We revisit the output-sensitive enumeration of maximal closed substrings (MCSs). Given a string of length $n$ over an integer alphabet, let $k$ denote its number of MCS occurrences. We combine a right-to-left sweep on the string with range-successor queries to obtain all MCSs with two deterministic word-RAM bounds: for any fixed $\varepsilon>0$, we can report all MCSs in $O(n\sqrt{\log n}+k\log^\varepsilon n)$ time using $O(n)$ words of space, or in $O(n\sqrt{\log n}+k\log\log n)$ time using $O(n\log\log n)$ words. Both bounds include preprocessing and report occurrences by their endpoints. In particular, these algorithms improve the known $O(n\log n)$ worst-case time bound for sufficiently small output.
Generative AI now produces images and poems that people cannot reliably distinguish from human-made work, so people often judge for themselves who made a work. We examine how these perceived-authorship judgments relate to aesthetic preference. In a public online AI Art Turing Test, 11,123 people judged whether each of 50 pictures (26 AI-generated) was made by a human and were asked for their favourite. Among the 9,152 respondents who judged their favourite, a picture was 22.1 percentage points more likely to be judged human-made when it was the respondent's favourite, holding respondent and picture fixed. The association was similar for AI-generated and human-made favourites, was positive for all 50 pictures, and appeared under every reported detection strategy. Respondents most negative about AI art chose fewer AI-generated favourites (47% against 60% among the most positive) but more often believed them human-made, partly because others also judged those pictures human. As a result, in every opinion group a third to two fifths of respondents named an AI-generated favourite they believed human-made. A reanalysis of a published experiment on AI-generated poetry, in which 230 people rated each poem before judging its authorship, found the same pattern, although the AI poems were rated higher. When provenance is hidden, liking and perceived human authorship travel together, consistent with aesthetic response serving as evidence of a human maker and with beliefs about the maker shaping enjoyment. The findings qualify research on bias against AI art, which typically uses explicit labels, and inform AI disclosure policy.
As generative artificial intelligence (AI) settles into everyday practices of inquiry, an old epistemological question returns in a new guise: when does a proposition a user takes from an AI output become something the user knows, and when is it rational to trust it as such? This paper argues that the characteristic failures of AI-mediated inquiry are precisely those that the justified-true-belief (JTB) tradition and its post-Gettier anti-luck refinements were built to diagnose, and it advances three claims. First, AI outputs inherit neither the presumptive warrant of testimony nor the reliabilist warrant of deterministic tools: unaudited trust in a calculator rests on a type-level reliability record, and for general-purpose generative models there is no type-stable process to which such a record could attach -- a practical form of the generality problem. Second, use that stops at obtaining an output turns generative AI into a Gettier machine: the user's fluency-based seeming of justification is disconnected from the truth of what is believed, and, as a clock running ten times too fast illustrates, sheer speed of production multiplies coincidental hits, inflating the appearance of reliability without improving the standing of any single hit. Third, the remedy is a three-step structure of use -- acquisition, understanding, independent verification -- whose first two steps answer to the classical belief and justification conditions and whose third answers to the post-Gettier sensitivity condition. The result is a stakes-indexed criterion of rational trust, justified true output (JTO), with an everyday test: "Had this been false, would I have noticed?" The criterion is defended against objections from over-demandingness, the extended mind, model improvement, and redundancy.
Deep neural networks (DNNs) are widely deployed in safety- and privacy-critical domains such as healthcare and finance, where models are trained on confidential data. This paper presents the first hardware-level attribute-stealing attack that bypasses state-of-the-art privacy defenses, revealing that accelerator optimizations intended for performance can directly undermine data confidentiality. SparseBleed requires no access to gradients, logits, model weights, ground-truth labels, power traces, DVFS manipulation, shared caches, or other channels. Instead, we show that sparsity-driven execution paths encode stable timing signatures that act as a surrogate for model confidence, enabling high-fidelity attribute inference from a single query. SparseBleed generalizes across heterogeneous backends and hardware engines, including Intel AVX2/AVX-512/AMX, NVIDIA Ampere Tensor Cores with 2:4 sparsity, and nine well-known sparse and dense libraries such as Intel oneDNN, oneMKL SparseBLAS, and NVIDIA cuSPARSE. We evaluate SparseBleed to infer sensitive input attributes and true output labels. On Texas-100X, SparseBleed infers sensitive input attributes such as Sex with 66.39% accuracy and recovers the true output label, Surgical Procedure, with 99.8% accuracy. Finally, we evaluate mitigations. Constant-time execution by disabling sparsity optimizations is expensive (e.g., up to $2\times$ higher latency on NVIDIA Sparse Tensor Cores and $\approx 25\%$ higher power in AMX microbenchmarks). By contrast, a padding defense that equalizes responses to the worst-case time masks timing leakage with only 7.24% average performance overhead across nine libraries on Texas-100X/Adult and near-zero extra energy.
Maris Basha, Yuhang Wang, Xiaoran Chen, Longbiao Cheng, Luca Yapura, Anja T. Zai, Mathieu Salzmann, Richard Hahnloser
原文摘要
Deciding which member of an animal group produced a vocalization and where that vocalization was produced are central to studying social communication but are rarely evaluated with physically verified ground truth. Existing resources either localize sounds without identifying the caller, or recognize individuals from isolated vocalizations without reliance on the 3D candidate geometry. We close this gap with V33DA, a dataset-centered multimodal benchmark for spatial vocal attribution in social zebra finches: it comprises 33,625 verified events coupling 5-channel audio with three camera views, calibrated 3D pose for every visible candidate, and per-bird radio telemetry across 10 individuals and three experiments. Caller identity is verified from an on-body accelerometer-derived vibration signal; this channel is withheld from benchmark models and used only by an on-body reference ceiling. Given a detected vocalization and the visible candidate birds, the task is to select the caller. We evaluate it under three regimes: session-disjoint testing on familiar birds from held-out recording dates, held-out-experiment transfer to a new cohort, and leave-one-individual-out evaluation on a bird whose vocalizations are held out from training. Reference methods expose a consistent gap between in-domain accuracy and transfer to unseen individuals: methods that can exploit familiar vocal identity perform well on known callers but degrade sharply on unseen ones, whereas candidate-aware methods preserving explicit spatial reasoning transfer substantially better. The on-body reference stays near-perfect across regimes, confirming the internal consistency of the accelerometer-verified labels. We release V33DA with fixed protocols and code for the evaluated methods, and V33DA++, an unevaluated companion set of verified overlapping-call events and +/-2 s context windows.
Fixed-wing unmanned aerial vehicles experience operating-condition transitions, asynchronous sensor sampling, and control-system compensation that can obscure fault signatures. Existing temporal detection methods have limited capacity to represent coupled faults across subsystems, track their evolution, and constrain language-model diagnoses with quantitative evidence. This paper proposes TGTV, an interpretable fault detection and traceability framework that integrates spatiotemporal graph learning with a large language model. A temporal convolutional network and a graph attention network with weighted fused adjacency extract multiscale temporal patterns and spatial dependencies among sensors. A hybrid temporal Transformer and variational autoencoder then evaluates window-level anomalies using prediction errors and reconstruction probabilities. Continuous-time Shapley attribution identifies critical sensors and high-contribution intervals while preserving the temporal continuity of fault evolution. A structured evidence package combines detection results, graph relationships, temporal attribution, sensor attribution, and operating context. An evidence-constrained diagnosis protocol uses this package to generate flight-level conclusions with traceable evidence identifiers. Cross-flight holdout experiments on the public ALFA fixed-wing flight dataset yield a window-level area under the precision-recall curve of 0.976, accuracy of 0.973, recall of 0.902, and F1 score of 0.920. Flight-level fault diagnosis achieves recall of 0.943 and an F1 score of 0.944. These results demonstrate accurate anomaly detection and quantitatively supported fault-tracing cues, providing a basis for interpretable post-flight fault review and analysis.
Robot-learning studies often compare position and velocity actions as though each label identified a single experimental treatment, yet the label alone does not determine what the robot executes. Before affecting robot motion, bounded policy output passes through filtering, physical scaling, reference construction, feedback, timing, and effort limits. We study how these omitted properties affect attribution under one shared learner and training budget. Across three contact-rich Factory tasks (PegInsert, GearMesh, and NutThread), we compare five joint-space interfaces, with ten independently trained policies per condition. We first remain within the same position-feedback law and vary the executed interface: substituting only the physical action scale as the directly modified component changes sample-median evaluation success by 0.645-1.000 across tasks, while removing an exponential moving average produces task-dependent changes. We then move to a cross-law comparison and choose a velocity scale that exactly matches the position interface's local raw-torque command gain. This calibration raises the sample median on GearMesh and NutThread, with no median change on PegInsert. Yet clear residual sample-median gaps remain on PegInsert and GearMesh, while the NutThread interval includes zero. Thus, even this explicit equality leaves feedback and held-action dynamics unmatched and does not establish position-velocity trained-outcome equivalence. Action-interface comparisons should therefore report the executed pathway and state the physical quantity they match.
Shuo Wen, Beixi Du, Edwin Meriaux, Junming Shi, Chloe Si, David Meger, Doina Precup, Gregory Dudek
原文摘要
We introduce The Prestige, a video understanding benchmark that probes cognitive visual reasoning through professional magic performances, where deliberate misdirection challenges models to track events and explain unexpected outcomes. The benchmark comprises open-ended questions ranging from temporal and causal understanding to robustness against adversarial questions. Evaluation of leading open-source and proprietary VLMs reveals a substantial gap between even the strongest model and human performance. We also find systematic discrepancies between human and automated grading and identify recurring model failure modes. Ablations further show that models can select correct multiple-choice answers even when they fail to generate correct explanations independently, motivating further investigations into question formats. Together, these findings highlight limitations in both VLM reasoning and its evaluation, underscoring the need for more reliable video understanding models and more rigorous benchmarks.
Heejeong Nam, Divyat Mahajan, Quentin Le Lidec, Randall Balestriero, David Klindt
原文摘要
The goal of a world model is to capture the dynamics of an environment. This has the great appeal that it allows agents to plan and simulate a course of actions with the help of a world model. For these uses to be reliable, however, the learned dynamics must behave well beyond the specific state-action combinations observed in expert rollouts. Despite this, the relationship between data coverage, dynamics generalization, and downstream planning performance remains underexplored. We study this relationship through the lens of compositional generalization: can a world model recombine familiar state and action factors into valid dynamics? To answer this, we introduce AnyGoal, a benchmark of goal-shifted task variants built from widely used world-model environments, together with an open-source library. Each task is designed to be within reach of a world model that has learned the underlying dynamics: individual state factors and actions remain within their training support, but their combinations are underrepresented in expert trajectories. Across multiple environments and world model families, we find that current models consistently exhibit failures under these recombinations. By examining dynamics and representation failures across model families, interaction regimes, and evaluation metrics, we study where current world models generalize and where challenges remain for goal-shifted prediction and planning. Together, this framework, benchmark, and analysis provide a basis for evaluating progress toward world models that generalize to new planning tasks.
Md Mahinur Alam, Kanita Jerin Tanha, Md Tanjin Alam, Shadman Sakib, Min Gyu Kim
原文摘要
In recent years, warehouse logistics optimization has increasingly focused on intelligent, data-driven systems to enhance operational efficiency. Conventional static product placement strategies are being progressively replaced by dynamic slotting approaches that can adapt to real-time variations in warehouse conditions. In this work, we present a novel framework that employs a reinforcement learning agent integrated within a FlexSim-based Digital Twin environment for real-time product placement optimization. The proposed agent continuously learns from streaming operational data, enabling it to reduce picking time while improving overall system throughput. The simulation environment models a four-zone warehouse layout with forklift-based material handling and leverages the Asset Administration Shell framework to ensure seamless interaction between the learning agent, physical asset, and the digital twin. Experimental evaluation demonstrates substantial performance gains, with the proposed approach achieving a 29.66\% reduction in average picking time, a 35.42\% increase in throughput, and an 18.54\% decrease in travel distance.
Justin Clarke, Isha Karn, Patrick Murray, KJ Hill, David Jensen
原文摘要
Large language models (LLMs) appear to represent and reason about a large range of phenomena. In particular, recent studies have examined whether LLMs can accurately reason about causation. One of the most commonly used benchmarks for evaluating formal causal reasoning in LLMs is CLadder, a framework for procedural generation of benchmark data sets that was released in 2023. Experiments with CLadder appear to indicate that LLMs can perform formal causal reasoning in a large percentage of generated examples. In this paper, we examine what can be reliably inferred when LLMs perform well on test sets produced by CLadder. We identify several examples of superficial and highly predictive patterns that the CLadder generator embeds in data sets it generates. We find that even simple models such as random forests perform well on CLadder-generated data sets after training on a sample of generated data. We also find that the performance of many commercial LLMs on CLadder test sets is extremely brittle and that removing large sections of the text that are critical for formal causal reasoning has almost no effect on performance. Given these results, the most likely hypothesis for the strong performance of state-of-the-art LLMs on CLadder is that they are exploiting superficial patterns learned from exposure to training data from the generator. Any inference that LLMs are exhibiting more complex causal reasoning must rule out this hypothesis.
When a language model returns engineering quantities with the status "solved", software that trusts it passes them on, even when the requested design is physically impossible. Such failures could reflect poor calculation, how the answer is requested, or a failure to judge feasibility. We set beam loads and disk speeds at known percentages above or below plastic collapse or burst, measured each model's limit calculation separately and tested 12 models in 4,860 calls, including follow-up experiments on new designs. We find that a model can recognize an impossible design when asked only whether it is feasible, yet accept the same design, reporting "solved", when asked for a full answer. Three models accepted beams 10 to 30% beyond collapse despite precise limit calculations in separate questions, and many accepted replies stated the violation. Asked only about feasibility, Opus 5 and Gemini 3.8 Flash rejected every impossible new design they were given, including those accepted in full answers, without false alarms. Rewording when to decline or changing the answer format raised rejection on the original beams, but not consistently on new designs. Software and benchmarks using a model's status should test feasibility judgment and full-answer rejection separately, and report the exact prompt.
We introduce Nemotron 3.5 Super and Nemotron 3.5 Lightning, open and efficient Mixture-of-Experts Hybrid Mamba-Transformer models designed for reasoning, long-context understanding, and agentic workloads. Super is a 120B-parameter, 12B-active multimodal model, while Lightning is a smaller 30B-parameter, 3B-active text model; both designed for efficient deployment. Both models combine large-scale distillation, long-context training, and native speculative decoding, while Super further incorporates Indic continued pretraining and agentic software-engineering mid-training. We develop a comprehensive post-training pipeline spanning supervised fine-tuning and reinforcement learning (RLVR and RLHF), complemented in Super by specialized teacher training and multi- teacher on-policy distillation (MOPD). This transforms Super from a text-only base model into a multimodal agent capable of reasoning over text, images, video, and interactive tool environments. We further explore test-time scaling through parallel agent execution and self-verification. Across training and deployment, we jointly optimize capability and efficiency. Indic continued pretraining improves Indian-language benchmarks by up to 9.82 percentage points, while agentic mid-training strengthens software-engineering capabilities and robustness across agent harnesses. The post-trained Super model achieves 51.78% on SWE-bench Pro, surpassing Nemotron 3 Ultra despite using approximately 4.6x fewer active parameters. Post-training MTP boosting improves speculative- decoding acceptance length by up to 21.8% under probabilistic draft sampling, while NVFP4 quantization achieves 99.62% median accuracy recovery relative to BF16. Inference optimizations enable up to 4,856 output tokens/s/GPU on GB200 while sustaining at least 50 tokens/s/user.
Libya possesses significant renewable energy potential that remains largely untapped. This paper presents PyPSA-Libya, a national-scale model built on the PyPSA-Earth framework, designed to explore optimal transition pathways for the Libyan power sector. We present two primary studies: (1) an electricity-only capacity expansion analysis comparing a zero-emission constraint against an unconstrained baseline at 36-cluster resolution, and (2) a sector-coupled hydrogen export study evaluating the levelized cost of exporting 70 TWh/year of hydrogen to Europe under three technological scenarios and a 100 EUR/tonne CO2 carbon tax regime. Results indicate that while solar PV dominates the domestic capacity mix, integration of hydrogen production significantly alters grid storage requirements. Furthermore, under carbon pricing, steam methane reforming with carbon capture (SMR-CC) frequently out-competes pure green electrolysis, establishing a lower-cost compliance pathway for cross-border export. The model, data, and configuration files are released under an open-source license to support transparent energy planning in Libya.
This study presents AiAWE, an open-source automated writing evaluation system that scores argumentative essays with a LoRA-adapted large language model deployed on local hardware. Using 480 TOEFL Independent Writing essays from a restricted-access Educational Testing Service (ETS) dataset, we fine-tuned Gemma-3-27B and LLaMA-3.3-70B under identical LoRA configurations on 120 essays from one prompt, with all hyperparameters fixed in advance, and evaluated them on the remaining 360 essays. Gemma achieved a root mean square error (RMSE) of 0.474 and a quadratic weighted kappa (QWK) of 0.828 on the half-point grid against the ETS scores. In paired comparisons, it significantly outperformed a fine-tuned GPT-3.5 baseline from prior work on every metric tested, showing that a locally deployable open-weight model can achieve competitive scoring performance in this setting. It also had significantly lower error than the larger LLaMA configuration, although, because the two models differ in family as well as size, this comparison does not isolate the effect of scale. Neither open-weight model showed a statistically detectable increase in error on the 240 essays written to a second, unseen prompt from the same task family. On a single consumer-grade GPU, the scoring model returned a complete response in a median of 14.2 s, demonstrating technical feasibility under the tested conditions. Only the scoring component was evaluated, not the generated feedback. The system is publicly accessible at https://app.awade.gec.waseda.ac.jp, and its code, LoRA adapters, and training configurations are openly released.
Macrocyclic peptides are an emerging modality that combines biologics-like affinity with small-molecule-like developability, yet their vast combinatorial space and multi-parameter objectives make lead optimization slow and challenging. Prior generative approaches for peptide lead optimization (e.g., PepINVENT) require chemists to pre-specify mutable positions for optimization, choices that are not always known a priori, and rely on static pretraining and optimization algorithms that limit the model's ability to generalize and effectively optimize peptide sequences. We introduce PepEVOLVE, a position-aware, dynamic framework that learns both where to edit and how to dynamically optimize peptides for multi-parameter improvement. PepEVOLVE (i) augments pretraining with dynamic masking and CHUCKLES shifting to improve generalization, (ii) uses a context-free multi-armed-bandit router that discovers high-reward residues, and (iii) couples a novel evolving optimization algorithm with group-relative advantage to stabilize reinforcement updates. During in silico evaluations, the router policy reliably learns and concentrates probability on chemically tractable sites that influence the desired peptide's properties. On a therapeutically motivated Rev-binding macrocycle benchmark, PepEVOLVE outperformed PepINVENT by reaching higher averaged permeability-weighted optimization scores of approximately 0.8 vs. 0.6, achieving best candidates with score of 0.97 (vs. 0.89), and converging in fewer steps under the task of optimizing permeability and lipophilicity with structure constraints. Overall, PepEVOLVE offers a practical, reproducible path to linear and macrocyclic peptide lead optimization when optimal edit sites are unknown, enabling more efficient chemical space exploration and improving design quality across multiple objectives to accelerate the design cycles.
Tanvir Hossain, Mahmudul Hasan, Kareem El-Araby, Lorenzo R. Godinez, Swarup Bhunia, Md Tauhidur Rahman, Tamzidul Hoque
原文摘要
Cybersecurity education has predominantly focused on software and network security, under the assumption that the underlying hardware is trustworthy. Threats introduced through the global semiconductor supply chain, together with vulnerabilities arising during design and integration, challenge the assumption that hardware can serve as an unquestioned root of trust. Sensor-enabled systems and machine-learning accelerators further expand the hardware attack surface. The undergraduate curriculum of electrical and computer engineering and computer science programs has not evolved accordingly. Core courses cover the design and the functionality of hardware while the security aspects remain overlooked, and dedicated hardware security (HS) courses are elective, scarce, and often restricted to graduate students. At the University of Florida, fewer than 7% of the undergraduate electrical and computer engineering students register for the introductory HS course each year. We present an integrated and economical approach that disseminates fundamental HS concepts within existing core courses through contextual learning. Nine hands-on modules were developed such that each security concept exercises an associated design concept of the core course, as a component of that course and not as a course of their own. Doing the security work is how the student exercises the design topic, so the modules are built to strengthen the host course and not only to add to it. Three were assessed across three offerings of the embedded systems course through one course survey and two conducted by an external evaluator, with 25 respondents. We report what those instruments measured and the limitations of the approach.
Multimodal large language models (MLLMs) are vulnerable to visual jailbreak attacks that bypass safety safeguards. Existing methods typically optimize isolated adversarial examples, limiting attack diversity, transferability, and generalization. We propose Jailbreak Connectivity (JC), a framework based on the observation that effective adversarial inputs can be connected through low-loss regions in image space. JC optimizes continuous adversarial paths: path sampling yields diverse jailbreak images, lightweight surrogate classifiers guide cross-model transfer, and a distribution-level objective enables universal attacks across harmful queries. On MM-SafetyBench, JC achieves average attack success rates of 84.01%, 85.77%, and 97.95% on Qwen3-VL-8B-Instruct, InternVL3.5-8B, and Gemma-3-12B-IT, respectively, exceeding the strongest baselines by up to 68.36 percentage points. In black-box evaluations, JC improves attack success rates over FigStep by 24.69 and 27.34 percentage points on GPT-5.6 and Gemini 3.1 Pro, respectively. These findings reveal structured vulnerabilities in MLLM safety alignment and inform more comprehensive security evaluation.
Generative engine optimization needs a scorer that ranks candidate pages by the share of an answer they will receive as citations. We measure pretrained rerankers, a hosted typed decision model (Jev), eight open decision models and distilled students on GEO-Bench queries answered by ten engines that receive the same candidate sources in their prompt. Where several engines answer the same query, they agree with one another at a within-group Spearman correlation of 0.704. Jev and two open 27B decision models, used without training, are the strongest scorers at 0.575 to 0.578. A Qwen3-Reranker-4B fine-tuned with low-rank adapters on teacher labels for other GEO-Bench queries, released as GEO Citation Ranker 4B, reaches 0.542 to 0.549 across three seeds, and its differences with the 27B models and with Jev are not distinguishable from zero on 229 queries. Most of the remaining distance between the student and Jev disappears when Jev scores one passage at a time, as the student does. Students trained on teacher labels and on observed engine citations reach the same level, and labels from Jev alone match the two-teacher target at 0.6B. We release the model, the benchmark, the predictions of every reported scorer and the code.
In finite-width deep neural networks, the empirical kernel $G$ evolves stochastically across layers. We develop a collective kernel effective field theory (EFT) for a simplified pre-activation residual model. Exact conditional Gaussianity of the residual increment gives finite-width discrete identities for $G$ and its sigma-kernel source. A projected $G$-only closure and a continuous-depth approximation define reference equations for the mean kernel $K_0$, the kernel covariance $V_4$, and the next-to-leading mean correction $K_1$. Numerical tests show that this closure captures the leading mean-kernel evolution, whereas its covariance and next-to-leading accuracy depend on sigma-kernel modes omitted from the $G$-only state. For the scalar ($N=1$) model with $\tanh$ activation, a Hermite--Galerkin extension agrees with the extrapolated $K_1$ value and reduces the $V_4$ endpoint error by $85$--$93\%$ relative to the $G$-only reference. These results clarify the range of validity of the equal-time Markov $G$-only description and identify the additional collective information required beyond it.
Bayesian change-point and segmentation models provide uncertainty-aware piecewise-constant representations of ordered data, but exact inference is often limited to narrow likelihood classes, single sequences, or index-uniform designs. We present BayesBreak, a modular offline Bayesian segmentation framework that separates local block scoring from global inference: each candidate block supplies a marginal likelihood and any needed moment numerators, while a dynamic program combines these scores to compute posteriors over segment counts, boundaries, and latent signals. For weighted exponential-family likelihoods with conjugate priors, block evidences and posterior moments are available in closed form from cumulative sufficient statistics, enabling exact sum-product inference for p(y|k), p(k|y), boundary marginals, and Bayes regression curves. We distinguish these summaries from the joint MAP segmentation, recovered by a separate max-sum recursion. BayesBreak supports design-aware partition priors for irregular observations, exact pooling across replicates with shared boundaries, and latent-template mixtures with exact EM updates. For non-conjugate GLM blocks, the same DP layer can use deterministic local approximations such as Laplace, variational methods, EP, or quadrature. We prove a posterior-odds stability bound: uniform per-block log-evidence error ε perturbs k-odds and boundary-odds by at most (k+k')ε and 2kε. Validation includes synthetic recovery, calibration, and scaling experiments, plus four real-data illustrations: well-log geology, array-CGH copy number, equity-return volatility, and CpG-atlas methylation.
Model distillation has become a major concern in the rapidly growing generative AI discourse. In its simplest form, distillation is the phenomenon where smaller firms can use data distilled from transcripts of conversations with powerful proprietary closed-weight models to train their own models, thus achieving reasonably high model quality but at much lower costs. Besides the legitimate safety and security concerns associated with distillation, another major issue is how it can also distort marketplace outcomes. This motivates us to initiate an exploration of the problem from an economic theory lens. We consider a setting with two players: an incumbent and a distilling competitor and characterize how the incumbent's model characteristics drive the incentives around how much the competitor should distill. We also consider the impact of distillation on key market outcomes like user-side surplus and the incumbent's utility. Our key contribution, however, is to study interventions for preventing or limiting distillation attacks. In this context, we study two settings: i) when it is possible, with a certain probability (which grows in the extent of distillation conducted), to detect an attack on the incumbent's model; and ii) when a distillation attack cannot be reliably detected. In the first setting, we show that penalty-based regulations can be very effective in completely deterring distillation, although it can have potentially adverse effects on market competition and user surplus. In the latter setting, we show that the incumbent can use self-distillation as a mitigation strategy. Importantly, the effectiveness of self-distillation depends on how cost-efficiently the incumbent can deploy a new self-distilled model on the market. On the positive side, self-distillation can preserve market competition thus protecting user surplus to some extent.
Victor Nikhil Antony, Zhili Gong, Yoonjae Kim, Chien-Ming Huang
原文摘要
We present M, an open-source, low-cost tabletop social robot platform aimed at lowering barriers to social robotics research by making robots easier to reproduce, modify, and deploy in real-world settings. M combines a modular mechanical design, multimodal sensing, and expressive yet mechanically simple actuation architecture with a ROS2-native software package that cleanly separates perception, expression control, and data management. The platform includes a simulation environment with interface equivalence to hardware to support easy sim-to-real transfer of interaction behaviors. We demonstrate extensibility through additional sensing/actuation modules and provide example interaction templates for storytelling and two-way conversational coaching. Lastly, we report real-world use of M in interaction design and week-long in-home deployments, showing how M can serve as a practical platform for longitudinal, reproducible social robotics research.
Single-linkage clustering (SLC) is a fundamental method for hierarchical data analysis. In the distance setting, a $k$-clustering produced by SLC can be obtained by computing a minimum spanning tree (MST) and deleting its $k-1$ heaviest edges. This naturally induces a cost profile for the SLC hierarchy: for each $k\in[n]$, we define $\mathrm{cost}_k$ to be the weight of the resulting $k$-component spanning forest, equivalently, the minimum total weight of any spanning forest with exactly $k$ connected components. The corresponding \emph{SLC cost profile} is $(\mathrm{cost}_1,\ldots,\mathrm{cost}_n)$, and the scalar quantity $\mathrm{cost}(G)=\sum_{k=1}^{n}\mathrm{cost}_k$ is the area under this profile. We study the problem of approximating these quantities in sublinear time. We assume that the input is a weighted graph $G$ of average degree $d$ with edge weights in $\{1,\dots,W\}$, accessed through adjacency-list queries; missing edges are treated as having infinite distance. Our main result is a sampling-based algorithm that outputs a succinct sketch of the entire SLC cost profile in the distance setting. The algorithm runs in $\widetilde{O}(d\sqrt{W}/\varepsilon^3)$ time and returns a sketch from which one can derive estimates $(\widehat{\mathrm{cost}}_1,\ldots,\widehat{\mathrm{cost}}_n)$ satisfying $\sum_{k=1}^{n}\bigl|\widehat{\mathrm{cost}}_k-\mathrm{cost}_k\bigr| \le \varepsilon\,\mathrm{cost}(G)$. Thus, we obtain an $\ell_1$ approximation to the full profile whose error is at most an $\varepsilon$-fraction of the area under the true profile. In particular, this yields a $(1\pm\varepsilon)$-approximation to $\mathrm{cost}(G)$ within the same running time. We also prove a nearly matching lower bound of $Ω(d\sqrt{W}/\varepsilon^2)$ queries for estimating $\mathrm{cost}(G)$.
Single-GPU deployment of 70B-parameter language models on an NVIDIA GPU is constrained by device memory, long-context throughput, and engineering integration cost. We cast single-GPU inference as a budget-aware design problem over these three axes and study how pruning, quantization, and KV-cache compression interact under realistic execution. Controlled ablations show that layer-wise pruning makes weight quantization more robust. KV-cache sparsification complements INT8 KV quantization by reducing memory without hurting decoding speed, while static vector quantizers often conflict with dynamic caching. Guided by these coupling results and explicit budget tracking, we assembled a practical pipeline and compressed a 70B model to about 33 GB, sustained about 57 tokens/s on 10k token prompts on a single A40, and kept absolute accuracy within 5% on common and reasoning benchmarks. We contribute design rules and a reproducible evaluation protocol that jointly report quality, memory, and end-to-end speed, and we provide a foundation for automated pipeline search under realistic single-GPU constraints.
Benchmarks such as WildfireSpreadTS (WSTS) have enabled data-driven forecasting of wildfire spread from multimodal satellite time series, but they suffer from limitations: size restricts them to large fires that burn for weeks, and they are from similar environments. We introduce AFSeq, a multimodal dataset of 1,194 recent fire events (2023 to 2026) from 8 different countries across three regions, dominated by the short, small fires that existing benchmarks underrepresent. Our dataset is the largest and most diverse to date for time-series-based wildfire spread prediction. We also introduce UTAE-AR, an extension of a classical U-Net with Temporal Auto Encoder, enhanced with recurrent modules before the decoder. This allows the model to not only predict the next day but the whole sequence in an autoregressive way. Moreover, our model can feed its own predicted fire mask back one day at a time, producing the day-by-day forecasts that single-step models cannot. Our results show that UTAE-AR surpasses state-of-the-art on both WSTS, WSTS+ and AFSeq for next-day accuracy while having 21.5% of their parameters, and being able to predict confidently at a longer horizon.