical gap in CUA evaluation and provides an effective testbed to measure whether agents can orchestrate GUI, CLI, and code operations across long-horizon real-world tasks.