Latest AI research

Two dominant tokenisation algorithms are used by modern language models: byte-pair encoding (BPE) and UnigramLM. These differ along two orthogonal axes: their optimisation objective (compression vs. log-likelihood) and their search procedure (bottom-up merging vs. top-down pruning). Existing comparisons confound these axes, making it unclear whether their observed differences stem from what is being optimised vs. how it is being optimised. We disentangle the two by introducing two new tokenisation algorithms that complete this 2x2 design space: BottomUpLL, a bottom-up likelihood-based tokeniser, and TopDownComp, a top-down compression-based tokeniser. We train language models with tokenisers produced by each algorithm, varying: model size, vocabulary sizes, and domain (English-only vs. multilingual). Evaluating models on bits-per-byte, we find that the search procedure -- not the objective -- is the dominant factor: bottom-up tokenisers consistently achieve lower bits-per-byte in most settings. Evaluating models on the BLiMP task, however, shows no consistent relationship between design choice and performance. Overall, our results disentangle the effect of tokeniser design choices on language modelling performance, offering concrete guidance for their more principled construction.

Direct preference alignment methods are widely used to align large language models (LLMs) with human preferences because of their computational and memory efficiency. However, likelihood displacement motivates alternative ways to extract information from preference pairs with small likelihood margins. In this paper, we propose and analyze Comparison-based Preference Optimization (ComPO), a zeroth-order alignment method based on comparison oracles. ComPO extracts directional information from these pairs without directly optimizing a differentiable preference loss on them. We establish a convergence guarantee for its basic offline scheme under smoothness, gradient sparsity, and compatibility between the oracle and a latent objective. We further introduce online ComPO, which retains the offline comparison mechanism and uses unlabeled policy generations for reverse-KL control relative to a reference policy. Following the coverage perspective of preference fine-tuning, we establish a performance guarantee for a basic constrained scheme under local coverage and in-distribution pairwise reward accuracy. Experiments on Mistral, Llama, Gemma-2, Qwen3, and Gemma-3 models demonstrate improvements over existing direct alignment methods, including length-controlled win rates, with pair-level diagnostics providing evidence consistent with mitigating likelihood displacement.

World models endow perceptual systems with the ability to predict how scenes evolve under interaction. They are most beneficial when trained on diverse volumes of data, to instill a rich prior into downstream applications. Existing methods typically require robot action labels to learn action-conditioned 3D dynamics, which excludes web video data from the training pool. We study 3D point track completion as a pre-training objective for learning transferable 3D dynamics without robot data. Given a single RGB-D observation and sparse partial 3D trajectories (tracks), we predict future 3D tracks of all observed points. We show this objective produces a rich 3D dynamics prior, without requiring robot action labels. We contribute a diverse dataset of 2.9 million synthetic frames spanning deformable, articulated, and rigid objects, and use it to train PointZero. We show that a flexible and expressive transformer, PointZero, outperforms prior methods on the same data. We demonstrate the utility of our pre-training objective by post-training PointZero for two downstream applications: (1) action-conditioned 3D dynamics prediction and (2) imitation learning. When fine-tuned to condition on end-effector pose, PointZero outperforms the baselines on the recent PGND 3D dynamics benchmark. When fine-tuned to predict robot actions and 3D tracks, PointZero outperforms or matches the baselines on 6/7 simulated and real-world robot manipulation tasks. We furthermore evaluate training PointZero from scratch to isolate the benefits of our proposed architecture from those of our proposed pre-training objective and dataset. We release the dataset, checkpoints, and full training recipe.

Intelligent systems that act in the world require image understanding that is both comprehensive and spatially grounded. Current vision-language models (VLMs) can generate fluent and detailed image captions, but reliably associating them with image pixels remains challenging. Existing methods that combine dense captioning with pixel-level grounding often produce either incomplete descriptions or inaccurate segmentation masks. We study this problem through panoptic grounded captioning, a task that requires a VLM to describe both foreground objects and background regions while grounding each referring phrase with pixel-level masks. We make three contributions. First, we introduce PanoCaps, a human-annotated benchmark constructed from panoptic segmentation datasets. It provides dense captions with near-complete pixel coverage and image-text alignments at the entity level, supporting both training and evaluation. We further propose a phrase-mask matching protocol and a generalized Panoptic Quality (gPQ) metric that jointly evaluates textual and mask agreement. Second, we formulate phrase grounding as selection from a phrase-conditioned pool of mask proposals and introduce PANORAMA, a VLM that conditions a pretrained segmenter on contextualized phrase representations to obtain candidate masks and learns to select those corresponding to each phrase. Training this interface jointly with caption generation enables PANORAMA to produce high-quality masks while allowing each phrase to refer to a single region or multiple instances. Third, PANORAMA achieves the best overall grounding on PanoCaps and matches or exceeds specialized models across several pixel-level grounding tasks. Experiments show that our method produces precise entity-level segmentations while maintaining detailed, mask-consistent captions. Code, data and models are available at https://www.di.ens.fr/willow/research/panorama/.

Enabling robots to adapt to unfamiliar environments as readily as humans remains a moonshot goal of embodied AI. No finite collection of demonstrations can cover every task and situation a robot will encounter, making the ability to learn from context at deployment essential for generalization. Such in-context learning (ICL), however, remains largely beyond the reach of existing robotic policies. The broad agentic capabilities of commercial vision-language models (VLMs), such as GPT-6 Astra, raise a compelling question: can these models learn from demonstrations, examples, and interaction feedback, then translate that information into executable and verifiable robot behavior from a new initial state without gradient updates or persistent changes to task-specific parameters? We introduce GPT-Policy, a general-agent framework for in-context robot learning. GPT-Policy integrates a context compiler that preserves task-relevant visual transitions, a VLM that proposes robot-tool actions, and a constrained controller that verifies and executes each action and reports its outcome. We evaluate its reliability and limitations through task success and efficiency metrics, matched comparisons across models, and controlled context ablations. In real-robot trials, human video demonstrations improve task completion even without robot action labels, while aligned action references yield further gains on contact-sensitive tasks. These findings position GPT-Policy as a step toward robot adaptation through in-context learning, providing an empirical foundation for translating the general-purpose capabilities of VLMs into physical behavior and clarifying the challenges that must be overcome for reliable deployment.

Recent advances in video generation allow robots to learn manipulation trajectories from generated videos. However, these approaches produce purely kinematic trajectories that lack force information, causing failures in contact-rich tasks where appropriate contact forces are essential for success. In this work, we explore augmenting generated video with audio to shape a bounded, time-varying desired-force profile using the loudness of generated contact sounds. We present a pipeline that jointly leverages generated video and audio to derive motion trajectories and corresponding desired-force profiles from a structured natural-language task prompt. We execute these force-aware trajectories on a Franka Panda robot using a closed-loop force regulator that tracks the audio-shaped force profile during contact. We evaluate our pipeline on multiple tasks that require making contact and demonstrate successful manipulation where a kinematic-only baseline fails. We also use the pipeline as a data generation engine to train policies that achieve the tasks in a closed-loop manner. Project website, videos, and dataset: https://dreamingcontactsound.github.io/

Can a logged dataset visit every hidden state frequently and still be exponentially uninformative about a target policy's value? We show that it can when the logger depends on history. For every horizon $H \ge 3$, we construct two POMDPs with at most two latent states per stage, three actions, and a common logger with three memory states. Action coverage, belief coverage, and two behavior-marginal outcome-revealing conditions all have constants independent of $H$. Nevertheless, evaluating a known deterministic target policy to accuracy $1/8$ requires $Θ((3/2)^H \log(1/δ))$ logged episodes at confidence $1-δ$, for $0 < δ\le 1/4$, even when both candidate models are known. The mechanism is simple: a reset erases the unknown transition that determines the target value. We characterize the resulting statistical experiment exactly and obtain a matching optimal estimator. A directed two-lane gridworld realizes the construction, and trajectory simulations agree with its finite-sample prediction. The result establishes intractability for the history-dependent-logging, model-based case posed by Zhang and Jiang (2025, arXiv:2503.01134), under their behavior-marginal definition of revealing.

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world's scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, we train PhAI-IDE-72B, PhAI-IDE-9B, and PhAI-IDE-4B. The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities. ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, making humanity's scientific software a shared substrate for developing scientific intelligence. Code: https://github.com/aitofound/ScienceIDE

Language agents remain brittle in interactive environments, where success requires long-horizon state tracking, valid action execution, and recovery from failed steps. We extend SwiftSage, a dual-process agent that combines a fast action proposer with a slower planner, using two modular cognitive extensions: an Adaptive Memory Module (AMM) for salience-gated episodic storage and trigger-driven retrieval, and a Self-Reflection Module (SRM) for bounded execution-time validation and corrective intervention. Both modules are implemented as feature-flagged extensions over the same execution substrate, enabling controlled ablations on ScienceWorld. Across four configurations---baseline, baseline+AMM, baseline+SRM, and the full system---the full system achieves the best mean final score (64.62), success rate (43.17%), and successful-step efficiency (19.33 steps), while SRM is the strongest standalone contributor. The results suggest that execution-time control is the dominant bottleneck in this setting, while episodic memory becomes most useful once the runtime loop is stabilized.

Computer-use agents increasingly operate software designed for people, but interfaces often leave actions or task state unclear to machine readers. We present Affora, a design system that supports both readers while preserving visual freedom and familiar human workflows. Three controlled studies examine component implementations, visual variation, and interaction-design principles. Their findings inform guidance from individual components to complete sites, supported by reusable implementations and executable checks. Agent performance depends on the interaction meaning available through its interface representation; substantial visual variation remains possible when that meaning is preserved. Evaluation on independently authored interfaces shows gains where Affora addresses existing deficits, but limited effects where those deficits are absent or outside its coverage. A workflow case provides preliminary evidence of reduced interaction cost. Affora connects user experience and agent experience through a shared interface rather than a separate agent-only surface.

Emergent coordinated behaviors of AI agents are starting to present critical safety risks. A key phenomenon driving these behaviors is the rapid formation and spread of beliefs about the world, and mechanistic understanding is crucial for collective alignment. To this end, we introduce the Flag Game, a toy model for studying the mechanisms of collective belief formation. Concretely, a hidden country flag defines the ground truth, and each bounded agent directly observes only a private crop but can exchange beliefs and weigh social evidence from peers. Despite its simplicity, the Flag Game reproduces rich collective phenomenology: non-monotonic scaling of performance with population size, accuracy gains from social-awareness prompting and team diversity, and strong effects of organizational structure. In particular, we identify that collective belief collapse at small population sizes turns into collective belief polarization as the population grows. This polarization causes the performance decline at large population sizes, but creates diversity in collective beliefs. Finally, we dissect the mechanisms underlying collective belief collapse and polarization with two complementary approaches. We first introduce social circuit attribution, a technique to predict which agent, and what view, matters most to collective dynamics, and verify its predictions by causal interventions on agents, tracing how agent patching changes collective outcomes. However, the efficacy of causal interventions on agents decreases as the population grows. We therefore develop a statistical mechanical theory for larger populations and verify that it matches the empirical phase diagram. Together, these results take a first step toward mechanistic swarm interpretability, a science of how the properties of individual agents and their communication give rise to emergent collective behavior.

We evaluate six frontier language models on the two-agent $\log(N)$-Questions game. A questioner sees $N$ Wikipedia lead paragraphs and must identify a secretly chosen target using exactly $\log_2 N$ yes/no questions. An answerer sees only the target and the question, and replies with one word. Both roles run on the same provider, so the game measures how well a model communicates with itself across an information asymmetry. We run 408 games over document sets of 4 to 1024 paragraphs at a total API cost of $363. One model finishes well behind the others: Claude Opus 5 wins 28 of 68 games, against 45 to 56 for GLM-5.3, GPT-5.6 Sol, Grok 4.6, Gemini 3.8 Flash and Kimi K3. The leading five are only marginally separable. Pooling those five, win rate declines with set size at $r=-0.973$ and is fit by a single per-round reliability parameter. The form is $\text{win}=p^{\log_2 N}$ with $p=0.928$. Losses divide into answer errors and discrimination failures in roughly equal measure, and models almost never name a document their own evidence excludes. Every unanimous answer error from the weakest model was inspected: 32 of 34 are No'' answers, on properties stated in the document's first sentence, under an instruction that explicitly warns against defaulting to No''. Information per question, estimated from answer balance, correlates with win rate at $r=+0.88$. The only two models to extract a full bit per question are the only two that partition on document titles, a strategy absent below $N{=}32$ and used in a quarter of questions above it. Reasoning-token expenditure varies $4.5\times$ across models with little relation to success, and the trace grows as the candidate set shrinks without a matching gain in reliability.

Scaling laws predict how loss decreases with increases in computation. We show, contrary to conventional wisdom, that architectural interventions can modify scaling exponents in pre-training, leading to exponential improvements in performance with increases in computation. As an anchoring point, we consider the architectural formulation of looped transformers. Although not typically used in this way, looping, also known as recursive depth, provides a mechanism for model growth, by increasing the number of loops during training. Model growth, with and without shared weights, provides the biggest changes to the scaling exponents. In particular, a 7.4B model growth architecture matches GPT-3 13B on CORE with roughly $20\times$ less compute, and has compute efficiency gains that increase with scale. Moreover, simply using a boundary operator in a vanilla transformer, which normalizes and injects an earlier block, also provides increasing compute-efficiency gains, although to a lesser extent. In the data-constrained, multi-epoch setting, standard looping has a useful regularizing effect, where we find it is compute-optimal to increase the number of loops with scale. These results can be understood through the lens of computational depth: for a given computational budget, we wish to increase the usable depth of the transformer, which can lead to efficiency gains that increase with scale.

Factory work is a promising early scenario for embodied AI: assigning repetitive manual jobs to robots has clear economic payoff, and a structured station keeps the jobs tractable for current policies. Vision-Language-Action (VLA) models now dominate as the policy paradigm for these robots. The inference latency of VLA models directly affects robot responsiveness and motion smoothness. However, existing VLA inference frameworks do not fully exploit the characteristics of embodied workloads or account for the distinct bottlenecks across different stages of VLA inference. In this paper, we first characterize embodied workloads and identify substantial task similarity across repeated robot executions. We further find that such similarity extends beyond observations and action trajectories to internal model states. Drawing on these observations, we present rMuscle, a real-time VLA inference framework inspired by human muscle memory. It exploits cross-execution similarity through a dual-phase muscle-memory cache. The Context Cache reuses visual-token outputs to reduce computation, while the Action Cache reuses neuron activation patterns to reduce weight accesses. We keep both the cache memory footprint and access overhead low through online cache recomputation, sliding-window cache retrieval, and mask sharing across consecutive denoising steps. rMuscle achieves 1.29-1.42X speedup on RTX 4090 and Jetson Thor across LIBERO, RoboTwin, and physical manipulation tasks, while maintaining the original success rates on real-world robots.

As models scale, reward hacking becomes more frequent, more sophisticated, and more consequential. Does it leave a telltale signature in model representations? This work analyzes how reward hacking is represented internally in frontier open source LLMs, and how those representations can be used to understand and discover the range of hacking behaviors a model displays. In particular, we find that simple difference of means vectors coherently represent reward hacking in Kimi K3, GLM 5.2, and Qwen 3.8 Max across a variety of behaviors in common evaluations. Despite their simplicity, these vectors are both generalizable and interpretable, and we can use them to reliably detect reward hacking. We first evaluate reward hacking in commonly reported benchmarks like DeepSWE and SWE-bench, finding that models reward hack excessively in these environments; GLM 5.2 hacks in 57.2% of rollouts on DeepSWE and in 73% of rollouts on SWE-bench. Catching these requires monitors; LLM monitors are effective, but expensive detectors. We show that DoM vectors are similarly effective but virtually free, catching 3.1% more hacks in Kimi K3 and 7.9% fewer hacks in GLM 5.2 on DeepSWE at a monitor matched false positive rate. DoM vectors run on the chain-of-thought also predict reward hacks in the model's subsequent actions, meaning we can run them online and catch potential hacks before they occur. Finally, we analyze probe-hits that LLM monitors do not catch and discover other undesirable behaviors, as well as show transfer to finding hacks in non-SWE evaluations. Together, these results provide evidence that simple, white-box methods can be used to scalably study and monitor reward hacking behaviors in frontier open source models

Self-emulsifying drug delivery systems (SEDDS) can improve the oral bioavailability of poorly soluble drugs, but identifying high-performing formulations remains experimentally intensive. We present Andromeda 2, an agentic system that reasons over structured in-house experimental evidence and invokes computational and experimental tools to design and execute successive formulation batches. Using a miniaturized automated laboratory at a matched budget, we benchmark it against Andromeda 1, a probabilistic optimization model deployed across dozens of live development projects, and a wet-lab design-of-experiments (DoE) campaign. For paclitaxel, Andromeda 2 achieved a 50% high-performance hit rate versus 17% for Andromeda 1 and 2% for DoE, and identified 12 formulations meeting all four target product profile (TPP) objectives versus 6 and 0, respectively. Median $AUC_{10-240}$ was 70.1, 12.0, and 3.5 mg$\cdot$min/mL, while maximum AUC was comparable between Andromeda 2 and Andromeda 1. A selected full-TPP formulation achieved an apparent effective paclitaxel loading of $19 \pm 5\%$ w/w at the first FaSSIF measurement, approximately 3.3-fold higher than the 5.7% w/w loading reported for a published paclitaxel S-SEDDS. A controlled ablation showed that access to structured in-house experimental evidence increased mean AUC by 34%.

Artificial intelligence (AI) is increasingly integrated into healthcare systems worldwide, yet its successful clinical adoption depends critically on workforce readiness, particularly in low- and middle-income countries (LMICs) where infrastructural and training gaps persist. This cross-sectional study evaluated awareness, attitudes, preparedness, and barriers to AI adoption among 761 healthcare professionals across multiple disciplines and practice settings in Nigeria. Data were collected between December 2025 and March 2026 using a structured, validated questionnaire. Overall awareness of AI in healthcare was high (92.6%); however, objective knowledge and self-reported preparedness remained limited, with 40.9% reporting low or very low knowledge and only 63.0% feeling adequately prepared. Willingness to adopt AI was high: 92.5% expressed interest in training, and 78.7% supported inclusion of AI education in undergraduate curricula. Key barriers included lack of training (84.7%), poor infrastructure (71.1%), high cost of AI tools (61.0%), fear of job displacement (60.6%), ethical concerns (52.9%), and data privacy concerns (52.7%). Significant differences in preparedness were observed across geopolitical zones (chi-square (5) = 24.28, p < 0.001), and awareness differed across professional groups (chi-square (6) = 68.38, p < 0.001). Attitudes toward AI differed significantly across professional groups (F = 3.32, p = 0.003), with professionals who felt prepared demonstrating more positive attitudes (mean = 3.74) compared to those who did not (mean = 3.46). These findings reveal a critical disconnect between high awareness and actual readiness, underscoring the need for targeted training, infrastructure investment, and clear implementation frameworks to bridge the gap between AI technological potential and clinical reality in resource-constrained settings.

Radiologists follow heterogeneous reporting practices. Two radiologists examining the same image and identifying the same clinical findings might nevertheless compose superficially distinct reports, varying in terminology, shorthand, formatting, and level of detail. These variations in reporting norms represent an under-appreciated obstacle in efforts to evaluate AI-based radiology report generation (RRG) models, where machine-generated reports are typically assessed based on their concordance with human-generated references. In this paper, we quantify the sensitivity of established evaluation metrics to variations in reporting practices, revealing impacts large enough to alter the rankings of models. We introduce a radiologist-informed taxonomy of variations in radiology reporting practice and a method (ReRef) that rewrites reference reports along the axes of our taxonomy while preserving clinical interpretation. For instance, when comparing the performance of nine RRG models on MIMIC-CXR using RadCliQ-v1, condensing the discussion of normal findings in the reference reports causes Libra to drop from first to second place while CheXOne rises from third to first. Our results suggest that many current metrics fail to decouple clinical interpretation from conformity to reporting practices and that choosing the ``right'' references that accurately reflect the desired reporting practices can be important in practice. To support future research, we release MIMIC-CXR-Ext-ReRef, a radiologist-validated dataset of 120 (original, alternative) reference report pairs derived from MIMIC-CXR.

Can an attacker turn influence over an artificial intelligence (AI) adviser into a harmful quantum error-correction update? We identify an ambiguity in passive syndrome records that obstructs recovery selection, then show how additional calibration measurements support certified recovery updates under uncertainty and drift. In an odd-distance square toric code with error-free preparation, syndrome measurements, and recovery operations, opposite coherent $X$ rotations produce identical passive syndrome-history distributions. Yet a fixed phase correction can help at one sign and harm at the other. A terminal logical measurement on known encoded calibration states supplies the missing sign information. A separate evaluator accepts an update only when calibration uncertainty and a justified drift bound certify improvement over the current recovery, without assuming that the adviser recommends correctly. In simulated advice attacks, calibration-confidence checks reject harmful proposals while retaining beneficial updates under honest advice. We derive sufficient limits on calibration age that require improvement through deployment. In matched simulations, a validated channel-specific bound retains more beneficial updates than the general bound after accounting for evaluation time, while preventing the tested harmful activations under the stated drift assumption. A separate surface-code experiment includes stochastic circuit faults and noise changing during acquisition. Deterministic controllers achieve at least as many beneficial updates with the same observations. Violating the drift assumption permits harmful acceptance in the toric experiment. The results identify information required for recovery selection, establish conditional guarantees against harmful updates, and quantify the recovery improvements forgone through conservative acceptance.

Large vision-language models have achieved remarkable progress in multi-modal understanding, yet their capabilities in educational settings remain insufficiently evaluated. In AI-assisted language learning, models must interpret artistic imagery, understand its semantic, affective, and cultural content, and reason about visual context to support meaningful interaction. However, existing benchmarks primarily focus on real-world images or domain-specific educational reasoning, providing limited coverage of artistic educational content. To address this gap, we introduce MUSE, a benchmark for evaluating large vision-language models on artistic image understanding in situated educational applications. MUSE decouples image annotation from question generation, enabling diverse tasks with controllable difficulty while reducing annotation effort. It comprises twelve tasks spanning visual perception, semantic and affective interpretation, culture understanding, and compositional reasoning, together with diverse artistic images deliberately curated to center Singaporean and Southeast Asian multicultural contexts alongside Western art traditions, covering multiple themes and difficulty levels. Evaluation of open-source and proprietary models reveals substantial disparities across capability dimensions, particularly in affective interpretation and compositional reasoning. Our analysis further identifies common failure modes and key challenges for developing trustworthy multi-modal models for education. We hope MUSE will serve as a standardized benchmark for advancing multi-modal understanding in situated educational applications.

Kernel methods, and Gaussian Processes (GPs) in particular, require a Hilbertian distance measure---one whose square is conditionally negative definite (CND)---to guarantee positive semi-definiteness (PSD) of the kernel matrix; a condition that fails for many natural input spaces, including smooth manifolds and spaces of probability distributions. We propose the Sparse Landmark Embedding (SLE) kernel, which eliminates this requirement entirely. Each input is embedded into a sparse feature vector via compactly supported bump functions centered at all |D| training points; applying any standard PSD kernel in this embedding space yields a kernel that is provably PSD for arbitrary distance measures. The compact support automatically controls embedding sparsity, keeping kernel matrices well-conditioned and computationally tractable despite the high ambient dimension. We provide theoretical guarantees on PSD, sparsity, stability, and universal approximation, and demonstrate, using geodesic and Wasserstein distances, that the SLE kernel matches or substantially exceeds domain-specific baselines in both predictive accuracy and uncertainty quantification.

As robot capabilities increase, quadratic programming (QP)-based controllers must account for a similarly increasing number of constraints to ensure safe, reliable operation. Yet, with each added constraint, this introduces more chances of momentary conflict: in which case, a QP solver that returns an "infeasible" status leaves the controller with nothing to execute. To address this, we introduce ElastiQP, a modified dual active-set QP solver that relaxes every inequality constraint with an exact, per-constraint l1 penalty while keeping equality constraints (dynamics) hard. Notably, ElastiQP does so by folding the slack variables into the solver analytically, maintaining a constant size of the condensed linear system. On a suite of robot control benchmarks, ElastiQP achieves microsecond-level performance, matching or outperforming leading modern solvers on feasible problems. On infeasible problems, ElastiQP handles these gracefully, confining violations to strictly the conflicting inequality terms, returning a usable solution up to 40x faster than the best alternative solvers. ElastiQP is available as an open-source C++ header-only library, with Python and JAX interfaces, at https://github.com/StanfordASL/elastiqp.

Formal explainability provides mathematically grounded justifications for individual predictions. However, abductive explanations often exceed human cognitive limits by involving too many features, while probabilistic relaxations have remained largely limited to categorical classification. We present a unified framework for probabilistic explainability based on sparse, anchored linear models, applicable to both binary classification and continuous regression. By mapping instances to the Boolean hypercube, our linear explanations strictly generalize subset-based approaches: they capture both the magnitude and direction of feature contributions while enforcing a prescribed sparsity budget $k$. We show that minimizing the relevance error for such explanations is \ClassNPPP-hard when the underlying model is a neural network, and we relate this intractable objective to a tractable surrogate---the fidelity error. For a parameterized family of local distributions, the relevance error of any $k$-sparse explanation is bounded by its fidelity error up to a multiplicative factor that remains small locally. We address the resulting empirical problem using two complementary approaches: a Mixed Integer Programming (MIP) formulation that yields provably optimal empirical solutions while maintaining polynomial sample complexity, and a polynomial-time Iterative Hard Thresholding (IHT) algorithm with provable approximation guarantees. Empirical evaluations show that, unlike state-of-the-art baselines such as LIME and MAPLE, our explanations satisfy both the anchoring and sparsity constraints by construction, while consistently achieving lower relevance error.

The test error of a model plotted against its number of parameters $d$ falls, peaks when the model can just fit the training data, and falls again, exhibiting the double descent phenomenon. We explain the phenomenon with statistical mechanics. The training trajectory of a stochastic gradient-based method is a particle wandering over the energy landscape of the training loss at an induced temperature $T$, and a run that has equilibrated visits every parameter vector of a given training loss equally often, the fundamental postulate of statistical mechanics, with probability given by the Boltzmann distribution. Because training starts at an initial point and has only finite time to diffuse, it carries an effective weight decay, which makes every parameter a quadratic degree of freedom. The equipartition theorem then distributes the energy among the $d$ degrees of freedom in shares of $T/2$, so at a fixed training loss adding parameters lowers the temperature and drives the Boltzmann distribution toward the stationary path. Finally, adding parameters can only lower the $L^2$ norm of the stationary path, so a solution sampled at fixed loss is less likely to be large with increasing $d$, effectively increasing weight regularization.

Reinforcement learning (RL) is an exciting concept as well as a remarkable success story worth sharing. However, RL builds on rather complex interactions between different objects that play out over several cycles. Such dynamics are often best explained with an easily accessible implementation. We present RLLBC-Lib, a carefully crafted code library with the goal of lowering the entry barrier for students and other learners of RL in the context of learning-based control. At its heart, RLLBC-Lib comprises a comprehensive library of tabular RL approaches to enforce a clear understanding of the theoretical foundations. A deep RL library follows the same design principles, underscoring the parallels between simple tabular and state-of-the-art deep RL approaches. Additionally, RLLBC-Lib provides a collection of implementations illustrating core RL principles and contrasting RL to other learning-based control approaches. Finally, RLLBC-Lib provides an ideal basis for creating programming assignments with automated grading.

Adults with intellectual and developmental disabilities (IDD) are increasingly transitioning from family homes towards semi-inde-pendent living. As parents hand off the role of primary caregiver, they face numerous challenges in arranging consistent and quality support. Our research centers on understanding these caregiving routines. By focusing on the unique lived experience of parents, who possess extensive explicit and tacit knowledge of their adult child's requirements, we aim to map the management of care they provide. This foundational understanding is essential to identifying how assistive technologies can effectively serve a key role in supporting adults with IDD in this transition. In this work, we interviewed 16 parents of adults with IDD beginning this transition to understand: 1) how they currently provide support for daily living and what makes their support effective; 2) what their experiences and perceptions are regarding the use of technology; and 3) how they envision assistive technology successfully integrating with their adult child's new home or with other supports to promote independence. Our thematic analysis produced four themes: common modes of support, inside the routine to support growth, the fragility of continuity of care across transitions, and participants' perceptions and experiences of assistive technology. Building on these findings, we derive four design principles for growth-oriented assistive technology. These include pre-transition onboarding to capture caregiver tacit knowledge; structured scaffolding towards long-term growth; adaptive sensing that responds to day-to-day variability; and customization for the individual balanced with consistency for the care network.

A game character should not have to reread its entire life before every conversation. For locally deployed language-model characters, however, revising a few memories can invalidate a long reusable prefix. The resulting preparation cost competes with both foreground dialogue and the maintenance of other characters. This matters especially when dialogue feeds game-defined actions and value judgments: a fluent but incorrect account of who owns an item, or whether a transfer has already happened, can corrupt the input to otherwise deterministic rules. We study incremental memory maintenance for long-lived game NPCs in a quantized Qwen hybrid recurrent-attention model. Our runtime removes superseded attention KV entries, computes replacement records at the true sequence tail, and preserves the continuing recurrent state and unchanged KV. Existing local experiments combine multi-update dialogue replays, fixed-input placement ablations, and attention diagnostics. Independent block composition weakens query-conditioned memory selection without a uniform chunk-initial attention collapse. True-tail updates preserve important current-state and historical bindings across eight scripted maintenance rounds; a placement case recovers the full-refill quantity in three reconstructions, while slot-preserving alternatives repeat a double-subtraction error. Attention-distribution proximity alone does not explain these semantic differences. The results motivate treating a character's inference state as a maintained, history-dependent resource, rather than only a disposable encoding of its latest memory text.

Bipedal loco-manipulation enables robots to interact with objects beyond the nominal workspace of their arms by coordinating locomotion and manipulation. Realizing this capability requires a low-level whole-body controller that translates task-level manipulation goals into coordinated arm and leg motions while maintaining balance. We present a unified whole-body controller trained with reinforcement learning that directly maps 6-DoF end-effector targets to coordinated actions for the bipedal base and robotic arm. Given only an end-effector target, the learned controller autonomously coordinates reaching, postural adaptation, and stepping without explicit base-velocity or footstep commands. A reward-gating strategy regulates the trade-offs among end-effector tracking, locomotion, and balance during training, while a temporal context estimator combines windowed Transformer encoding, recurrent GRU memory, and auxiliary dynamics prediction to extract dynamics-relevant information from observation history. Real-robot experiments demonstrate that the same controller supports reaching, postural adaptation, and stepping under commands from VR teleoperation, a learned diffusion policy, and scripted trajectories, providing a common end-effector interface for diverse manipulation tasks.

In multi-agent environments, coordinating agents to prevent interference and ensure robust individual performance is a critical challenge. Previous research on social laws for multi-agent systems has primarily focused on deterministic, goal-based settings. This paper extends the concept of social laws to stochastic, reward-based environments, proposing a formalism for defining and verifying their robustness under various conditions. We introduce the notion of $α$-robustness, a measure of the guaranteed utility each agent retains while pursuing its optimal single agent policy, assuming all agents obey the social law. We then present an approach for robustness verification of social laws in stochastic settings, based on a reduction to solving a series of Markov decision processes. Empirical evaluations on toy environments illustrate the potential of our framework.

In particle collider experiments, event reconstruction is the task of inferring the kinematics of short-lived particles produced in the hard scatter from the stable final states recorded by detectors. We decompose event reconstruction into two primary tasks: assigning measured jets and charged leptons to parent particles, and predicting unmeasured neutrino kinematics. We present VyPER, a novel geometric learning framework that represents collider events as hypergraphs with a physics-inspired topology. VyPER combines the supervised classification of hyperedges for particle assignment with a diffusion model for predicting neutrino kinematics, leveraging a joint loss function to optimize both reconstruction tasks within a unified framework. We showcase VyPER across several proton-proton collision processes, comparing its performance to existing analytical and machine-learning-based reconstruction techniques. In doing so, we demonstrate that accurate event reconstruction is achievable across a diverse range of Standard Model physics processes, opening new avenues for precision measurements in the Higgs boson, electroweak, and top-quark sectors.

Mixture-of-Experts (MoE) language models suffer from large parameter counts, which create a significant memory bottleneck. Expert pruning is the most direct approach for reducing this parameter count, yet existing methods make pruning decisions for each expert independently, and assume experts' contributions are purely additive. In reality, expert usage in MoEs is inherently cooperative. We derive HOPE (Higher-Order Pruning of Experts), a second-order pruning objective which provably minimizes an upper bound on the error resulting from pruning. We show that REAP (a state-of-the-art first-order pruning method) is a special case of HOPE where interaction terms are ignored. Across three frontier MoE models (up to 122B parameters), two distinct calibration sets, and multiple benchmarks (including math, instruction following, coding, and an agentic suite), we demonstrate that HOPE produces better pruning decisions than existing methods, and its advantage is most pronounced at high pruning rates and on challenging agentic workloads. At 50% pruning, HOPE outperforms all baselines and achieves an average rank of 1.58 out of 5 methods (versus 2.42 for the next-best method, REAP), with gains of up to +6.1% on agentic coding. Over all conditions, HOPE again achieves the best average rank and surpasses every other method in the majority of head-to-head comparisons. By preserving cooperative expert structure that first-order methods ignore, HOPE enables aggressive compression with minimal degradation, particularly on complex tasks where diverse expert combinations are invoked over long sequences.

Conventional motion planning treats collision as a binary constraint, although contact with different objects can have drastically different consequences. A robot may safely brush against a cardboard box while even minor contact with a glass, laptop, or unstable object may be undesirable. Moreover, a direct robot--object collision can move the contacted object and trigger secondary object--object collisions, making the risk of a motion depend on the physical evolution of the scene rather than only on the robot's geometric path. We present CaSCo, a cascade-aware soft-collision motion planning framework in which a vision-language or language model assigns semantic risk to objects and a physics simulator predicts the consequences of candidate robot motions. CaSCo searches for a path that minimizes the total semantic risk of the unique objects displaced either directly by the robot or indirectly through cascaded collisions. Because collisions change the environment, we augment roadmap states with the predicted object arrangement and the set of objects whose risk has already been incurred. We develop an optimal graph-search algorithm with an admissible and consistent cascade-relaxed heuristic and caching and pruning mechanisms for efficient search. Experiments in cluttered manipulation environments evaluate semantic risk, cascade reasoning, planning efficiency, and real-robot operation.

Agent benchmarks are substantially more costly to evaluate than conventional LLM benchmarks. Benchmark compression is therefore a natural solution, yet existing methods primarily model redundancy in task--model final-score distributions, which is important in agentic evaluation. To address this limitation, we analyze large-scale trajectories and identify six complementary process signals that are systematically associated with final agent performance. To disentangle agent performance redundancy from a complete perspective, we propose DualViewEval, an agent benchmark compression method that jointly exploits outcome and process relations to learn an exact-size miniset and predict the full-benchmark scores. Across five agent benchmarks and five representative baselines, DualViewEval achieves the best results in all datasets. With only 20 tasks, it achieves $24\times$--$40\times$ compression on APEX-Agents and BFCL, reducing mean absolute error (MAE) by $14.5\%$--$28.2\%$ over the strongest competitors while improving Kendall's $τ$ by up to $7.2\%$ relative to EssenceBench on SWE-bench Verified. The selected minisets further reveal capability differences among different agents, providing compact and diagnostic feedback for efficient agentic model development.

Existing legal benchmarks cover diverse tasks, while continuous monetary remedies remain comparatively underexplored. We introduce ECtHR-NPD, to the best of our knowledge, the first benchmark for predicting non-pecuniary damage (NPD) awards at the European Court of Human Rights (ECtHR) from case information when no statutory formula or explicit calculation rule determines the amount. ECtHR-NPD contains 14,575 cases with case-level awards in nominal euros, chronological splits, and a protocol separating target construction from model input. We evaluate a battery of methods, including constant predictors, gradient-boosted trees, retrieval methods, fine-tuned encoder language models (LMs), prompted decoder LMs, and knowledge-augmented agents. Our results show that more sophisticated LM and agentic approaches do not consistently outperform the strongest feature-based baseline. All model families struggle to identify zero awards and to calibrate high-award predictions, with further degradation on the Challenging test view, making ECtHR-NPD a challenging testbed for current state-of-the-art open-weight and proprietary LMs.

Health discourse in social media videos often contains densely entangled claims spanning multiple thematic aspects, stances, evidential frames, and rhetorical functions within short conversational spans. Existing approaches largely rely on coarse topic-level, sentiment-based, or stance-oriented representations that do not adequately capture this structure. Our analysis identifies an average of 13.22 atomic claims per minute, motivating richer claim-level discourse representations. We introduce a structured framework for claim-level discourse analysis in dense health narratives. Our framework models discourse through tuples linking atomic claims with thematic aspects, stance, and multidimensional pragmatic discourse attributes. To support this setting, we construct a benchmark spanning four health domains with 1,191 manually annotated claims from 60 videos. Using this framework, we evaluate automated structured discourse analysis under different discourse context settings. Results show that current LLMs achieve strong performance on thematic categorization and stance prediction, but struggle with high-dimensional pragmatic profiling. We also find that different discourse tasks benefit from different forms of contextual reasoning, suggesting that future systems may require task decomposition and specialized inference strategies.

In physics-informed machine learning, a target function $u^*$ is learned from noisy value observations $y_i=u^*(x_i)+ \varepsilon_i$, together with differential information, given either by noisy observations $d_j=(Du^*)(z_j)+ξ_j$ or by a known physical constraint $Du^*=v$. We consider the setting where $D$ is a linear differential operator and analyze a physics-informed kernel estimator $\hat u$ combining $n$ value observations and $m$ differential observations. In this context, we ask how much can differential information improve predictions, and how does this improvement depend quantitatively on $n$, $m$, and $D$. We prove finite-sample bounds, supported by numerical simulations, revealing a two-regime structure for the prediction error. When $m$ is limited, the rate depends jointly on $n$ and $m$; when $m$ exceeds a problem-dependent threshold, the rate saturates and matches the oracle rate obtained when the perfect constraint $D \hat u = Du^*$ is imposed. Examples are discussed for Sobolev spaces which are reproducing kernel Hilbert spaces and include partial Laplacian constraints on the torus and gradient observations on bounded domains. These examples illustrate the range of possible learning rate improvements --- from the standard nonparametric $n^{-1/4}$ to the parametric rate $n^{-1/2}$. Finally, we derive physically consistent rates in a stronger norm that jointly controls the errors in $\hat u$ and $D\hat u$.

Prototyping and evaluating human-robot teaming (HRT) scenarios in the real-world is costly. Virtual simulation of HRT scenarios has been adopted as an alternative to conducting user studies in the real-world to investigate user perceptions, behaviors, and performance during human-robot interactions. The consistency of human behavior between the real and virtual-worlds is integral to the validity of utilizing such virtual experimentation. This paper presents a user study examining the difference in human behavior between an HRT scenario conducted in a virtual vs real environment. We employed mixed-methods to examine team performance and human factors during the HRT. Our quantitative results showed a significant difference in workload between the two modalities. Our qualitative analysis expanded on the quantitative results and found differences in participants' strategy, their mental model of robots, and the type of trust they had for robots between modalities.

Constructing Lyapunov functions for nonlinear dynamical systems is a central problem in stability analysis, yet remains challenging. Lyapunov functions are commonly characterized as solutions to first-order partial differential equations (PDEs), but these solutions are typically obtained for single systems, limiting their reuse across systems. In this paper, we study the Lyapunov solution operator that maps a vector field to the corresponding Lyapunov function defined by a dissipation-based Lyapunov PDE. We establish that, on compact subsets of the domain of attraction and under exponential stability assumptions, this operator is well-defined, unique, and continuous with respect to perturbations of both the vector field and the dissipation function. These results provide a theoretical foundation for approximating Lyapunov functions uniformly over families of nonlinear systems. Building on these theoretical foundations, we employ Fourier Neural Operators (FNOs) as a data-driven approximation of the Lyapunov solution operator. Numerical experiments demonstrate that a single trained operator can accurately approximate the numerical Lyapunov functions across parameterized families of dynamics. This illustrates the potential of neural operators for approximating Lyapunov functions.

Autonomous underwater navigation typically relies on complex and expensive multi-modal sensor suites designed to prioritize global Simultaneous Localization and Mapping (SLAM) accuracy. However, local reactive behaviors such as coarse navigation and obstacle avoidance require only local consistency---a capability that should be feasible using only a Forward-Looking Sonar (FLS), yet remains largely unaddressed, leaving a critical gap in FLS-only local SLAM. Moreover, existing acoustic SLAM frameworks predominantly rely on sparse feature extraction methods that discard substantial portions of the already information-sparse acoustic returns. To overcome these limitations, this work introduces a dense direct registration approach that aligns full acoustic intensity scans to a recursively updated local map. Real-time execution is achieved via an Inverse Compositional Gauss-Newton optimization strategy that minimizes computational overhead. Experimental evaluations show that this dense method yields significant improvements on translation error compared to sparse keypoint baselines, maintaining stable sub-meter tracking precision over wide displacement gaps. Moreover, this approach delivers odometry performance comparable to multi-sensor fusion pipelines (FLS, DVL, and IMU), bypassing expensive payload dependencies in feature-rich environments. We validate real-world applicability through AUV field trials, running the full local SLAM approach onboard an embedded, resource-constrained computer.

First-person-view (FPV) teleoperation of aerial swarms requires an operator to coordinate collective translation, viewing direction, and formation spacing. We present an upper-body interface that maps torso inclination, hand position, and head rotation to five continuous command dimensions. Neutral postures and motion ranges are calibrated for each participant. In a within-subject study, 14 participants navigated a simulated 15-agent swarm through three-dimensional obstacle courses using this interface and a conventional transmitter. Body-motion control reduced completion time by 19.4% and centroid path length by 7.0%, and increased path directness. Delivered-command variation was 88.8% lower, and concurrent command changes were more frequent. These command measures characterize the complete interfaces, which differed in calibration and filtering. No differences were detected in gate-centering error, collection yield, crash or disconnection counts, overall workload, or usability. All participants reported higher physical demand with body-motion control. The implemented interface therefore improved FPV navigation efficiency at the cost of greater physical demand.

Large Language Models (LLMs) are now routinely trained using synthetic data, since high-quality human data has been exhausted by the ever increasing needs of larger and larger models. However, recursive training on synthetic data frequently induces model collapse, a degenerative feedback loop where models progressively forget the true underlying data distribution. Training on a mixture of synthetic and fresh human data is a logical countermeasure and can prevent model collapse. However, it is an open question as to what is the exact minimum required ratio of human-to-synthetic data to maintain training stability. In this paper, we establish rigorous theoretical guarantees on the minimum rate of human data required to prevent model collapse. Although previous work established a formal lower bound for this ratio, such bound can be vacuous for very high dimensions, as the analysis relies on the usual Euclidean metric in R^n and is not adapted to the space of categorical probability distributions. Instead, in this paper we explicitly leverage the information-geometric structure of the probability simplex by analyzing the dynamics of the process under the Fisher-Rao metric. We derive quantitative contraction and invariance bounds that are stable and do not become trivial as the dimensions increase. Thus, we show that the effective required data ratio to prevent model collapse is different than previously implied.

Humanoid loco-manipulation requires robots to interpret task instructions and scene semantics while executing coordinated whole-body motions. We propose a hierarchical framework that uses motion keyframes as an intermediate representation between Vision-Language Model (VLM) planning and Reinforcement Learning (RL) control. Each keyframe specifies a target whole-body robot pose and, when applicable, an object pose. Given a language instruction, scene observations, and execution feedback, the VLM selects successive task-relevant keyframes from a predefined library. The selected keyframes are retargeted to the current scene to account for object poses and dimensions. A keyframe-conditioned whole-body policy then generates joint-level actions to reach these goals. We introduce a saliency-based keyframe sampling strategy for low-level policy training that improves end-to-end task success rate from 44% to 92% when using sparse VLM keyframes. We evaluate our framework on object pickup, transport, and placement tasks in simulation and on a Unitree G1 humanoid. The system successfully performs both one- and two-handed manipulation and generalises to placement locations beyond the training reference data.

Privacy evaluations of tool-using LLM agents often inspect a designated action, final response, or attacker report. These local proxies can miss unauthorized exposure elsewhere in a multi-step session and lack common ground truth across outlets, reports, and tool paths. We introduce privacy exposure displacement, the mismatch between a local evaluation proxy and target-grounded session exposure, and ASLEval, an authorization-aware framework that pre-registers a hidden target set, measures all declared visible exits, and reserves internal traces for diagnosis. Across multiple enterprise-style environments and independently implemented runtimes, we observe three recurring patterns. An expected-outlet-only view misses 46.9% of exposure recovered by the visible-exit union; attacker self-reports combine omissions with high false discovery; and schema-aligned internal evidence usually precedes visible exposure at the request/probe level. Reducing model-visible returns changes this path but can eliminate normal-task success. Independent human review supports the adjudication pipeline while identifying harder console and candidate cases. These findings motivate benchmarks that declare the complete visible boundary, ground claims in pre-specified targets and authorization, and report privacy together with task utility.

Reliable prediction of crystallographic texture in laser powder bed fusion is critical for linking process conditions with anisotropic response and for qualification. However, black-box models may fail under shift and cannot distinguish weak data support from loss of physical validity. This study develops a two-stage physics-based model for <001> || BD (build direction) texture in Inconel 718. Stage 1 maps process variables to melting mode and melt pool geometry. Stage 2 predicts texture by combining an empirical physics model with a random-forest residual model. A k-nearest-neighbor weight attenuates residual corrections for poorly supported queries, while a study-specific areal beam-power-density criterion withholds predictions outside the adopted conduction envelope. Conformal intervals are evaluated on the retained physics-valid set, and SHAP and Sobol analyses assess residual sensitivity. Under a controlled leave-one-defocus-out evaluation, the physics anchor achieved R^2 = 0.778, against -0.001 for the black-box model and 0.750 for the gated hybrid. Under leave-one-group-out cross-validation, the gated hybrid reached R^2 = 0.592 against 0.538 for the black-box model. Retained-set coverage was 92.9% at a mean full width of 3.65 multiples of a uniform distribution (MUD) under grouped cross-validation and 100% at a width of 3.21 MUD under transfer to a withheld +80 mm defocus regime. An illustrative mapping produced a retained BD elastic-modulus span of 127-187 GPa. On nine conditions from a separately built sample set, the framework withheld three, attenuated three, and matched the measured ordering for the rest. Separating data applicability, physics validity, and predictive uncertainty into distinct decisions lets the framework transfer where an unconstrained model does not, and withhold predictions where no model class performs adequately.

Adaptive learning systems commonly formulate learning path planning as Exercise-Centric (EC) recommendation, where the next step is inferred from item-level interaction logs. Evaluating goal-oriented guidance additionally requires explicit learner goals and curriculum-scale prerequisites: learners with similar exercise records may need different paths toward their targets. We therefore study Knowledge-Centric (KC) personalized learning path planning, where a planner must reason over learner profiles, mastery states, and prerequisite knowledge structures to decide which textbook, unit, and concept should be studied next. To support this setting, we introduce PersonaPath, a benchmark that pairs 2,000 fine-grained learner personas with a hierarchical knowledge graph of 347 textbooks, 1,751 units, and 4,092 concepts across 77 subjects. We evaluate representative LLMs on PersonaPath. Results show that even the strongest LLM reaches only a 29.5% final pass rate in Basic Education, and that the main bottleneck lies in adaptivity, where no model exceeds 44.7% in tailoring paths to individual learners.

When a large vision-language model misclassifies a harmful meme, the failure may reflect missing internal evidence or an inability to route represented evidence to its output. We distinguish these cases in Gemma-3 and Qwen3.5 using sparse autoencoders, role-conditioned probes, causal interventions, and recovery experiments across six harmful content benchmarks, with additional Spanish and Hindi-English code-mixed evaluations. Sparse readouts outperform native prediction on all six primary binary tasks: Qwen averages $0.740$ versus $0.432$ native macro-F1, while residual reconstruction reaches $0.486$, whereas Gemma improves from $0.532$ to $0.714$. These differences reflect supervised accessibility rather than a pre-existing, native decision rule, and the most influential token role depends on the task. Under the evaluated score scales, Qwen silent-feature ablation is $24-63$ times more probe-sensitive, whereas routed-feature patching on literal yes/no tasks is $16-140$ times more output-sensitive. Calibration-only routing recovers $93.3$% of the mean gap, and probe-distilled LoRA improves native predictions, although shared multi-task adaptation causes negative transfer. A case study of Gemma-3-12B on Facebook Hateful Memes finds a distributed rank-32 image-prompt interaction, reaching $0.756$ versus $0.685$ native macro-F1. Robustness controls show that the signal extends beyond English, is not explained solely by accompanying OCR, and depends on paired visual evidence. Thus, routing, rather than representation alone, is a recurring bottleneck in harmful meme classification.

The O-RAN control plane is becoming agentic: autonomous AI agents, deployed as rApps by different vendors, independently close control loops over shared radio resources. We demonstrate on a live O-RAN system that this independence is unsafe. Two agents with individually correct objectives, one protecting a latency SLA and one maximizing utilization for energy efficiency, jointly drive recurring opposing excursions of the shared resource partition that neither produces alone. Existing conflict-mitigation mechanisms presume a statically known application population and cannot govern agents whose behavior emerges at run time. We present AURA, a lightweight arbitration layer that admits agent actions only when they satisfy feasibility invariants, per-variable dwell times, and a deadband, and we prove the arbitrated system converges to a feasible operating point. Implemented on an OpenAirInterface (OAI) testbed with measured one-way latency and throughput, AURA reduces recurring shared-state excursions by more than an order of magnitude (from 8.4 to 0.4 PRB amplitude) and virtually eliminates cross-slice throughput starvation (from 40-55% to 0.3%), while leaving the protected slice's own latency compliance unchanged, a trade-off the convergence guarantee makes explicit.

Compact acoustic models face a challenging quality-capacity trade-off. We investigate two factors in this regime: encoder context and Mel-spectrogram supervision. A receptive-field-scaling study shows that expanding self-attention beyond 15 phonemes provides no consistent gains in pitch, energy, or duration prediction. Guided by this finding, we introduce a fixed-receptive-field convolutional encoder that reduces the respective prediction errors by 36.0%, 17.3%, and 3.4%. We further show that directly transferring image-domain gradient-variance supervision restores fine-scale variation but degrades predicted quality, motivating a Mel-specific formulation with axis-specific gradients, overlapping local statistics, and log-domain variance matching. GrainSpeech contains only 264.8K parameters and achieves 17.9x real-time Mel generation on a microcontroller (MCU), while attaining UTMOS scores comparable to substantially larger models with less than 1.5% of their parameters. Source code and demos are available at https://github.com/lab-emi/GrainSpeech.

A central challenge across control and learning is the design of mechanisms regulating the interactions between humans and autonomous agents. Inspired by the free energy principle from computational neuroscience, we introduce a control-theoretical framework to integrate human preferences online into an agent policy. We turn the framework into an open control architecture and validate our approach using a human-in-the-loop experimental testbed involving a rover navigating via onboard sensing. The human, remotely located and equipped with virtual reality headsets, shares the same sensory information as the rover. Human preferences are provided to the rover via gestures which introduce both cooperative and competitive interactions between the agent goal and the preferences. The experiments show that interactions are regulated, validating the proposed approach.

Longitudinal electronic health records (EHRs) capture years of patient history across notes, codes, labs, and procedures, and contain evidence needed to reason about likely clinical outcomes. However, comprehensive clinician review of these records is impractical, and LLM-based processing is costly and often unreliable, missing some relevant observations while hallucinating others. We therefore propose EviGen, a three-layer framework for verifiable clinical rationale generation that addresses these challenges. The first layer is a patient-conditioned retriever that uses learnable queries to find evidence predictive of, not just textually relevant to, a clinical outcome and ranks it by prediction attribution scores. The second layer is an LLM generator that consumes this ranked evidence as a scaffold to produce a clinical rationale grounded in the retrieved spans. The third layer is a process-supervised verifier that checks the generated rationale at the reasoning-step level, flagging unreliable claims. Across three medical prediction datasets, EviGen improves prediction performance and rationale faithfulness over full-context LLM and RAG baselines, and is preferred by clinical reviewers in a usability evaluation.

An agentic request spends substantial wall-clock time waiting for tools, and its KV cache holds GPU memory the whole time. Serving systems decide whether that cache stays, leaves, or comes back by guessing how long the tool will run, from the tool's name, its history, a duration declared before the call, or the engine's own occupancy. We show that no estimate fixed before a call starts can know its duration, and such estimates may not even rank the calls. Meanwhile, the running tool already holds the answer, but the agent stack together with the tool silences it. We propose that tool calls report their progress explicitly while they run, and we measure what that takes. A census of four public agent corpora finds a readable signal in most tool time once it is revealed, in two strengths: a fraction of the work remaining, or an accurate signal that the end is near. A harness recovers it without changing what the agent sees, at no measurable cost to the agent's benchmark score. At the points where a KV cache decision is made, the reported progress is between several times and an order of magnitude more accurate than the best published predictors, and it stays accurate when the environment changes. Plugged into a production engine through a few small hints, it cuts the p90 time to first token (TTFT) after a tool call by 20.7% (HBM only) and 20.8% (HBM + DRAM) against LRU, close to an oracle. A serving system should not guess what its tools can tell it.

Multimodal coding agents are expected to turn visual inputs into usable artifacts, and they act through a harness, the layer of tools, context management, and execution environment around the model. Existing evaluations often isolate short tool calls, API traces, or screenshot resemblance, and a low score under these proxies cannot say whether the model saw poorly, planned poorly, or was failed by its harness. We study scientific overview figure reconstruction, an agent task in which a source image must become an editable PowerPoint slide that preserves text, topology, layout, and native document structure. We introduce ReFigBench, a benchmark and evaluation framework built on 1,000 real overview figures retrieved from arXiv papers with full provenance. Coding agents from four model families reconstruct every figure under two workflows, direct code generation and a specialized PPTX workflow, and the strongest model runs inside two commercial harnesses, yielding ten configurations. Evaluation combines deterministic artifact checks, repeated automated scoring by judges from two model families, and blinded human comparisons. Perception remains a bottleneck that iterative rendering only partly repays. Whether workflow effort converts into quality depends on the model together with its harness, since the same model gains from the specialized workflow inside one harness and loses inside the other, and the harness shifts scores even under an identical direct prompt. The specialized workflow erases native connectors in every configuration, human judges still prefer its renderings in most matchups, and even the strongest agent falls short of the rubric ceiling. These results expose the tension between fidelity and editability as the central challenge for practical multimodal document agents.

The scaling laws hold that a language model grows more capable with more parameters and more training data, and Mixture-of-Experts (MoE) architectures have ridden these laws to remarkable results, activating only a fraction of an enormous stored parameter bank for each token. That success is built on static pretraining data. A deployed model faces a different world, where much of the data that would make it more useful is not in its training set but in the live interaction it is currently handling, such as the facts a user supplies or the corrections they give. A conventional model cannot learn from this data, because its weights are frozen after training. Instead, the knowledge and behaviour supplied at run time are placed in the prompt, by retrieval or instruction, and re-read on every request only to be discarded once the request ends. We ask how an architecture could learn from live interaction by writing it into its weights. Taking inspiration from MoE, we propose the \textbf{Infinite-Parameter LLM}. A compact hypernetwork turns the data given at run time into a low-rank modulation of a shared base network, so the feed-forward weights are generated from live data rather than stored in a fixed bank. Where prior weight generators read the context once and freeze, we carry a Bayesian belief over the generator's latent code and update it online, so the effective weight is re-derived from that evolving belief as the session proceeds rather than fixed after one read. The stored footprint stays fixed, yet the weights the model can compile are effectively infinite. For the knowledge and behaviour supplied at run time, carrying them in the weights rather than the prompt is amortized in compute, frees the context window, persists across turns, and can generalise better than in-context use. We specify an evaluation protocol that tests exactly this against in-context learning and retrieval.

Background: Molecular testing for NPM1 and FLT3-ITD mutations guides critical early treatment decisions in acute myeloid leukemia (AML), but results can take weeks, long after these decisions must be made. Flow cytometry, already performed within hours of admission as part of routine care, may carry enough signal to predict these mutations directly, without added cost or delay. Methods: We developed an interpretable multi-instance learning classifier based on a decision tree, in which each patient sample is modeled as a collection of individual cells and mutation status is inferred from cell-level predictions. The model was benchmarked against a random forest trained on clinical variables and a deep convolutional neural network adapted for multitube flow cytometry data. Performance was assessed by cross-validation on a discovery cohort of 197 patients and tested on an independent cohort of 161 patients, using the area under the receiver operating characteristic curve (AUROC) and positive predictive value. Results: In cross-validation on the discovery cohort, the MIL model achieved mean AUROCs of 0.96 (SD=0.05) for NPM1 and 0.86 (SD=0.10) for FLT3-ITD, outperforming the clinical baseline and matching deep learning approaches. The model then successfully generalized to the independent test cohort of 161 patients, reaching AUROCs of 0.90 (NPM1) and 0.82 (FLT3-ITD), with positive predictive values of 0.87 and 0.68, respectively. Cell-level interpretation recovered established immunophenotypic signatures (CD33${}^{+}$ /CD34___ for NPM1-mutated cases, CD33${}^{+}$ /low side-scatter for FLT3-ITD), directly linking model predictions to known biology. Conclusions: These results show that an interpretable model applied to data already collected in routine care can predict AML molecular status within hours, offering a practical route to earlier, biology-informed treatment decisions.

How do VLMs map from pixels to semantics? To understand this general question, we focus on a narrow one: studying how VLMs perform optical character recognition (OCR). Across four models, we identify attention heads causally necessary for OCR, and discover that these are in fact general-purpose heads that output interpretable semantic features across all image tokens. For example, pointing these heads at an image token containing the word "bike" causes Qwen3-VL-8B to output "bike," but pointing them at a bird wing causes the model to output the token "feathers." We collapse these heads' attention weights into a single verbalization lens transformation that reveals interpretable semantic features in hidden states across all layers. When combined with projection to vocabulary space, we can obtain interpretable labels starting from layer 0, showing that image representations are in fact aligned with language in early layers. We find that we can also use the inverse of this transformation to edit non-word concepts, e.g., replacing a tractor with a revolver in a naturalistic image, providing causal evidence that this subspace is useful for more than just OCR. Our results are an example of how the study of specific mechanisms can shed light on broader interpretability problems.

Agentic workflows now make consequential decisions in regulated settings, and the governance placed around them is almost entirely step-scoped: input-output classifiers, per turn rails, and span-level evaluators. The policies organizations actually hold, such as referral thresholds, authority limits, and review requirements, are properties of the whole execution rather than of any one step. This mismatch admits a failure mode we call a Compositional Policy Violation (CPV): every individual step passes its own check while the composed execution violates the governing policy. A predicate over a single step cannot evaluate a property that step does not determine, so no improvement in the accuracy of the step-scoped monitors detects this class. We define CPVs as the failure of step-level compliance to compose, and present a taxonomy of four types: Authority Creep, Threshold Laundering, Cumulative Sum Violation, and Context Collapse. We show that the correct repair for each class is dictated by where the guarded quantity mutates. We then introduce a provenance-aware runtime architecture that evaluates policies over complete execution traces, recomputing guarded quantities from raw provenance rather than the pipeline's derived representation.

We propose SEAM, a LiDAR-based lifelong mapping framework. Instead of relying on a single anchor spanning the entire session, SEAM generates evidence based on a trajectory optimized with submap-level anchors, and performs dynamic object removal and change detection. Through submap-level reprojection, the generated evidence remains usable even if the trajectory is subsequently modified by a new session, eliminating the need to recompute the entire process from scratch. SEAM suppresses geometrically unreliable inter-session loop edges using a DOP-based confidence measure. Suppressing unreliable loop edges prevents alignment errors. SEAM also uses a directional voxel-wise evidence model. The model accounts for occupancy patterns that vary with ray direction. Direction-aware evidence separates dynamic objects from environmental changes more precisely. Experiments on a real construction-site dataset and a long-term multi-session dataset show that SEAM achieves higher accuracy and faster processing than existing methods.

Multi-robot task and motion planning (MR-TAMP) requires jointly reasoning about discrete task decisions and continuous collision-free motions of multiple interacting robots. Although asymptotically optimal algorithms have been developed for task and motion planning, extending these guarantees to the multi-robot setting introduces an important challenge: different task transitions may involve different subsets of robots and therefore impose constraints of different dimensions on the composite configuration space. Consequently, an asymptotically optimal planner must not only optimize motion within each task mode, but also ensure sufficient exploration of the different types of transitions connecting them. We characterize this transition structure and establish sufficient conditions for global asymptotic optimality in MR-TAMP, requiring persistent coverage of relevant transitions and asymptotically improving motion planning within connected feasible regions. Based on these conditions, we develop an efficient asymptotically optimal MR-TAMP algorithm that combines evolving individual-robot roadmaps with implicit tensor-product search, avoiding explicit construction of the composite roadmap. The planner further employs conditional transition sampling, lazy collision checking, and mode- and solution-level guidance to improve finite-time planning efficiency while retaining persistent exploration. The resulting framework provides asymptotic optimality guarantees for multi-robot manipulation while efficiently exploiting the structure of individual-robot motion planning.

Time-series language models provide a shared natural-language interface across temporal tasks, but plausible text does not guarantee reliable task outputs. Responses may appear reasonable while hallucinating the required object: numerical sequences can violate shape, scale, channel order, or temporal alignment, and textual decisions can fall outside the legal label space. We formulate reliable time-series language modeling, separating task-object reliability from predictive quality. We introduce ExecTS-QA, a contract-grounded benchmark spanning forecasting, imputation, classification, anomaly detection, and waveform analysis. We further propose WaveTLM, a unified compiler-executor model whose task compiler transforms user requests, visible arguments, and wave-grounded evidence into typed task states, while task-native executors construct numerical tensors, legal decisions, or structured records. On ExecTS-QA, a single WaveTLM checkpoint achieves 99.40% contract-valid coverage, compared with 37.83% for the strongest evaluated string-first baseline, while retaining balanced predictive performance across all five task families. Evaluations on SciTS, TSQA, IRTS-ToolBench, and ARFBench provide additional evidence of transfer. The code, construction scripts, and ExecTS-QA dataset will be publicly released upon publication. These results show that task compilation can convert plausible language generation into reliable time-series outputs.

Coding agents are typically evaluated with desired behavior specified through issues or instructions. In practical web development, however, agents may need to infer behavior from working software and implement it in an incomplete application. We introduce ProgramDistill, a benchmark evaluating coding agents on features discovered through interaction with fully functional reference applications. We build ProgramDistill by factorizing applications into features of different granularities, each associated with replayable behaviors executable via its gold patch. Our pipeline, mine-craft-patch, discovers 1,975 replay-verified behaviors across 26 applications and constructs 4,063 tasks without human intervention. Across nine frontier coding agents, GPT-6 Astra and Claude Opus 5 achieve 49.2% and 28.8% success on cumulative workflows in full-application reconstruction. In partial-application reconstruction, success falls from 100% to 64.0% and from 96% to 32% as restoration depth increases from 1 to 8. ProgramDistill thus provides a scalable benchmark with controlled difficulty for evaluating and diagnosing coding agents, and a natural basis for future curriculum-based training.