arXiv:2606.09711v1 Announce Type: new Abstract: Reward hacking is usually studied after it becomes visible, once a model earns high proxy reward while failing the intended task. We instead study what proxy RL teaches before that failure appears. We introduce Proxy Reward Internalization and Mechanistic Exploitation (PRIME), a learned capability to assess task correctness, predict proxy acceptance, and reason about exploitable proxy--gold gaps. In coding RL environments with exploitable pytest rewards, we measure PRIME through chain-of-thought monitoring, direct probes, and activation-level concept vectors. We find that PRIME emerges in a staged sequence before sustained reward hacking, and that its current direct-probe score forecasts later hack onset and severity even when the visible hack rate is still low. PRIME also adapts when the evaluator changes, retargeting to whichever proxy--gold gap remains rewarded and persisting when gold reward suppresses overt hacking, and ablating its activation directions reduces hacking. Across checkpoints, in-domain PRIME tracks out-of-domain misalignment. Together these results suggest that exploitable proxy RL amplifies a proxy-internalization capability upstream of visible hacking, making PRIME a candidate early-warning signal for broader alignment risk.
Science Journals
arXiv:2606.09730v1 Announce Type: new Abstract: Large language models are increasingly expected to handle complex, long-horizon real-world tasks whose context demands can grow without bound, yet model context windows remain inherently finite. Recent work explores a paradigm where a main agent decomposes tasks and dispatches subtasks to subagents, which execute and return only summarized results, conserving the main agent's context budget. However, performing this well requires delegation intelligence: the ability to decompose complex tasks, determine when and what to delegate, and integrate returned results into the ongoing workflow. Training data for this capability is scarce in naturally occurring text, and to our knowledge, how to synthesize such data and train models to acquire this capability remains largely unexplored in the open-source community. To bridge this gap, we present a preliminary exploration targeting deep research, a representative long-horizon agent task. Specifically, we design a harness that guides the model toward high-quality task decomposition and delegation, while constraining subagents to return results properly to support the main agent's workflow. The harness-guided trajectories naturally encode correct delegation decisions, which we use as supervised fine-tuning data to internalize delegation intelligence into model weights. Our resulting model, SearchSwarm-30B-A3B, achieves 68.1 on BrowseComp and 73.3 on BrowseComp-ZH, the best results among all models of comparable scale. We will release our harness, model weights, and training data to facilitate future research.
arXiv:2606.07108v2 Announce Type: replace Abstract: Recent advances in Large Reasoning Models (LRMs) demonstrate remarkable performance improvements by iteratively reflecting, exploring, and executing complex tasks, yet suffer from inefficiencies due to redundant reasoning, known as "overthinking". Existing methods to mitigate this issue either rely on static difficulty estimates or require task-specific training, and thus fail to adapt to the dynamic complexity during reasoning. In this work, we empirically show that the problem difficulty evolves dynamically throughout the reasoning process and is linearly encoded in the LRM's step-level embeddings. Building on this insight, we propose DyCon, a training-free framework that leverages latent step-level representations to explicitly model the evolving task difficulty, enabling the dynamic control of reasoning depth to mitigate the overthinking issue. Extensive experiments conducted on four models ranging from 4B to 32B, and across twelve benchmarks in math reasoning, general question answering, and coding tasks demonstrate that DyCon significantly enhances reasoning efficiency by reducing redundant steps without sacrificing accuracy or generalization. Code is available at https://github.com/yu-lin-li/DyCon.
arXiv:2606.07252v2 Announce Type: replace Abstract: Long multimodal document question answering is limited by which evidence reaches the reader, rather than by the quantity retrieved. In lengthy documents, findings often recur across figures, captions, and introductory sentences, causing similarity based retrievers in modern multimodal retrieval-augmented generation (RAG) systems to allocate resources to near-duplicates while overlooking complementary evidence. This work introduces a retriever that selects evidence as a Constrained Dominant Set (CDS) on a query-augmented affinity graph, offering three advantages that similarity ranking does not. First, the query is encoded as a hard structural constraint, ensuring that every selected element is directly connected to the question through the cluster anchor. Second, the relevance-redundancy balance is determined automatically by a spectral bound, eliminating the need for manually tuned trade offs required by diversity-aware selectors. Third, the selection process achieves a global equilibrium via replicator dynamics, thereby avoiding the distortions introduced by greedy heuristics. The method is inherently graph-based and does not require training. Using a Qwen3-VL-32B reader, CDS establishes a new state of the art on VisDoMBench ($66.99$ average) and improves over the no-retrieval baseline by $37.1$ points on VisDoMBench and $4.8$ on MMLongBench-Doc.
arXiv:2606.07419v2 Announce Type: replace Abstract: Recovering 3D human poses for multiple individuals from different camera views is a fundamental bottleneck for analyzing interacting behaviors. Existing self-supervised approaches leverage synthetic catalogues of 3D poses; however, this leads to poor generalization in real-world scenarios due to distribution shifts. We therefore introduce DisPOSE, a self-supervised framework that approximates the inherently discrete multi-view person-assignment problem as a generative diffusion process over the space of polystochastic tensors. By employing differentiable Sinkhorn projections during denoising, our model learns to guide solutions toward valid and feasible assignments based on 2D image priors. The complete 3D skeletons of localized individuals are then regressed using a Hypergraph-Convolutional Decoder that explicitly models relational structures and articulated joints across multiple views. The proposed approach outperforms current state-of-the-art self-supervised methods on standard datasets and demonstrates strong performance on a newly proposed benchmark featuring highly occluded scenes from surgical operating rooms. Our diffusion-based localization demonstrates high label efficiency, retaining 99% of its performance with only 10% of the pseudo-labels. Notably, disentangling the assignment and root regression components while maintaining differentiability makes DisPOSE nearly agnostic to different camera arrangements.
arXiv:2203.07194v5 Announce Type: replace-cross Abstract: The theory of $(\infty,1)$-categories can be developed synthetically in an augmentation of homotopy type theory introduced by Riehl--Shulman. Central to their development is an additional type forming operation called extensions. The original article sketches the semantics of this formal system, explaining how the simplicial homotopy theory can be used to reason about $(\infty,1)$-categories presented using the Segal space model. However, they leave it open to demonstrate the strict stability of extension types. We prove this using the splitting method of Voevodsky, later generalized by Lumsdaine--Warren to local universes. The practical upshot is that this system has semantics in simplicial objects of an $\infty$-topos, and thus can be used to prove theorems about internal $\infty$-categories in the sense of Martini--Wolf.
arXiv:2503.22697v3 Announce Type: replace-cross Abstract: Decoding sensory experiences from neural activity to reconstruct human-perceived visual stimuli and semantic content remains a challenge in neuroscience and artificial intelligence. Despite notable progress in current brain decoding models, a critical gap still persists in their systematic integration with established neuroscientific theories and the exploration of underlying neural mechanisms. Here, we present a novel framework that directly decodes fMRI signals into textual descriptions of viewed natural images. Our novel deep learning model, trained without visual information, achieves state-of-the-art semantic decoding performance, generating meaningful captions that capture the core semantic content of complex scenes. Neuroanatomical analysis reveals the critical role of higher-level visual cortices, including MT+ complex, ventral stream visual cortex, and inferior parietal cortex, in visual semantic processing. Furthermore, category-specific analysis demonstrates nuanced neural representations for semantic dimensions like animacy and motion. This work provides a more direct and interpretable framework to the brain's semantic decoding, offering a powerful new methodology for probing the neural basis of complex semantic processing, refining the understanding of the distributed semantic network, and potentially developing brain-inspired language models.
arXiv:2509.11485v3 Announce Type: replace-cross Abstract: Labyrinthine stripe patterns are common in many physical systems, yet their lack of long-range order makes quantitative characterization challenging. We investigate the evolution of such patterns in bismuth-doped yttrium iron garnet (Bi:YIG) films subjected to a magnetic field annealing protocol. A U-Net deep learning model, trained with synthetic degradations including additive white Gaussian and Simplex noise, enables robust segmentation of experimental magneto-optical images despite noise and occlusions. Building on this segmentation, we develop a geometric analysis pipeline based on skeletonization, graph mapping, and spline fitting, which quantifies local stripe propagation through length and curvature measurements. Applying this framework to 444 images from 12 annealing protocol trials, we analyze the transition from the "quenched" state to a more parallel and coherent "annealed" state, and identify two distinct evolution modes (Type A and Type B) linked to field polarity. Our results provide a quantitative analysis of geometric and topological properties in magnetic stripe patterns and offer new insights into their local structural evolution, and establish a general tool for analyzing complex labyrinthine systems.
arXiv:2605.13768v2 Announce Type: replace Abstract: This is the second part of the work investigating quantized matrix multiplication (MatMul). In part I we considered the case of calibration-free quantization, whereas here we discuss the setting where covariance matrix $\Sigma_X$ of the columns of the second factor is available. This setting arises in the ubiquitous task of weight-only post-training quantization of LLMs. Weight-only quantization is related to the problem of weighted mean squared error (WMSE) source coding, whose classical (reverse) waterfilling solution dictates how one should distribute rate between coordinates of the vector. We show how waterfilling can be used to improve practical LLM quantization algorithms (GPTQ), which at present allocate rate equally. A recent scheme (known as ``WaterSIC'') that only uses scalar INT quantizers is analyzed and its high-rate performance is shown to be (a) basis free (i.e., characterized by the determinant of $\Sigma_X$ and, thus, unlike existing schemes, is immune to applying random rotations); and (b) within a multiplicative factor of $\frac{2\pi e}{12}$ (or 0.25 bit/entry) of the information-theoretic distortion limit. GPTQ's performance, in turn, is affected by the choice of basis, but for a random rotation and actual $\Sigma_X$ from Llama-3-8B we find it to be within 0.1 bit (depending on the layer type) of WaterSIC, suggesting that GPTQ with random rotation is also near optimal, at least in the high-rate regime.
arXiv:2606.08135v1 Announce Type: new Abstract: Repository-level code generation with Large Language Models (LLMs) remains challenging, primarily due to complex dependencies and limited context windows. Recent approaches adopt retrieval-augmented generation (RAG) and the planning mechanism to reuse potential callee functions in the repository. However, these approaches often suffer from two limitations: lack of test-driven behavioral guidance during planning and overlooking the implementation logic embedded in repository code during reuse. As a result, generated plans may not align with expected behaviors, and retrieved functions may not be effectively reused. In this paper, we propose TICoder, a novel repository-level code generation framework that improves both planning and reuse. TICoder introduces a test-driven iterative planning mechanism that leverages test cases as behavioral specifications to refine implementation steps. Furthermore, TICoder employs an implementation-aware code reuse strategy, which retrieves potential callee functions using a dual-view similarity that captures both functional and implementation aspects. We then identify relevant usage patterns through a dual-stage selection strategy, combining structure-based clustering and perplexity-based filtering. We conduct extensive experiments on widely used repository-level code generation benchmarks with various LLMs. Experimental results demonstrate that TICoder outperforms state-of-the-art (SOTA) methods, achieving an average improvement of 11.52%.
arXiv:2606.08138v1 Announce Type: new Abstract: Eukaryotic DNA replication must remain robust under thermal, chemical, and genotoxic stress despite large fluctuations in replication dynamics. Here, we develop a lattice-based stochastic Monte Carlo framework for whole-genome replication in Saccharomyces cerevisiae at single base-pair resolution, incorporating probabilistic origin firing, replication fork-speed distributions, and a time-dependent limiting factor that governs the availability of cellular replication resources. The model is benchmarked quantitatively against experimental replication profiles before being applied to stress conditions, and reproduces diverse replication stress responses using only two effective parameters. Importantly, the analysis reveals that replication fork-speed heterogeneity underlies the emergence of Erlang-distributed S-phase durations and rare, anomalously prolonged replication events observed experimentally in Escherichia coli and human cell lines, while predicting similar behavior in S. cerevisiae. The framework further predicts non-monotonic thermal behavior, power-law scaling under hydroxyurea stress, and total replication-time dynamics under diverse genotoxic conditions.
arXiv:2606.09511v1 Announce Type: new Abstract: Self-supervised data curation provides a pathway to scaling and improving the generalization capabilities of machine learning models. By leveraging self-supervised learning (SSL) for data curation, the demand for massive training datasets required by foundation models can be effectively met. SSL greatly alleviates the costs associated with annotation and manual dataset curation while minimizing the need for human oversight. However, the integrity of SSL-curated datasets must be rigorously checked, as reliance on anonymous and unvetted external sources can substantially increase the risk of data poisoning. In this paper, we propose a Poisoned Data Detector (PDD), an active defense mechanism designed to ensure the integrity of SSL-curated datasets prior to foundation model training. PDDs are designed using a combination of the pretrained ImageBind model and traditional classifiers, including Random Forest (RF), k-Nearest Neighbors (KNN), Naive Bayes (NB), and Support Vector Machines (SVM). We rigorously evaluated PDDs using 176,200 images from three diverse datasets and three different adversarial attacks encompassing both in-distribution and out-of-distribution scenarios. Notably, SVM-PDD achieves superior performance for both in-distribution (Set3-Set5) and out-of-distribution (TrueFace and 140K RealFace) datasets. Our design demonstrates strong scalability and enables the rapid integration of new adversarial attack detectors through an ensemble approach.
arXiv:2511.18493v4 Announce Type: replace-cross Abstract: The significant variability in cell size and shape continues to pose a major obstacle in computer-assisted cancer detection on gigapixel Whole Slide Images (WSIs), due to cellular heterogeneity. Current CNN-Transformer hybrids use static computation graphs with fixed routing. This leads to extra computation and makes it harder to adapt to changes in input. We propose Shape-Adapting Gated Experts (SAGE), an input-adaptive framework that enables dynamic expert routing in heterogeneous visual networks. SAGE reconfigures static backbones into dynamically routed expert architectures via a dual-path design with hierarchical gating and a Shape-Adapting Hub (SA-Hub) that harmonizes feature representations across convolutional and transformer modules. Embodied as SAGE with ConvNeXt and Vision Transformer UNet (SAGE-ConvNeXt+ViT-UNet), our model achieves a Dice score of 95.23% on EBHI, DSC scores of 92.78% and 91.42% on GlaS Test A and Test B, respectively, and 91.26% DSC at the WSI level on DigestPath, while exhibiting robust generalization under distribution shifts by adaptively balancing local refinement and global context. SAGE establishes a scalable foundation for dynamic expert routing in visual networks, thereby facilitating flexible visual reasoning. Project page: https://oxyzgiahuy.github.io/sage/
arXiv:2512.05517v3 Announce Type: replace-cross Abstract: The ferroelectric nematic (NF) phase and related polar liquid-crystalline phases form a new class of strongly polarized yet fluid soft matter. Well-recognized heliconical ferroelectric phases such as NTBF and SmC_P^H exist, but their electro-optic functionality in smectic systems has been largely unexplored. Here, we report a newly designed single-component achiral molecule exhibiting a hierarchical polar phase sequence: SmAF-NTBF-HEC-SmC_P^H. The key feature of the SmC_P^H phase is its ability to form a stable macroscopic orientation without any alignment layers and to enable continuous, reversible pitch modulation over a wide spectral range at ultralow electric fields, approximately one-third of those required for conventional heliconical nematics. The system exhibits characteristic electro-optic properties in the sub-kilohertz range, differentiating it from heliconical materials controlled by dielectric-elastic balance at higher frequencies. In addition, the SmC_P^H phase's polar heliconical smectic structure facilitates enhanced second-harmonic generation via favorable phase matching, thereby establishing a simple and robust platform for low-voltage photonic applications.
arXiv:2601.07013v2 Announce Type: replace-cross Abstract: Traditional filtering algorithms for state estimation -- such as classical Kalman filtering, unscented Kalman filtering, and particle filters -- show performance degradation when applied to nonlinear systems whose uncertainty follows arbitrary non-Gaussian, and potentially multi-modal distributions. This study reviews recent approaches to state estimation via nonlinear filtering based on conditional normalizing flows, where the conditional embedding is generated by standard MLP architectures, transformers or selective state-space models (like Mamba-SSM). In addition, we test the effectiveness of an optimal-transport-inspired kinetic loss term in mitigating overparameterization in flows consisting of a large collection of transformations. We investigate the performance of these approaches on applications relevant to autonomous driving and patient population dynamics, paying special attention to how they handle time inversion and chained predictions. Finally, we assess the performance of various conditioning strategies for an application to real-world COVID-19 joint SIR system forecasting and parameter estimation.
arXiv:2602.15519v3 Announce Type: replace-cross Abstract: Target speech extraction (TSE) typically relies on pre-recorded high-quality enrollment speech, which disrupts user experience and limits feasibility in spontaneous interaction. In this paper, we propose Enroll-on-Wakeup (EoW), a novel framework where the wake-word segment, captured naturally during human-machine interaction, is automatically utilized as the enrollment reference. This eliminates the need for pre-collected speech to enable a seamless experience. We perform the first systematic study of EoW-TSE, evaluating advanced discriminative and generative models under real diverse acoustic conditions. Given the short and noisy nature of wake-word segments, we investigate enrollment augmentation using LLM-based TTS. Results show that while current TSE models face performance degradation in EoW-TSE, TTS-based assistance significantly enhances the listening experience, though gaps remain in speech recognition accuracy.
arXiv:2603.18955v3 Announce Type: replace-cross Abstract: The Solvability Complexity Index (SCI) provides an extensional limit-height formalism for recovering a target map $\Xi$ from finite samples of an evaluation interface $\Lambda\subseteq\mathbb C^\Omega$ by finite-height towers of pointwise limits. We first give a foundational analysis of what this extensional framework does and does not determine. We show that the SCI separation axiom is equivalent to a factorization of $\Xi$ through the full evaluation table, and we isolate the minimal logical role of $\Lambda$ as an information interface. To connect the SCI to Type-2 computability and Weihrauch reducibility, we give an effective enrichment for countable $\Lambda$ by viewing the evaluation table image $I_{\Lambda}\subseteq\mathbb{C}^{\mathbb{N}}$ as a represented space and factoring $\Xi$ as $\widehat{\Xi}$. We then define the Weihrauch-SCI rank of a problem as the least number of iterated limit-oracles needed to compute it in the Weihrauch sense, i.e. the least $k$ such that $\widehat{\Xi}\le_{W}\lim^{(k)}$, and prove well-posedness and representation invariance of this rank. A central negative result is that the unrestricted raw type-G SCI model (arbitrary post-processing of finite oracle transcripts) is generally not a computability model in the Type-2/Weihrauch sense: finite-query factorizations collapse raw type-G height, and analytic non-Borel decision problems yield examples with raw $\mathrm{SCI}_G=0$ but infinite Weihrauch-SCI rank. We therefore distinguish the raw extensional SCI from implemented SCI variants, where the indexed approximation table is required to be realized uniformly by a chosen class of operations. To recover a robust bridge, we introduce an intermediate SCI hierarchy by restricting the admissible deepest-level post-processing to regularity classes (continuous/Borel/Baire).
arXiv:2605.14211v3 Announce Type: replace Abstract: Long-horizon embodied tasks remain a fundamental challenge in AI, as current methods rely on hand-engineered rewards or action-labeled demonstrations, neither of which scales. We introduce ASH, an agentic system that learns an embodied policy from unlabeled, noisy internet video, without reward shaping or expert annotation. ASH follows a self-improvement loop; when it gets stuck, ASH learns an Inverse Dynamics Model (IDM) from its own trajectories, and uses its IDM to extract supervision from relevant internet video. ASH uses unsupervised learning to identify key moments from large-scale internet video and retains them as long-term memory -- allowing it to tackle long-horizon problems. We evaluate ASH on two complementary environments demanding multi-hour planning: Pokemon Emerald, a turn-based RPG, and The Legend of Zelda: The Minish Cap, a real-time action-adventure game. In both games, behavioral cloning, retrieval-augmented and zero-shot foundation-model baselines plateau, while ASH sustains progression across our 8-hour evaluation. ASH reaches an average of $11.2/12$ milestones in Pokemon Emerald and $9.9/12$ in Legend of Zelda, while the strongest baseline gets stuck in both environments at an average of $6.5/12$ and $6.0/12$ milestones, respectively. We demonstrate that self-improving agents are a scalable recipe for long-horizon embodied learning.
arXiv:2604.16512v2 Announce Type: replace Abstract: We propose a novel variational method to compute a highly accurate global signed distance function (SDF) to a given point cloud. To this end, the jump set of the gradient of the SDF, which coincides with the medial axis of the surface, is explicitly taken into account through a higher-order variational formulation that enforces linear growth along the gradient direction away from this discontinuity set. The eikonal equation and the zero-level set of the SDF are enforced as constraints. To make this variational problem computationally tractable, a phase field approximation of Ambrosio-Tortorelli type is employed. The associated phase field function implicitly describes the medial axis. The method is implemented for surfaces represented by unoriented point clouds using neural network approximations of both the SDF and the phase field. Experiments demonstrate the method's accuracy both in the near field and globally. Quantitative and qualitative comparisons with other approaches show the advantages of the proposed method.
arXiv:2604.23868v3 Announce Type: replace-cross Abstract: We investigate an analytic theory of force-dipole hydrodynamics in a viscous membrane coupled to an infinite surrounding fluid, focusing on quenched (orientation-fixed) dipoles. While the single-dipole flow exhibits the known Saffman crossover from a near-field $v\sim r^{-1}$ to a screened far-field $v\sim r^{-2}$, we show that this crossover induces a qualitatively new reorganization of dipole--dipole interactions. For two identical quenched dipoles, the near-field dynamics is exactly solvable and effectively one-dimensional, with a fixed line of centers and linear evolution of the squared separation. In the far field, the system remains integrable but becomes intrinsically two-dimensional, with coupled radial and angular dynamics and an exact first integral. For pullers, the angular dynamics drives alignment toward an attracting manifold, leading to universal late-time collapse $R\sim (t_c-t)^{1/3}$, in contrast to the near-field scaling $R\sim (t_c-t)^{1/2}$. The Saffman crossover thus reorganizes the Hamiltonian phase-space structure of dipolar interactions and produces a transition from effectively one-dimensional to fully coupled dynamics, providing a minimal framework for aggregation in viscous fluid membranes.
arXiv:2605.06692v3 Announce Type: replace-cross Abstract: Coherence resonance (CR) characterizes noise-induced regularity in excitable systems, yet its evaluation in quiescent biological media is often obscured by flattened energy landscapes and complex nonlinear dynamics. In this study, we investigate the stochastic dynamics of a 3D Sherman-Rinzel-Keizer (SRK) model driven by multiplicative Feller noise. We show that traditional extremal evaluations of CR encounter a "bathtub effect", a broad resonance valley that can lead to statistical inaccuracies. To address this, we propose a logarithmic centroid extraction method, which filters out stochastic jitter and recovers the underlying adiabatic Kramers scaling with high linearity. Furthermore, we identify the physical boundary where this adiabatic approximation breaks down under the strong-noise limit. Extending our analysis to gap-junction coupled systems, we observe a noise-induced transition from sub-threshold physiological shivering (characterized by statistical correlation but negligible functional output) to macroscopic functional synchronization. Our results provide a mathematical framework for extracting optimal noise intensities in broad energy valleys and offer insights into how quiescent biological systems utilize stochastic fluctuations for functional recovery.
arXiv:2605.09851v3 Announce Type: replace-cross Abstract: Ball lightning remains one of the most enigmatic atmospheric phenomena, characterized by its long lifetime, ability to penetrate materials, and stable spherical structure. Here we propose a novel theoretical framework interpreting ball lightning as a three-dimensional projection of a high-dimensional topological soliton. The system is described by a nonlinear Schr\"odinger equation with attractive interactions, stabilized by a non-zero topological charge. Through comprehensive numerical simulations of the three-dimensional Gross-Pitaevskii equation, we verify the model's core predictions: (1) long-lived stability protected by topological invariants, (2) low transmission probability due to wavefunction orthogonality, and (3) energy and size scales consistent with observational data. The soliton lifetime $\tau \sim \hbar/\Gamma$ naturally explains the observed second-scale durations. Our work provides a self-consistent physical explanation for ball lightning while offering concrete pathways for experimental realization of three-dimensional topological solitons in Bose-Einstein condensates and nonlinear optical systems. This theoretical framework gains additional support from recent experimental breakthroughs in laboratory generation of ball-lightning-like structures.
arXiv:2510.16028v4 Announce Type: replace Abstract: Neural networks increasingly run on hardware outside the user's control (cloud GPUs, inference marketplaces). Yet ML-as-a-Service reveals little about what actually ran or whether returned outputs faithfully reflect the intended inputs. Users lack recourse against service downgrades (model swaps, quantization, graph rewrites, or discrepancies like altered ad embeddings). Verifying outputs is hard because floating-point(FP) execution on heterogeneous accelerators is inherently nondeterministic. Existing approaches are either impractical for real FP neural networks or reintroduce vendor trust. We present TAO: a Tolerance Aware Optimistic verification protocol that accepts outputs within principled operator-level acceptance regions rather than requiring bitwise equality. TAO combines two error models: (i) sound per-operator IEEE-754 worst-case bounds and (ii) tight empirical percentile profiles calibrated across hardware. Discrepancies trigger a Merkle-anchored, threshold-guided dispute game that recursively partitions the computation graph until one operator remains, where adjudication reduces to a lightweight theoretical-bound check or a small honest-majority vote against empirical thresholds. Unchallenged results finalize after a challenge window, without requiring trusted hardware or deterministic kernels. We implement TAO as a PyTorch-compatible runtime and a contract layer currently deployed on Ethereum Holesky testnet. The runtime instruments graphs, computes per-operator bounds, and runs unmodified vendor kernels in FP32 with negligible overhead (0.3% on Qwen3-8B). Across CNNs, Transformers and diffusion models on A100, H100, RTX6000, RTX4090, empirical thresholds are $10^2-10^3$ times tighter than theoretical bounds, and bound-aware adversarial attacks achieve 0% success. Together, TAO reconciles scalability with verifiability for real-world heterogeneous ML compute.
arXiv:2606.09802v1 Announce Type: new Abstract: We consider a variant of the linear contextual stochastic multi-armed bandits, where the learner must provide recommendations to a group of users, each having its personalized preference vector, and in the presence of context distributions that are drifting over time. Under practitioner-friendly assumptions, we reduce this setting to linear bandit with stationary mean but heteroskedastic and non-stationary noise. We further study the case when the learner must ensure the mean reward of each decision must exceed that of a baseline strategy $\boldsymbol{\pi}_0$ at each decision step. We introduce Dri-MED, an algorithm inspired from the linear version of the MED strategy, and carefully adapted to handle the non-stationary heteroskedastic noise. We show that the instance-dependent regret scales as $\tilde{\mathcal O}\left(\frac{\kappa}{\tilde{\Delta}}d^2(\log(T)\right)$, where $\tilde{\Delta}$ is the constraint-aware sub-optimality gap subject to policy $\pi_0$, with variance-aware multiplicative term $\kappa$ that we carefully handle using heteroskedastic regression. We further show Dri-MED enjoys $\tilde{\mathcal{O}}(d)$ expected constraint violations. Our numerical results suggest that Dri-MED significantly outperforms conservative baselines that ignores the drift and preference structure.
arXiv:2606.08751v1 Announce Type: new Abstract: Accurate quantification and uptake measurement in PET are critical for assessing disease progression and supporting clinical decision-making. While high-count PET provides reliable image quality, the associated radiation dose and prolonged acquisition remain significant clinical concerns, motivating the adoption of low-count protocols. Diffusion-model-based methods have demonstrated strong potential for restoring low-count PET to near high-count quality, but their iterative sampling procedure becomes prohibitively expensive when applied to high-resolution 3D PET volumes, introducing substantial inference latency that limits practical clinical deployment. To address these challenges, we propose a training-free Global-Local Skipping Strategy that accelerates diffusion model-based 3D PET denoising while simultaneously improving reconstruction quality. The proposed method is plug-and-play and directly applicable to pre-trained diffusion models without retraining or architectural modification. Specifically, we introduce: (i) a global denoising step skipping strategy that initializes the reverse diffusion process from an intermediate denoising step using a noise-consistent transformation of the low-count input, substantially reducing the number of required denoising steps; and (ii) a local feature reuse shortcut that reuses slowly-varying high-level U-Net features across neighboring denoising steps, further reducing per-step computation while preserving image fidelity. We evaluate the proposed approach on multiple PET tracers from in-house and public datasets, including 18F-FDG PET, 68Ga-DOTATATE PET, and 18F-PSMA PET, demonstrating consistent acceleration of over an order of magnitude alongside improved or comparable reconstruction performance relative to the full-step baseline. Blinded reader studies further confirm enhanced clinical confidence and perceived diagnostic quality.