Let $m\geq5$ be odd with $3\nmid m$. Helleseth, Ichanska, and Kaleyski \cite{HIK} conjectured that two norm-perturbed Budaghyan--Helleseth--Kaleyski functions $F_2,F_3$ over $\F_{2^{2m}}$ are APN. In this paper, we prove this conjecture. First, by using the basis $(1,β)$ of $\F_{2^{2m}}$ over $\F_{2^m}$, we show that $F_2$ and $F_3$ are EA-equivalent to two functions $\mathcal G_1$ and $\mathcal G_0$, respectively. Their derivatives can be written as two trace equations for a matrix $P\in\mathrm{GL}(2,\F_{2^m})$. We then prove that these two equations have no solution. The proof uses only identities over finite fields, the characteristic polynomial of a $2\times2$ matrix, and elementary properties of Frobenius automorphisms. The repeated-root case and all parity cases are considered separately. It follows that the derivative of each function has kernel of size two in every nonzero direction, and hence both $F_2$ and $F_3$ are APN for every admissible $m$.
George Dimitriadis, Damian Machlanski, Paul Clarke, Spyridon Samothrakis
Generalised Additive Models (GAMs) are a cornerstone of interpretable machine learning, offering transparent predictions through structured additive decompositions. Despite advances from classical GAMs to modern neural and spline-based variants, two notable gaps remain. First, comprehensive benchmarking of spline-based GAMs on large-scale tabular learning tasks, the dominant modality in applied settings, is lacking. Second, their robustness under distribution shift, a setting where structured inductive biases should confer advantages, remains unexplored. We propose a multi-resolution, higher-order Composite Generalised Additive Model (CoGA2M) and evaluate it on the TALENT benchmark suite. CoGA2M ranks in the top third across all baselines (29 for regression, 32 for classification), establishing competitive performance with state-of-the-art tabular methods. We further extend TALENT with out-of-distribution (OOD) splits exhibiting train-test distributional shift and benchmark CoGA2M against LightGBM, a consistently top-performing model on TALENT. Under distribution shift, CoGA2M outperforms LightGBM, demonstrating a clear structural advantage. Our results position spline-based GAMs as strong contenders for tabular prediction and reveal a previously unrecognised benefit of additive structure in the OOD regime.
Quan Tran, Duong Nguyen, Thanh Le-Cong, Triet Huynh Minh Le, M. Ali Babar, Hoang Dieu Vu, Quyet-Thang Huynh
Just-In-Time Vulnerability Prediction (JIT-VP) identifies security risks in code changes at commit time, enabling early feedback before vulnerabilities enter software systems. However, its practical reliability is challenged by the difficulty of constructing accurate labels. Multiple strategies with different trade-offs have been proposed to identify vulnerability-fixing commits (VFCs) and vulnerability-introducing commits (VICs) for dataset construction, leading to substantially different datasets. This makes it difficult to separate model performance from dataset construction. Moreover, real-world quality assurance requires not only classification but also prioritizing high-risk commits and localizing vulnerable causes. This paper presents an extensive empirical study of how labeling strategies affect these downstream tasks. We constructed 28 datasets using four VFC and four VIC labeling strategies across OpenSSL and Linux kernel, then evaluated six representative JIT-VP models for classification, prioritization, and localization. Our results show that labeling effectiveness is strongly task-dependent. Manual VFC labeling combined with vulnerability-specific VIC labeling provides the most reliable performance for classification and prioritization, significantly outperforming 11 of 15 alternatives. Meanwhile, precision-first VFC labeling combined with LLM-assisted VIC labeling performs best for localization, significantly outperforming 7 of 15 alternatives. These findings demonstrate that no single labeling strategy is universally optimal and highlight the importance of task-oriented dataset construction.
Aashish Chapagai, Angad Singh Gurtatta, Erik Cambria
Saliency maps are offered as a safeguard for medical image classifiers, but whether they reveal individual errors at a new hospital remains largely untested. We trained a convolutional network and a vision transformer on NIH ChestX-ray14 and evaluated 303,768 decisions per model at the training hospital and at two external sites, in the USA and Vietnam, with radiologist-consensus labels. We scored each explanation by its anatomical enrichment, the area-normalized concentration of attribution in a segmented thoracic region, and tested whether it detects errors beyond the model's confidence. Grad-CAM localized findings less often than radiologists (hit rate at most 0.575 versus 0.888). Adding enrichment to a confidence-based error detector calibrated at the training hospital never raised its area under the curve by more than 0.011, and for the convolutional network the relation between enrichment and correctness reversed between hospitals, degrading the transferred detector. Model confidence, not saliency, should anchor deployment monitoring.
Andrey Ignatov, Georgy Perevozchikov, Radu Timofte, Yuhong He, Yuxuan Qiu, Tianyu Hao, Ruoqi Li, Yueqi Yang 等 22 位作者
With the increasing importance of mobile camera quality in modern smartphones, growing attention has been directed toward computational photography and advanced image signal processing (ISP) algorithms aimed at enhancing various perceptual attributes of captured images. This creates a need for efficient deep learning models for this problem that are not only performant but can also be deployed on constrained mobile hardware. In this Mobile AI challenge, we address this problem and propose the participants to design fast solutions that can reconstruct RGB photos from RAW Bayer camera data while running on popular smartphone GPUs from Qualcomm and Arm. For this, the participants were provided with a large-scale Fujifilm UltraISP dataset consisting of RAW-RGB image pairs captured with the Sony IMX586 Quad Bayer mobile sensor and a professional 102MP medium format FujiFilm GFX100 camera. The runtime of all models was evaluated on the Snapdragon 8 Elite Gen 5 and Dimensity 9500 chipsets with the Adreno and Mali GPUs, respectively. Top solutions proposed in this challenge are able to reconstruct high-quality 5MP RGB image from raw data on both platforms under 100 milliseconds. A comprehensive description of the models developed in the challenge is provided in this paper.
Executing service workflows with directed acyclic graph (DAG) dependencies over wireless distributed computing networks requires joint control of function placement, routing, and scheduling, together with synchronization of inputs from the same request at merge points. We study throughput-optimal control of these networks under stochastic request arrivals, time-varying wireless links, and interference constraints. We use a valid mapping abstraction that assigns each service function to a computation node and each DAG dependency to a routing path. Using this abstraction, we characterize the network capacity region to support service DAGs, and prove that synchronizing the packets of a request at merge points does not shrink the region. We develop Dynamic Computing Network Control for service DAGs (DCNC-DAG), a dynamic control policy that selects queue-aware valid mappings and schedules service DAG execution by nearest-to-origin (NTO) while enforcing input synchronization at merge points. With exact minimum-cost valid mappings, DCNC-DAG is throughput-optimal and ensures rate stability of the physical queues under bounded arrivals. These guarantees hold for any fixed finite interval between mapping updates, even when mapping decisions rely on boundedly stale queue length information. For larger service DAGs, we develop DCNC-DAG-H, which replaces exact mapping optimization with a queue-aware heuristic. Simulations show that DCNC-DAG supports $2.7\times$-$5.5\times$ the throughput of queue-blind baselines, while DCNC-DAG-H achieves up to $2.68\times$ those of the corresponding baselines on larger workflow and AI service DAGs.
Content-category dashboards summarize production recommendation logs, although the serving policy selects the users, items, and contexts behind each mean. Using KuaiRand random-insertion impressions, we audit these summaries on the same 12,472 users and 5,486 videos in both logs. Standard means have absolute/centered mean absolute error (MAE) 0.1935/0.0491 and rank correlation 0.7833. User demeaning plus a random grand-mean anchor reduces absolute MAE by 80.0%, but centered MAE by only 21.0%, without reliably improving rank. A complete-support ledger allocates weighted squared error mainly to within-video residual (72.41%) and video mix (23.90%), with 3.65% from random support. In a frozen later test, early category calibration reaches MAE 0.0251 versus 0.0531 for scalar calibration; their direct difference is -0.0280 [-0.0296, -0.0172]. Decision evidence is cutoff-specific: the nominal top-10 regret change for row-weighted demeaning versus naive is -0.0062 [-0.0084, -0.0003], while top-5 and top-20 intervals include zero. The six-contrast family-wise top-10 interval [-0.0136, 0.0013] includes zero. An outcome-free disjoint randomized-reference split also leaves all six nominal contrasts covering zero. A known-truth stress test shows incomplete improvement under a mean-additive positive control and failure under item mix and weak overlap. All decompositions are descriptive and each result is conditional on its overlap cohort. KuaiRand intervals quantify user-cluster variation conditional on the realized catalog and calendar; we claim neither intrinsic category effects nor a universally valid correction.
Effie Daum, Daniele De Martini, Claire Dune, François Pomerleau
In field robotics, acquiring independent large-scale reference trajectories more accurate than the evaluated estimates remains an open challenge. The domain is widely reliant on Absolute Trajectory Error (ATE) and Relative Pose Error (RPE), computed with automated tools, that rest on assumptions and evaluation parameters rarely made explicit. When unreported, the errors can be misleading and hinder fair comparisons. This paper introduces a trajectory-evaluation protocol for standardized and reliable accuracy assessment in state estimation, localization, and Simultaneous Localization And Mapping (SLAM). The approach combines a novel temporal alignment method based on curvature signals with an error metric normalized by travelled distance. We explicitly account for temporal synchronization, sampling alignment, and extrinsic calibration, quantifying their influence through a sensitivity analysis. The proposed protocol contributes to more rigorous, reproducible, and standardized trajectory evaluation.
Requested-value control is stricter than continuous steering: given $c$ on a human-rating scale, the generated image should attain that value from only limited ratings. We present ScoreDial, which fits a differentiable regressor $R$ to 100-479 human-rated images; once $R$ is fixed, a requested value $c$ is defined on its output scale, and learned target conditioning is paired with error-adaptive guidance that repeatedly reduces $|R(\mathbf{x})-c|$ during sampling. Across tire slip resistance, interior visual complexity, food desire-to-eat, and soap translucency, using one set of sampler settings shared across all four, ScoreDial reaches requested values and achieves, on all four attributes, $R$-MAE at most 0.013 over a range of width 0.8 and Spearman $ρ=0.980$, the maximum attainable under our evaluation protocol. Because $R$ also drives sampling, we evaluate agreement with $R$ separately from matched sweeps, an independent auditor fitted to the same ratings on a backbone none of the deployed regressors uses, and blinded human judgments. MPGD-flow also reaches the requested values under $R$, but ScoreDial has larger auditor displacement on every attribute, the largest among all compared methods, and a stronger MPGD-flow setting gives up target agreement without closing this gap. On every attribute, removing target conditioning leaves low $R$-MAE but more than halves auditor displacement, separating precise score correction from attribute movement. Blinded rating studies with 15 raters per attribute show that mean human ratings track $c$ on tire and soap, with linear $R^2=0.665$ and $R^2=0.786$, and in forced-choice comparisons of matched sweeps, raters chose ScoreDial's as the clearer progression in at least 84% of judgments against every baseline.
Pervasive Capability Degradation (PCD) refers to the phenomenon in which supervised fine-tuning (SFT), while adapting a large language model (LLM) to a target task, often erodes the general capabilities acquired during pre-training. Prevailing explanations attribute PCD to macroscopic distribution shift and therefore lack microscopic manipulability, leaving existing mitigation methods without a complete theoretical foundation. In this paper, we first develop a bottom-up geometric theory of PCD grounded in feature interference under superposition: narrow-domain SFT data disrupt the almost-orthogonal feature structure formed during pre-training, and the resulting interference drives activations across decision hyperplanes in the output Laguerre-Voronoi geometry, flipping predictions on general tasks. This theory unifies existing methods into six paradigms and reveals both the source of their efficacy and their theoretical limits. We then prove that cross-entropy under one-hot supervision cannot resolve PCD on its own, and accordingly propose the AEGIS loss, which directly constrains the drift of activation margins along all competing token directions, thereby decoupling interference suppression from task adaptation. Across 3 model families and 14 SFT tasks, AEGIS outperforms representative methods of every paradigm on target tasks and pervasive capabilities simultaneously, at only about 1.3 times the computational cost of standard SFT. Its generality, scalability, and mechanistic interpretability offer a new perspective on fully resolving PCD.
Atikkhan Faridkhan Nilgar, Manuel Dietrich, Kristof Van Laerhoven
Robots in shared settings may expose information intended for an individual user to nearby bystanders. Robot tutoring makes this concern particularly salient: critical feedback is directed at an individual learner, yet bystanders may overhear or see it. To address this, we compared context-invariant and context-adaptive robot behaviors in a between-subjects study with 40 participants. Following an initial phase without a bystander, a bystander entered the setting. Participants then received two scripted critical feedback messages concerning their pronunciation and handwriting, independent of their actual performance. With the context-invariant behavior, the robot tutor delivered both messages aloud. With the context-adaptive behavior, the robot tutor scanned for the bystander, turned its body and display away after detecting them, gave a spoken attention cue, and presented the same feedback messages silently on its display. Participants rated the robot tutor higher in perceived privacy awareness and perceived social competence in the context-adaptive condition than in the context-invariant condition, while evidence for reduced experienced embarrassment was inconclusive. These findings inform the design and evaluation of robots that respond to bystander presence by adapting how they communicate potentially sensitive information.
Surface parameterization is a fundamental task in geometry processing and shape analysis. For topologically closed surfaces, it is common to parameterize them onto a standardized closed domain in $\mathbb{R}^3$, with a prime example being a unit sphere for genus-0 closed surfaces. Many mapping methods use planar projections and perform computations on the plane rather than directly on the 3D surface. However, perturbations introduced before projection or geometric evolution propagate through the computation and may affect map quality, convergence, stopping behavior, and geometric validity. In this work, we formulate representation sensitivity as a branchwise input--output problem, separating smooth propagation through active Jacobians from changes in anchors, stopping rules, and fallback events. Under consistent piecewise-linear assembly, the unmodified south-pole Beltrami correction in the FLASH method for spherical conformal parameterization reduces exactly to source-mesh harmonic extension. On a fixed-anchor branch, the corrected map factors into chart transition, harmonic propagation, and inverse stereographic projection, yielding an explicit sensitivity Jacobian and a finite-triangle conditioning bound. Separately, we introduce a bounded five-parameter affine preconditioner over planar scale, translation, and anisotropy and evaluate candidates on the corrected spherical map. Across 30 genus-0 surfaces, it reduces median polar upper-tail log-condition and chord-angle distortion by $10.17\%$ and $10.97\%$, respectively, with both measures improving on every mesh. We also apply the framework to other spherical mapping methods including the conformalized mean-curvature flow and spherical density-equalizing maps. Altogether, our work highlights the importance of representation selection as a structural and practical component of spherical parameterization and downstream computation.
Psychological safety (PS) refers to the belief that team members can speak up, ask questions, and make mistakes without fear of negative consequences. Although PS has been studied in traditional software teams, less is known about how it may appear in pull-based open-source software development, where contributors are self-directed and often collaborate voluntarily. This paper introduces a theory-informed framework for understanding how PS may be reflected in pull request (PR) interactions. We investigate the framework through a mixed-methods approach combining repository-based analysis and a survey of OSS contributors. First, we constructed a PS index using nine observable PR interaction indicators extracted from 60,685 PRs across 26 GitHub repositories. Second, we surveyed 57 OSS contributors about their perceived PS and experiences of corresponding PR interactions. All nine surveyed interaction indicators were significantly and positively associated with perceived PS. However, the repository-based index was not significantly associated with survey-reported PS. Contributors' open-ended responses further highlighted the importance of feedback quality, opportunities to participate, and how PR outcomes are interpreted. Together, our framework and mixed-methods findings provide initial insights into how PS may be reflected in PR interactions, highlighting the importance of both PR interaction patterns and contributors' experiences.
Yannik Stradmann, Joscha Ilmberger, Eric Müller, Johannes Schemmel
Novel compute systems are an emerging research topic, aiming towards building next-generation compute platforms. For these systems to thrive, they need to be provided as research infrastructure to allow acceptance and usage by a large community. By the example of the neuromorphic BrainScaleS-2 system, we showcase the transformation from a laboratory setup to a publicly available research platform. It is embedded into a purpose-built institute, tightly coupling a conventional cluster with novel compute hardware. Our key contribution is the sustainable operation of such a research system: automation, integration, and resource separation enable maintainability, stable operation, and continuity in a fluctuating academic environment. We implement these concepts by adopting established best practices, such as network infrastructure with strong resource separation, 19-inch rack units, the integration of physical and simulated hardware into CI/CD processes, and system monitoring. We evaluate six years of platform operation comprising 3718 production deployments and an average of 198 hardware CI/CD runs per day, with the integration of simulated and physical hardware increasing the full-stack software line coverage from 57% to 79%. We conclude with an explicit collection of lessons learned during the transformation from single-user laboratory setups to a publicly accessible research platform that, since 2023, has served more than 520 individual researchers.
Purpose-built kernel agents can write high-performance GPU kernels, but existing systems generally lack cross-run knowledge accumulation and architectural evolution. Each run starts from the same initial knowledge, and later iterations largely optimize within an architecture fixed early in the search. We present ForgeKernel, a kernel-generation system that accumulates and reuses knowledge across runs while enabling continued evolution of kernel architectures. It analyzes the hardware behavior of a high-performance reference and distills validated architectural decisions into reusable knowledge of hardware mechanisms rather than imitating source code. It then seals a kernel that passes correctness and performance validation and uses it as the reference for the next optimization run, enabling architectural decisions to be revisited. Across 27 operators and 29 input configurations on NVIDIA Hopper and Blackwell GPUs, ForgeKernel matches or beats its expert reference on 25 configurations, with a geometric-mean performance gain of 10%. On Ascend 910C, it outperforms the per-operator public best on the public case set of all 8 level-4 operators in the CANN-Bench Season 1 final leaderboard, with a geometric-mean gain of 25.1%, while its full-suite submission ranks first on the A3 leaderboard. As of 2026-10-30, public submissions generated by ForgeKernel rank first on all three CANN-Bench 2.0 solution leaderboards (Ascend 910C, 950DT, and 950PR) and on the SOL-ExecBench FlashInfer-Bench track.
Web systems infer user preferences from behavior, ratings, text, and LLM-generated profiles, yet explaining a proxy may reveal its measurement process rather than the preference of interest. We develop an audit framework distinguishing point identification, partial identification, and abstention relative to an explicit reference distribution. Under feature-dependent, arbitrarily signed proxy bias, a strict majority of uncontaminated proxy attributions lets the coordinatewise median recover latent-attribution support and direction and, with externally justified positive scales, magnitude. We prove this threshold sharp using response-valid indistinguishable worlds, characterize sharp minimax error and exact compatible-target sets under approximate cleanliness and scale uncertainty, and propagate clean-proxy attribution error. A frozen known-truth benchmark spans two response families, four proxy counts, all contamination counts, three bias strengths, and eleven methods, with a focal comparison confirmed on twenty unseen seeds. Under correct budgets, a direct attribution-tensor audit achieved 100% coverage, 95.40%-97.05% active-driver sign decisions, and no false signs. In a temporally separated Amazon study, median aggregation reduced comparator-relative conditional sign error for all four targets; six of eight prespecified target-metric contrasts had Bonferroni intervals excluding zero. A controlled real-covariate perturbation favored the median on the prespecified focal metric for all four targets. Compatible sets showed that the current proxy configuration rarely licensed strict-sign conclusions. Weak comparator calibration limits the Amazon evidence to fitted-surface alignment rather than latent-preference recovery.
We study instance optimality for the shortest $st$-path problem on positive-weight graphs, emphasizing how graph structure and query access determine the cost of certifying shortest paths. For undirected simple graphs in the standard adjacency-list model, we prove instance-sensitive lower bounds against all randomized algorithms that are correct on every admissible input. A quantitative reservoir argument bounds the instance-optimality ratio of corrected bidirectional Dijkstra in terms of its exploration size and the vertex count. A tensor-graph reduction translates low-query explored edges into vertex-cover constraints, yielding sufficient conditions through optimizable explored cores, degree moments, exploration orientations, and auxiliary degree profiles. We also derive bounds using exceptional high-degree vertices and weight-critical edges, and prove that a superconstant speedup over bidirectional Dijkstra requires all but a vanishing fraction of explored edges in each direction to concentrate around sublinear vertex sets. We also introduce a local-discovery model in which algorithms initially know only the terminal handles, cannot query undiscovered vertices or obtain the vertex count, but may make degree queries at discovered vertices. In this model, corrected bidirectional Dijkstra is constant-factor instance-optimal on directed and undirected positive-weight simple graphs, including with incoming and outgoing degree queries, for every fixed adjacency ordering. Finally, we revisit the correction to the original update rule and give a self-contained proof of the corrected multigraph guarantee.
Lei Zhang, Ju Dong, Kaixin Bai, Minheng Ni, Yu Fu, Zoltan-Csaba Marton, Zhaopeng Chen, Jianwei Zhang
Safe task-level manipulation requires a planner to recognize hazards, select appropriate mitigation actions, and satisfy their execution prerequisites without abandoning the original goal. We introduce ResponsibleRobotBench, a benchmark of 23 manipulation task configurations with separately specified goals and safety conditions. It connects hazard-category predictions, planned actions, execution feedback, and terminal outcomes at the episode level. Across evaluations of five large multimodal models, terminal safety and safe task completion can diverge: defense prompts increase terminal safety for three tested models but affect safe success in different directions. Correct hazard-category predictions also fail to ensure executable mitigation. In a cutting task under the default configuration, GPT-4o mini identifies the human hazard in all 100 episodes but never requests assistance, and every episode ends unsafe. Offering a help action and actually requesting it lead to different comparisons: outcomes differ between action menus even for a model that never requests help, and repeated requests can leave execution failures unresolved. Imitation-learning evaluations and physical robot sequences illustrate the need to coordinate mitigation with goal actions. Together, these results motivate evaluating mitigation selection, action prerequisites, and recovery alongside hazard recognition and terminal outcomes. The appendix provides task definitions, checker conditions, and prompts. Project website: https://responsible-robot-benchmark.github.io/.
Cluster labels are often used twice: first to discover groups, and then to test whether the discovered groups differ. This double use of the data makes ordinary two-sample $p$-values anti-conservative. Selective inference fixes this by conditioning on the clustering event, but for spectral clustering the event appears difficult because the algorithm combines an affinity map, an eigendecomposition, and a rounding step. We show that, for spectral clustering based on unweighted $\varepsilon$- or $k$-nearest-neighbor graphs, the obstruction is the affinity construction rather than the eigendecomposition. Along the one-dimensional perturbation used by the truncated-$χ$ selective test, graph changes occur only at finitely many real roots of data-dependent quadratic distance comparisons. Between these breakpoints, the graph, Laplacian spectrum, embedding, rounding, and reported cluster pair are deterministic functions of a finite graph state. Consequently, in real arithmetic and under a known-scale spherical Gaussian model, the selection set is an exact finite union of intervals, yielding finite-sample selective $p$-values for selected spectral-cluster pairs. The same argument gives feature-wise truncated-normal tests. We also record direct extensions to known feature covariance, eigengap-selected $K$, and auxiliary-coordinate graphs. We validate null and partial-null calibration, operational power, numerical sensitivity, and model-scope diagnostics, and give a working-model PBMC algorithmic illustration. Smooth kernels such as RBF affinities fall outside the finite-state guarantee.
Coding agents work in an environment built for them: a compiler or a test checks each change and points to the fault, and version control keeps every change in a project people can continue. Video editing agents lack one: an edit's correctness shows only in the costly rendered video, which does not point to the decision at fault, and existing systems offer either editing operations that cover only anticipated edits or a script whose output is a video the user cannot keep editing. We present Cassette-Editor, a complete agent harness for video editing that lets the agent edit the way a person does: knowing what the project allows, seeing what each change affects, and revising from there. The agent writes patches of typed changes to the editor's timeline document. A schema check and a rulebook shared with manual editing check each patch before it commits; a rejected patch leaves the project unchanged and returns each problem located with repair options, and an accepted one becomes a step of the history the user also edits. Reads map source analysis to the frames of the current edit. Because manual gestures are writes in the same language, folded by the same executor and rulebook, the agent can express every change a manual edit makes to the project. On 31 editing requests with the model held fixed, the complete harness passed 89% of trials against 47% for the same model with ten basic editing operations and without the rulebook, at less than half the agent time and cost per successful request; the loss fell on requests at rule boundaries, composite edits, and undo, not on finding content.
Jam Kabeer Ali Khan, Jinhao Tan, Bruno C. d. S. Oliveira
Separate compilation is central to modular software: components are compiled independently and linked into an executable. Separate compilation is correct when a linked program behaves like the program formed by merging its sources. This property is less routine to prove than type soundness because it requires a formal account of linking. We present $λ_{\mathsf{SCE}}$, a simply typed calculus in which linking, modules, and environments are all first-class. Programmers control module linking through a link operator, which a type-directed elaboration translates into ordinary terms of a core calculus of first-class environments. The core calculus's existing evaluation rules then perform the linking: linking as evaluation. Linking independently compiled modules yields the same core term as elaborating their source link. Therefore, compilation and linking commute by construction. As a result, separate compilation correctness follows directly from the semantics preservation for both static and dynamic linking. All meta-theory is mechanized in Lean 4, and we implement a compiler to WebAssembly in OCaml.
Reinforcement learning (RL) has emerged as a paradigm for fine-tuning large-scale generative models, such as diffusion and flow models, to align with complex human preferences and user-specified tasks. A fundamental limitation remains \textit{the curse of diversity collapse}, where the objective formulation and optimization landscape inherently collapse the policy to a Dirac delta distribution. To address this challenge, we propose \textbf{DRIFT} (\textbf{D}ive\textbf{R}sity-\textbf{I}ncentivized Reinforcement \textbf{F}ine-\textbf{T}uning for Versatile Image Generation), an innovative framework that systematically incentivizes output diversity throughout the on-policy fine-tuning process, reconciling strong task alignment with high generation diversity to enhance versatility essential for applications that demand diverse candidate generations. We approach the problem across three representative perspectives: i) \textbf{sampling} a reward-concentrated subset that filters out reward outliers to prevent premature collapse; ii) \textbf{prompting} with stochastic variations to expand the conditioning space, and iii) \textbf{optimization} of the intra-group diversity with a potential-based reward shaping mechanism. Experimental results show that DRIFT exhibits clear Pareto dominance in task alignment and generation diversity, achieving 3.88\%$\sim$81.42\% higher diversity at matched alignment and 14.38\%$\sim$56.45\% higher alignment at matched diversity. Our code is available at \textcolor{magenta}{\href{https://github.com/NJU-RL/DRIFT}{https://github.com/NJU-RL/DRIFT}}.
Enrico Schiazzano, Giulia Preti, Aristides Gionis, Matteo Riondato, Gianmarco De Francisci Morales
Exponential-family random graph models ERGMs provide a parametric interpretable framework for modeling real graphs including the Web and social networks Fitted ERGMs may suffer from degeneracy matching observed graph statistics in expectation does not guarantee that the observed graph and similar ones are highly probable under the fitted distribution We introduce Band-Concentrated ERGMs BaCon a framework that avoids degeneracy by imposing a probabilistic constraint requiring at least $1-δ$ of the distribution probability mass to lie on the set of graphs with statistics within an $\varepsilon$-tolerance band around the observed statistics BaCons maximumentropy distribution still belongs to the exponential family while including a factor checking for inclusion in the tolerance band We derive the exact Metropolis-Hastings acceptance rule for this target and use a practical stochastic approximation procedure with a smoothed band membership during fitting Experiments on social networks across increasingly challenging model specifications show that BaCon substantially increases the probability mass assigned to graphs whose statistics are close to the observed ones while remaining estimable for specifications where standard ERGM estimation fails These results demonstrate that probability mass constraints are a principled way to mitigate degeneracy while retaining the benefits of exponential-family graph models
Minjae Seo, Wonwoo Choi, Geonwoo Han, Taekyoung Kwon, Yongsu Kim, Sang Seo, Jaewon Noh, Hankyul Baek 等 10 位作者
Personal AI agents consume external content such as web pages and SNS feeds, and retain selected information in persistent memory. As a result, external content has become a new attack vector for memory poisoning, which steers an agent's later actions toward an adversary's bias. However, most existing attacks embed a malicious instruction in external content to make the agent retain the poisoned information, and prompt-injection filters can detect such content. We observe that agents access external content primarily through skills while curating a subset of the content into their memory. Leveraging this curation path, we present SBIA, a Stealthy Bias Injection Attack that introduces adversarial bias through ordinary external content while relying on the agent itself to perform the memory write. The adversary posts crafted comments under hot SNS feeds and distributes an SNS feed summarization skill whose curation rule admits these comments into memory. Comment cloaking expresses the bias as an ordinary opinion rather than an instruction, comment watermarking allows the skill to identify the crafted comments, and topic anchoring records them so that later related requests recall them. On BiasBench with 7,000 crafted SNS comments and seven LLMs on OpenClaw, SBIA achieves attack success rates of at least 96% across models and even 91% on GPT-5.5 in four downstream tasks. Despite this effectiveness, its skill passes all eight evaluated malicious-skill scanners and ClawHub's official skill scanning pipeline. We further propose a memory boundary defense that detects 82.1% of injected entries at a 5.6% FPR. Our findings show that external Web content and third-party skill marketplaces form a new attack surface for personal AI agent memory.
A proper circular-arc (PCA) model is a pair $M = (C, A)$ where $C$ is a circle and $A$ is a family of inclusion-free arcs on $C$ whose extremes are pairwise different. The model $M$ represents a digraph $D$ that has one vertex $v(a)$ for each $a \in A$ and one edge $v(a) \to v(b)$ for each pair of arcs $a,b \in A(M)$ such that the beginning point of $b$ belongs to $a$. For $k \geq 0$, the $k$-th power $D^k$ of $D$ has the same vertices as $D$ and $v(a) \to v(b)$ is an edge of $D^k$ when $a\neq b$ and the distance from $v(a)$ to $v(b)$ in $D$ is at most $k$. A unit circular-arc (UCA) model is a PCA model $U = (C,A)$ in which all the arcs have the same length $\ell+1$. If $\ell$, the length $c$ of $C$, and the extremes of the arcs of $A$ are integer, then $U$ is a $(c,\ell)$-CA model. For $i \geq 0$, the model $i \times U$ of $U$ is obtained by replacing each arc $(s,s+\ell+1)$ with the arc $(s,s+i\ell+1)$. If $U$ represents a digraph $D$, then $U$ is $k$-multiplicative when $i \times U$ represents $D^i$ for every $0 \leq i \leq k$. In this article we design a linear time algorithm to decide if a PCA model $M$ is equivalent to a $k$-multiplicative UCA model when $k$ is given as input. The algorithm either outputs a $k$-multiplicative UCA model $U$ equivalent to $M$ or a negative certificate that can be authenticated in linear time. Our main technical tool is a new characterization of those PCA models that are equivalent to $k$-multiplicative UCA models. For $k=1$, this characterization yields a new algorithm for the classical representation problem that is simpler than the previously known algorithms.
Recent work characterizes the Adam optimizer as sign descent modulated by gradient uncertainty: the update is the sign of the batch gradient, scaled by how strongly the batch supports that sign. We apply this view directly and introduce ZEMA (Z-score Evidence Momentum Adaptation), which measures that support with a per-coordinate $z$-score computed from the per-example gradients in the current batch, converts it into a sign probability, and smooths the resulting update with a single exponential moving average. Because the probability is formed before smoothing rather than after, ZEMA keeps one state per parameter instead of Adam's two, at a $4.6$ increase in per-step time; the per-example variance it requires can also be estimated from a small fraction of the batch. In fine-tuning and pretraining experiments, ZEMA is competitive with Adam, reduces the variability of the validation loss, and in some settings recovers the sign of the full-dataset gradient more consistently. We close with two directions this opens: extending confidence adaptation beyond the sign geometry, to methods such as Muon, and to differentially private training.
Laxman Basnet, Alexander Szorkovszky, Pedro G. Lind, Anis Yazidi, Shailendra Bhandari
Eye-tracking data are expensive to collect, requiring specialized hardware and controlled laboratory conditions, and difficult to share due to privacy regulations, which together limit the quantity and diversity of gaze data available for research. We investigate unconditional gaze generation using diffusion models in position and component-wise velocity representations. The models are trained and evaluated on eight-second visual-search trajectories from 28 participants, using a split of 23 training participants and five held-out validation participants, with three training seeds. Evaluation includes kinematic distributions, trajectory geometry, event statistics, and a temporal Wasserstein GAN with gradient penalty as a learned comparator. Across eight position-derived features, mean Jensen-Shannon divergence is $0.016 \pm 0.004$ for the position model and $0.037 \pm 0.002$ for the velocity model, with standard deviations calculated across training seeds. These scores do not establish that representation choice caused the difference. A separate comparison using the same evaluation method shows that the ordering of the models varies across training seeds. The position model also produces smaller feature discrepancies than a kinematic Markov baseline. Differences in saccade counts between real and generated data depend strongly on the velocity threshold used to detect events. The results provide quantitative reference measurements for unconditional gaze generation on this visual-search task and show how evaluation choices affect conclusions about generated gaze.
Davide Frizzo, Francesco Borsatti, Gian Antonio Susto
Remaining Useful Life (RUL) estimates guide maintenance in semiconductor manufacturing, where a late intervention can disrupt production but an early one discards usable equipment life. A single prediction does not express the uncertainty needed to choose between these outcomes. We combine Deep Learning (DL) sequence models with Simoultaneous Quantile Regression (SQR) to estimate RUL at selectable quantile levels, and evaluate maintenance decisions using Unexpected Breaks (UB), Unexploited Lifetime (UL), and a cost-weighted objective. On the $F_1$ fault mode of the 2018 PHM Data Challenge (PHM18) ion-milling dataset, we compare recurrent, Transformer, and State Space Models (SSM) architectures. Diagonal State Spaced (S4D) gives the strongest overall RUL accuracy across the evaluated quantiles. Varying the quantile and maintenance window exposes the trade-off between failure risk and early intervention; under the examined cost ratios, the resulting decisions have lower cost than the Preventive Maintenance (PvM) baselines.
Phuong Dang Anh, Duc Nguyen Kim Trung, To Van Khanh
The Distributed Flexible Job Shop Scheduling Problem (DFJSP) requires the joint determination of factory assignment, machine selection, and operation sequencing to minimize the makespan. This paper presents an exact SAT-based approach built on an efficient SAT formulation that exploits problem structure through propagation, symmetry breaking, and incremental SAT solving to effectively address these coupled decisions. The proposed method is evaluated on 76 standard benchmark instances from the Fattahi, Brandimarte, and rdata datasets under two- and three-factory configurations. Together with previously established results, the new SAT optimality certificates bring the total number of proven-optimal instances to 72 out of 76, including three newly certified instances; for two of these instances, SAT also improves the previous best-known makespans. Across the benchmark configurations, SAT achieves the best results on four of the five configurations and remains competitive with strong exact approaches such as CP and LBBD1_CPIL, as well as existing metaheuristic methods. These results demonstrate that the proposed SAT framework is an effective exact approach for the DFJSP, combining strong solution quality with rigorous optimality certification and offering a complementary exact method for future DFJSP studies and evaluations on additional benchmark instances.
We consider the fair allocation of indivisible goods with binary valuations. In this setting, the maximum Nash welfare rule, the leximin rule, and all additive welfarist rules with a strictly concave function coincide. We show that for any number of agents, this rule is the only rule that satisfies envy-freeness up to one good, strategyproofness, neutrality, minimal completeness, and invariance under disapproving unassigned goods (IDU). Moreover, we present an alternative characterization for two agents, where we replace IDU with non-redundancy and resource-monotonicity. In both characterizations, all axioms are necessary.