arXiv:2607.14418v1 Announce Type: new
Abstract: Ad-load design is a central supply-side decision in sponsored search: more sponsored slots can raise revenue, but may crowd out organic results and degrade user outcomes. We study this trade-off using a large-scale randomized field experiment on an Android app store, where over five million users are exposed to one through six sponsored slots. Increasing ad load raises revenue by up to 43%, but reduces total search conversions by up to 5% and daily engagement by up to 2.2%. These average effects mask substantial heterogeneity: additional slots generate large revenue gains for high-ad-conversion queries, but little or negative marginal revenue for low-conversion queries. The trade-off also shifts within query as advertiser composition changes, such as brand-advertiser presence. Motivated by these findings, we design and deploy a novel adaptive algorithm -- exploration-augmented Locally Adaptive Ad Load (e-LAAL). e-LAAL combines LAAL, a model-free query-level decision rule that updates ad-load recommendations using recent outcomes, with static exploration arms that maintain support and provide fixed-policy counterfactual benchmarks. We provide a finite-time dynamic-regret guarantee for the e-LAAL architecture. In a platform-level production deployment serving 22.3 million users and 77.6 million searches, e-LAAL improves the empirical revenue--conversion trade-off relative to deployed static benchmarks and outperforms uniform and historical query-dependent static benchmarks.
Science Journals
arXiv:2607.14429v1 Announce Type: new
Abstract: This paper proposes twelve candidate Experience-Led Commerce design principles for high-constraint, relational fashion e-commerce, surfaced through design-led induction while building VogueDrop, a multi-vendor prototype. The principles address multi-entry discovery, experience continuity, relational exploration, preference sovereignty, evidence-scoped correspondence, recommendation-time feasibility, customer-compatible commercial ranking, adaptive but stable workspaces, attributable transaction authority, outcome-linked learning, shared composition authoring, and accountable human to agent handoff. The paper uses fashion as an intentionally bounded domain in which fit, composition, material, identity-sensitive preference, seller fragmentation, visual correspondence, and delivery timing make the interaction breakdown observable; it does not claim universal applicability across e-commerce. Each candidate principle is paired with a prespecified hypothesis, primary behavioural outcome, and rejection condition. A formative critical-incident study and a preregistered matched-interface experiment are specified, with user-experience and platform-facing outcomes reported separately. No empirical superiority is claimed before those studies are completed.
arXiv:2508.15777v3 Announce Type: replace
Abstract: While color harmony has long been studied in art and design, a clear consensus remains elusive, as most models are grounded in qualitative insights or limited datasets. In this work, we present a quantitative, data-driven study of color pairing preferences using controlled hue-based palettes in the HSL color space. Participants evaluated combinations of thirteen distinct hues, enabling us to construct a preference matrix and define a combinability index for each color. Our results reveal that preferences are highly hue dependent, thereby challenging classical color harmony theories. Yet, when averaged over hues, statistically meaningful patterns of aesthetic preference emerge, with certain hue separations perceived as more harmonious. Strikingly, these patterns align with hue distributions found in natural landscapes, pointing to a statistical correspondence between human color preferences and the structure of color in nature. Finally, we analyze our color-pairing score matrix through principal component analysis, which uncovers two complementary hue groups whose interplay underlies the global structure of color-pairing preferences. Together, these findings offer a quantitative framework for studying color harmony and its potential perceptual and ecological underpinnings.
arXiv:2509.00761v4 Announce Type: replace
Abstract: Large language models are increasingly deployed for legal question answering, where evaluations typically focus on multiple-choice accuracy. This measure overlooks a common failure: whether the citation source attached to an answer exists and supports the rule the system attributes to it. We present L-MARS, an open multi-agent legal QA system with agentic search and judge-driven evidence checks, and audit it claim by claim against its cited source. Each atomic claim is labelled with a six-class taxonomy and scored with strict-ALCE under cross-provider judging, where the answerer and verifier come from different model families. On a stratified 100-question Bar Exam audit, retrieval barely moves accuracy, yet the multi-turn judge loop lifts strict citation F1 from 0.13 (naive RAG) to 0.25 and cuts the no-citation rate from 34% to 13%. We further introduce Faith-Search, a post-draft step that re-verifies and repairs unreachable citations; it drops the unreachable rate below 1% but does not improve F1 over the multi-turn loop, so we report it as a targeted reachability intervention rather than a faithfulness breakthrough. A 50-question LegalSearchQA case study confirms the picture: retrieve-then-draft pipelines saturate near 0.75 citation F1, while a single-agent web-search baseline collapses to 0.22 under external audit.
arXiv:2510.09416v4 Announce Type: replace
Abstract: Learning on temporal graphs has become a central topic in graph representation learning, with numerous benchmarks indicating the strong performance of state-of-the-art models. However, recent work has raised concerns about the reliability of benchmark results, noting issues with commonly used evaluation protocols and the surprising competitiveness of simple heuristics. This contrast raises the question of which characteristics of the underlying graphs temporal graph learning models actually use to form their predictions. We address this by systematically evaluating eight models on their ability to capture eight fundamental characteristics related to the link structure of temporal graphs. These include structural characteristics such as density, temporal patterns such as recency, and edge formation mechanisms such as homophily. Using both synthetic and real-world datasets, we analyze how well models learn these characteristics. Our findings reveal a mixed picture: models capture some characteristics well but fail to reproduce others. With this, we expose important limitations. Overall, we believe that our results provide practical insights for the application of temporal graph learning models and motivate more interpretability-driven evaluations in graph learning research.
arXiv:2510.14854v4 Announce Type: replace
Abstract: Magnetic induction (MI) communication (MIC) has emerged as a promising candidate for underground communication networks due to its excellent penetration capabilities. Integration with Space-Air-Ground-Underground (SAGUI) networks in next-generation mobile communication systems requires a well-defined network architecture. A recent discovery in MIC research, MI fast fading, remains in its early stages and presents unique challenges. This paper provides a comprehensive survey on through-the-earth (TTE) MIC, covering MI applications, channel modeling, point-to-point MIC design, relay techniques, network frameworks, and emerging technologies. We compare various MIC applications to highlight TTE-specific challenges and review the principles of channel modeling, addressing both MI slow fading and MI fast fading, along with its potential impact on existing MIC theories. We conduct a fine-grained decomposition of MI channel power gain into four distinct physical parameters, and propose a novel geometric model to analyze MI fast fading. We also summarize MI relay techniques, examine crosstalk effects in relay and high-density networks, and explore key research tasks within the OSI framework for a holistic MI network protocol in SAGUI. To bridge the gaps identified, we propose a MIC framework that supports TCP/IP and Linux, enabling full implementation of existing and emerging MIC solutions. This framework empowers researchers to leverage Linux resources and deep learning platforms for accelerated development of MIC in SAGUI networks. Remaining research challenges, open issues, and promising novel techniques are further identified to advance MIC research.
arXiv:2510.21324v2 Announce Type: replace
Abstract: Chest X-ray (CXR) plays a pivotal role in clinical diagnosis, and a variety of task-specific and foundation models have been developed for automatic CXR interpretation. However, these models often struggle to adapt to new diagnostic tasks and complex reasoning scenarios. Recently, LLM-based agent models have emerged as a promising paradigm for CXR analysis, enhancing model's capability through tool coordination, multi-step reasoning, and team collaboration, etc. However, existing agents often rely on a single diagnostic pipeline and lack mechanisms for assessing tools' reliability, limiting their adaptability and credibility. To this end, we propose CXRAgent, a director-orchestrated, multi-stage agent for CXR interpretation, where a central director coordinates the following stages: (1) Tool Invocation: The agent strategically orchestrates a set of CXR-analysis tools, with outputs normalized and verified by the Evidence-driven Validator (EDV), which grounds diagnostic outputs with visual evidence to support reliable downstream diagnosis; (2) Diagnostic Planning: Guided by task requirements and intermediate findings, the agent formulates a targeted diagnostic plan. It then assembles an expert team accordingly, defining member roles and coordinating their interactions to enable adaptive and collaborative reasoning; (3) Collaborative Decision-making: The agent integrates insights from the expert team with accumulated contextual memories, synthesizing them into an evidence-backed diagnostic conclusion. Experiments on various CXR interpretation tasks show that CXRAgent delivers strong performance, providing visual evidence and generalizes well to clinical tasks of different complexity. Code and data are valuable at this \href{https://github.com/laojiahuo2003/CXRAgent/}{link}.
arXiv:2511.15912v2 Announce Type: replace
Abstract: The Nab (Neutron a b) experiment is designed to measure the beta-antineutrino angular correlation in free neutron $\beta$ decay with an ultimate precision goal of 0.1%, providing input for tests of Cabibbo-Kobayashi-Maskawa (CKM) matrix unitarity. This measurement is performed via detection of electrons and protons in delayed coincidence using custom large-area segmented silicon detectors. We present the characterization of one such detector system to establish the proton energy and timing response, using a dedicated proton accelerator. The detected proton peak was studied for 25 keV, 30 keV, and 35 keV incident protons on a set of detector segments and multiple cooling cycles over a one year period. Ionization losses were consistent with models of the detector dead layer with thicknesses less than 100 nm. The detected proton peak was stable within the uncertainty from energy calibration (0.25 keV). The rise times of detector pulses from $^{109}$Cd and $^{113}$Sn conversion electron sources were used to extract the impurity density profile and establish a precise model for the detector timing response. The observed impurity density profile varied from $(2 \pm 2) \times 10^9$ cm$^{-3}$ at the center to $(26 \pm 2) \times 10^9$ cm$^{-3}$ at the edge. This impurity density profile was then used to characterize systematic effects in proton time-of-flight measurements due to detector pulse-shape effects; the resultant proton timing systematic uncertainties were below 0.3 ns, which is sufficient for the Nab experiment.
arXiv:2607.14510v1 Announce Type: new
Abstract: Industrial time series serve as the foundation for Prognostics and Health Management (PHM) to ensure the reliability and safety of industrial equipment such as aero-engines. However, existing approaches are typically limited to single-modality modeling, which restricts their generalization in complex scenarios. Although recent advances in large language models (LLMs) provide new opportunities for multimodal learning, bridging continuous time-series signals and discrete textual semantics remains an open challenge. To this end, we propose VLT, a multimodal foundation model that jointly models time-series, frequency-spectrum visual representations, and textual knowledge. A key insight is to utilize the frequency spectrum as a visual bridge to connect continuous temporal signals with discrete semantics. Specifically, a Time-aware Mixture-of-Experts (Time-MoE) is designed to capture heterogeneous temporal dynamics, while a Frequency-Text Augmented Learner enables joint modeling of spectral and semantic features within a shared representation space. Furthermore, a time-centric gradient alignment mechanism is introduced to mitigate cross-modal optimization conflicts via gradient normalization and reliability-aware dynamic reweighting. Extensive experiments on multiple industrial datasets demonstrate that VLT outperforms state-of-the-art methods, achieving superior robustness and generalization under few-shot, noisy, and incomplete-modality settings.
arXiv:2511.21352v4 Announce Type: replace
Abstract: As computational resources continue to increase, the storage and analysis of vast amounts of data will inevitably become a bottleneck in computational fluid dynamics (CFD) and related fields. Although compression algorithms and efficient data formats can mitigate this issue, they are often insufficient when post-processing large amounts of volume data. Processing such data may require additional high-performance software and resources, or it may restrict the analysis to shorter time series or smaller regions of interest. The present work proposes an improved version of the existing \emph{Sparse Spatial Sampling} algorithm ($S^3$) to reduce the data from time-dependent flow simulations. The $S^3$ algorithm iteratively generates a time-invariant octree grid based on a user-defined metric, efficiently down-sampling the data while aiming to preserve as much of the metric as possible. Using the sampled grid allows for more efficient post-processing and enables memory-intensive tasks, such as computing the modal decomposition of flow snapshots. The enhanced version of $S^3$ is tested and evaluated on the scale-resolving simulations of the flow past a tandem configuration of airfoils in the transonic regime, the incompressible turbulent flow past a circular cylinder, and the flow around an aircraft half-model at high Reynolds and Mach numbers. $S^3$ significantly reduces the number of mesh cells by $35 \%$ to $95\%$ for all test cases while accurately preserving the dominant flow dynamics, enabling post-processing of CFD data on a local workstation rather than HPC resources for many cases.
arXiv:2607.15192v1 Announce Type: new
Abstract: In this paper, we construct an efficient higher-order local time integration scheme for spatially discretized linear Friedrichs' systems. In particular, our interest is in problems where only a few of the mesh elements are small while the majority of the elements is much larger. The special combination of two methods like the leapfrog method on the coarse part of the mesh and the Crank-Nicolson method on the fine part as was done in Hochbruck, Sturm 2016 and Hochbruck, K\"ohler 2022 is not suitable for higher-order time integration. Therefore, we suggest to approximate the solution of the linear systems arising in each time step by a preconditioned Krylov subspace method, e.g., the quasi-minimal residual method by Freund and Nachtigal 1991. The techniques developed here for linear problems also carry over to nonlinear problems, where linear systems of the same type arise within a Newton-type iteration.
Motivated by the analysis of locally implicit methods by Hochbruck and Sturm 2016, we show how to construct a preconditioner in such a way that the number of iterations required by the Krylov subspace method to achieve a certain accuracy is bounded independently of the diameter of the small mesh elements. We prove this behavior by using Faber polynomials and complex approximation theory.
The cost to apply the preconditioner consists of the solution of a small linear system, whose dimension corresponds to the degrees of freedom within the fine part of the mesh (and its next coarse neighbors). If this dimension is small compared to the size of the full mesh, the preconditioner is very efficient.
We conclude by verifying our theoretical results with numerical experiments for the fourth-order Gauss-Legendre Runge--Kutta method.
arXiv:2607.15200v1 Announce Type: new
Abstract: Reinforcement learning has proven effective for improving reasoning in large language models, but extending it to Masked Diffusion Language Models (MDLMs) remains challenging due to the intractability of the log-likelihood estimation. Existing approaches approximate this log-likelihood by modeling only the token predictions, ignoring the order in which positions are unmasked during generation. We observe that MDLM generation involves two decisions at each step: what tokens to place at each masked position and which positions to remask. We formalize this as a two-stage action MDP, showing that the policy gradient naturally decomposes into a token term and a masking term. Combining optimization of both terms leads to state-of-the-art outcomes on mathematical reasoning and coding benchmarks, with scores of 87.1% on GSM8K and 53.4% on MBPP.
arXiv:2512.12427v2 Announce Type: replace
Abstract: Many aerial tasks involving quadrotors demand both instant reactivity and long-horizon planning for obstacle avoidance, energy efficiency, or trajectory tracking. High-fidelity models enable accurate control but are too slow for long horizons. Low-fidelity planners scale but cannot directly control the system, necessitating cascaded architectures. Prevailing hierarchical approaches plan with a simplified model and use a high-fidelity controller for tracking, yet this decomposition is inherently suboptimal. The controller is limited by the coarse plan, and conventional MPC alternatives shorten the horizon to stay real-time feasible. We present UNIQUE, an MPC architecture that replaces this hierarchical stacking with temporal cascading. The planning problem is formulated as the second-tail horizon of a single multi-phase MPC, rather than being solved separately. We align costs across horizons, derive feasibility constraints for the point-mass planning model, and introduce transition constraints that convert high-fidelity states into meaningful low-fidelity states. Parallel point-mass and mixed-integer solvers address nonconvexities while incorporating progressive 3D obstacle smoothing over the planning horizon. In simulations and real flights, under equal computational budgets, UNIQUE improves closed-loop tracking by up to 75% compared with standard MPC and hierarchical baselines. Ablations and Pareto analyses confirm performance gains across variations in horizon, constraint approximations, and smoothing schedules.
arXiv:2512.13993v3 Announce Type: replace
Abstract: Discretized versions of optimization problems over continuous arguments are routinely solved at a single fine resolution, incurring a per-iteration cost that grows, often superlinearly, with the number of grid points. This paper analyzes a multiscale method that instead solves a hierarchy of increasingly fine dyadic discretizations. Linear interpolation of each coarse solution warm starts the next finer scale using any q-linearly convergent update rule as the inner solver. Each coarse problem is a consistent discretization of the continuous problem. Structural properties such as convexity and smoothness are preserved. For problems with Lipschitz-continuous solutions, two variants of the method converge to the fine-scale solution with explicit error bounds. The fine-scale solution in turn approximates the continuous solution once the grid is sufficiently fine, with quantified constants. The total cost to reach a fixed accuracy is provably lower than that of single-scale optimization whenever the cost of one update grows at least linearly in the problem size. Numerical experiments on probability density demixing problems, including geological survey data, show four- to sevenfold speedups while using a fraction of the memory.
arXiv:2512.14008v2 Announce Type: replace
Abstract: Masked Discrete Diffusion Models (MDMs) have achieved strong performance across a wide range of multimodal tasks, including image understanding, generation, and editing. However, their inference speed remains suboptimal due to the need to repeatedly process redundant masked tokens at every sampling step. In this work, we propose Sparse-LaViDa, a novel modeling framework that dynamically truncates unnecessary masked tokens at each inference step to accelerate MDM sampling. To preserve generation quality, we introduce specialized register tokens that serve as compact representations for the truncated tokens. Furthermore, to ensure consistency between training and inference, we design a specialized attention mask that faithfully matches the truncated sampling procedure during training. Built upon the state-of-the-art unified MDM LaViDa-O, Sparse-LaViDa achieves up to a 2x speedup across diverse tasks including text-to-image generation, image editing, and mathematical reasoning, while maintaining generation quality.
arXiv:2601.09243v2 Announce Type: replace
Abstract: Gaussian Splatting has emerged as a powerful representation for high-quality, real-time 3D scene rendering. While recent works extend Gaussians with learnable textures to enrich visual appearance, existing approaches allocate a fixed square texture per primitive, leading to inefficient memory usage and limited adaptability to scene variability. In this paper, we introduce adaptive anisotropic textured Gaussians (A$^2$TG), a novel representation that generalizes textured Gaussians by equipping each primitive with an anisotropic texture. Our method employs a gradient-guided adaptive rule to jointly determine texture resolution and aspect ratio, enabling non-uniform, detail-aware allocation that aligns with the anisotropic nature of Gaussian splats. This design significantly improves texture efficiency, reducing memory consumption while enhancing image quality. Experiments on multiple benchmark datasets demonstrate that A TG consistently outperforms fixed-texture Gaussian Splatting methods, achieving comparable rendering fidelity with substantially lower memory requirements. Project page: http://github.com/Rickyeeeeee/A2TG.
arXiv:2601.17471v2 Announce Type: replace
Abstract: Continuous fuzzing platforms such as OSS-Fuzz uncover large numbers of vulnerabilities, yet the subsequent repair process remains largely manual. Unfortunately, existing Automated Vulnerability Repair (AVR) techniques -- including recent LLM-based systems -- are not directly applicable to continuous fuzzing. This is because these systems are designed and evaluated on a static, single-run benchmark setting, making them ill-suited for the diverse, noisy, and failure-prone environments in continuous fuzzing.
To address these issues, we introduce PatchIsland, a system for Continuous Vulnerability Repair (CVR) that tightly integrates with continuous fuzzing pipelines. PatchIsland employs an ensemble of diverse LLM agents. By leveraging multiple LLM agents, PatchIsland can cover a wider range of settings (e.g., different projects, bug types, and programming languages) and also improve operational robustness. In addition, PatchIsland utilizes a two-phase patch-based deduplication to mitigate duplicate crashes and patches, which can be problematic in continuous fuzzing.
In our internal evaluation, PatchIsland repaired 84 of 92 vulnerabilities, demonstrating strong repair capability. In the official AIxCC competition, the system operated with no human intervention in a fully autonomous environment and successfully patched 31 out of 43 vulnerabilities, achieving a repair rate of 72.1\%.
arXiv:2607.15033v1 Announce Type: new
Abstract: Neural video coding has advanced rapidly, achieving competitive compression performance while also enabling real-time coding speed. Yet, existing codecs exhibit severe rigidity when deployed in dynamic environments, failing to adapt to different video content, user requirements, and quality preferences. First, to meet the real-time constraint, they discard explicit motion estimation and motion compression, thereby losing the ability to adapt temporal prediction to motion complexity and bitrate constraints. Second, their spatial bit allocation strategy is coarse and, once trained, is fixed. It cannot adapt to dynamic user requirements at test time, preventing users from freely controlling the spatial distribution of bits. Third, they cannot adapt their quality preference to varying application requirements without deploying separate models. We address all three limitations within a single real-time neural video codec--URVC, transforming a rigid system into a unified framework with temporal, spatial, and perceptual adaptivity. First, we propose a rate-aware adaptive temporal prediction method that generates diverse prediction candidates through a multi-candidate architecture and couples candidate selection directly to rate-distortion optimization. Second, we propose a decomposition-based spatial rate control method that achieves finer-grained spatial bit allocation through feature decomposition and separate quantization, and allows users to perform direct spatial rate control at test time without retraining. Third, we propose a perceptual switching method that only requires learning a secondary module bank alongside a frame generator, enabling a codec to switch between signal fidelity and perceptual quality modes.
arXiv:2607.14581v1 Announce Type: new
Abstract: Recent advances in large language models (LLMs) and their extension to vision-language models (VLMs) have made it easier to combine text and images for tasks such as report generation. Existing VLMs in medicine typically focus on 2D images (chest X-rays), and their extension to 3D imaging has been difficult because of the lack of paired 3D imaging-text data. Thus, we introduce a new method for creating a 3D image-text dataset for brain oncology using 3D MRI scans of glioma and meningioma cases. We use a cooperative system in which several LLMs work together to generate and check reports, ensuring that they are accurate and clear. By leveraging the new 3D MRI-text dataset, we further build a VLM that converts MRI scans into tokens and aligns them with text instructions. Our VLM performed better in report generation and visual question answering tasks than other 2D and 3D methods. Our method not only improves the quality of reports but also helps with better diagnosis and treatment in brain oncology.
arXiv:2602.15280v2 Announce Type: replace
Abstract: Combining conversational AI with refreshable tactile displays (RTDs) offers significant potential for creating accessible data visualization for people who are blind or have low vision (BLV). To support researchers and developers building accessible data visualizations with RTDs, we present a multimodal data interaction architecture along with an open-source reference implementation. Our system is the first to combine touch input with a conversational agent on an RTD, enabling deictic queries that fuse touch context with spoken language, such as "what is the trend between these points?" The architecture addresses key technical challenges, including touch sensing on RTDs, visual-to-tactile encoding, integrating touch context with conversational AI, and synchronizing multimodal output. Our contributions are twofold: (1) a technical architecture integrating RTD hardware, external touch sensing, and conversational AI to enable multimodal data interaction; and (2) an open-source reference implementation demonstrating its feasibility. This work provides a technical foundation to support future research in multimodal accessible data visualization.
arXiv:2607.14586v1 Announce Type: new
Abstract: In goal-directed embodied navigation, where an agent must locate a specified target in an unseen environment, 3D scene understanding and navigation reasoning must work in concert. Current approaches transmit 3D scene information to vision-language models (VLMs) through text, suggesting a representation gap in our tested configurations; a controlled ablation confirms that direct embedding-level transfer significantly outperforms the evaluated text serialization formats. We introduce SoftNav, which injects entity-level 3D continuous representations -- one token per detected object or frontier -- into a VLM's hidden space as soft tokens through a lightweight projector. With the 3D encoder and VLM frozen, only ~1,200 samples and ~17M trainable parameters are needed. On HM3D-OVON, SoftNav achieves 74.2%/68.3%/66.7% SR across three splits, surpassing all prior methods in both SR and SPL; the same navigation policy transfers zero-shot to GOAT-Bench (67.2% SR), SG3D (47.2% s-SR), and real-world robot deployment without retraining or architectural modification. Injecting 3D scene tokens directly into VLMs bridges the representation gap, enabling transferable navigation with minimal training.
arXiv:2602.16312v4 Announce Type: replace
Abstract: This work presents a novel agglomeration-based multilevel preconditioner designed to accelerate the convergence of iterative solvers for linear systems arising from the discontinuous Galerkin discretization of the monodomain model in cardiac electrophysiology. The proposed approach exploits general polytopic grids at coarser levels, obtained through the agglomeration of elements from an initial, potentially fine, mesh. By leveraging a robust and efficient agglomeration strategy, we construct a nested hierarchy of grids suitable for multilevel solver frameworks. The effectiveness and performance of the methodology are assessed through a series of numerical experiments on two- and three-dimensional domains, involving different ionic models and realistic unstructured geometries. The results demonstrate strong solver effectiveness and favorable scalability with respect to both the polynomial degree of the discretization and the number of levels selected in the multigrid preconditioner.
arXiv:2603.01140v2 Announce Type: replace
Abstract: Conventional image denoising models often inadvertently learn spurious correlations between environmental factors and noise patterns. Moreover, due to high-frequency ambiguity, they struggle to reliably distinguish subtle textures from stochastic noise, resulting in over-removed details or residual noise artifacts. We therefore revisit denoising via causal intervention, arguing that purely correlational fitting entangles intrinsic content with extrinsic noise, which directly degrades robustness under distribution shifts. Motivated by this, we propose the Teacher-Guided Causal Disentanglement Network (TCD-Net), which explicitly decomposes the generative mechanism via structured interventions on feature spaces within a Vision Transformer framework. Specifically, our method integrates three key components: (1) An Environmental Bias Adjustment (EBA) module projects features into a stable, de-centered subspace to suppress global environmental bias (de-confounding). (2) A dual-branch disentanglement head employs an orthogonality constraint to force a strict separation between content and noise representations, preventing information leakage. (3) To resolve structural ambiguity, we leverage Nano Banana Pro, Google's reasoning-guided AI image generation model, to guide a causal prior, effectively pulling content representations back onto the natural-image manifold. Extensive experiments demonstrate that TCD-Net outperforms mainstream methods across multiple benchmarks in both fidelity and efficiency, achieving a real-time speed of 104.2 FPS on a single RTX 5090 GPU.
arXiv:2607.14618v1 Announce Type: new
Abstract: CPUs are the most universal target for on-device LLM inference, but existing low-bit quantization methods offer either coarse operating points or fine-grained mixed precision that is difficult to execute efficiently on CPUs. We present PolyQ, a CPU-oriented compiler/quantization co-design for activation-aware channel-wise bit allocation under a user-specified average-bit budget. PolyQ assigns per-channel bit-widths from $\{2,3,4,8,16\}$, then uses a compile-time model compiler to permute and cluster channels into bit-homogeneous blocks, generate SIMD- and LUT-compatible kernels, and merge compatible permutations across operators to keep layout regularization off the runtime path. This turns fine-grained budget fitting into a practical fractional-bit deployment method for CPU-only inference. Across Falcon-H1-3B, Llama2-13B, and Qwen3-32B on WikiText-2, PolyQ provides stable quality scaling from 3--6\,b and improves perplexity by 2.4--32.1\% over prior methods at a 3\,b target. End-to-end measurements on three representative CPUs -- workstation, laptop, and mobile -- show that compiler layout regularization reduces activation reorder traffic by up to 70.8\%, prefill latency and decode throughput scale nearly proportionally with the configured bit budget, and energy/token overhead stays below 2\% relative to an optimized LUT-based back-end. These results show that fractional-bit CPU deployment is practical, predictable, and energy-efficient across diverse edge targets.
arXiv:2607.14630v1 Announce Type: new
Abstract: Layer-wise post-training quantization of large language models minimizes each layer's reconstruction error in isolation, allowing quantization errors to accumulate across depth and causing severe degradation in extreme low-bit regimes. We formulate quantization as a joint optimization over the discrete codes and scales of all layers, driven by two mechanisms: (i) cross-layer error compensation, which maintains the network-level accumulated error through the recursion e_{l+1} = A_l e_l + q_l, with a propagation operator A_l derived from the layer's input differential and a local quantization residual q_l evaluated at teacher features; and (ii) finite-sample feature-statistics matching, which aligns means, projected covariances, and centered empirical kernels between the full-precision and quantized networks under relative normalization. We prove that instantiating the propagation operator as a finite difference of the quantized network makes the recursion exact for arbitrary nonlinear layers, enabling an efficient forward-difference implementation. Binary weights are optimized via a mirror-descent parameterization u = tanh(beta*z) with annealed inverse temperature and group-wise log-scales. On Qwen2.5-1.5B with 1.125-bit group-binary weights, error compensation alone reaches a perplexity ratio of 9.56 +/- 0.15 over the FP16 teacher, outperforming logit distillation (14.09 +/- 0.53; 32 percent relative, more than 8 sigma over 3 seeds) and layer-local reconstruction by two orders of magnitude. The same objective transfers unchanged to 4-bit quantization (1.060 vs. 1.088 for layer-local). Out-of-domain evaluations (C4, CNN/DailyMail) show the advantage of error compensation grows off-domain, while statistics matching keeps feature-statistics discrepancy low off-domain (0.42-0.88 vs. 1.41-2.99 without it), revealing a complementary division of labor between the two mechanisms.