Science & Mathematics

Goal reasoning (GR) is the process by which intelligent agents continually reason about the goals they are pursuing, which may lead to goal change, for instance when opportunities arise, or when events cause discrepancies among expected and actual outcomes of actions planned to achieve goals. In this paper we investigate whether Hierarchical Task Network (HTN) planning can provide a uniform implementation of the Goal Lifecycle, a flexible process model for goal reasoning in dynamic environments. The Goal Lifecycle progressively refines goals by adding constraints to them through a series of strategies (e.g., {\it formulate, select, expand}) and uses other strategies (e.g., {\it defer, re-expand, repair}) to resolve the current goal when an event impacts it. We present a novel GR agent architecture that employs HTN planning in this Goal Lifecycle. In addition to using HTN planning for the \textit{expand} and \textit{re-expand} strategies, as is frequently done in GR and planning research, we also use it in the \textit{formulate} strategy, to obtain a uniform planning approach that provides several benefits (e.g., the possibility to formulate new goals in view of achieving higher level overarching goals). We also explain how we use an enhanced HTN action representation to process discrepancies among expected and actual action outcomes that can trigger the transition from one goal to another. Finally, we empirically demonstrate that our GR architecture enables simulated autonomous vehicles to adaptively conduct missions in dynamic environments.

Digitally controlled water distribution systems must maintain service during disruptions, yet restoring water delivery does not necessarily ensure hydraulic recovery. This study evaluates cyber-induced pump outages under hypothetical environmental operating stress using two EPANET benchmark networks (Net1 and Net3) simulated with the Water Network Tool for Resilience. The experimental design comprises 288 simulation records (240 distinct configurations), including 192 verified attack scenarios, with four- and twelve-hour outages, six hydraulic configurations, and four mitigation strategies. All attack scenarios satisfied pump-shutdown verification and pre-attack hydraulic-equivalence criteria. Twelve-hour outages produced an average of 404.96 m$^3$ of unmet demand in Net1, while Net3 maintained full modeled demand delivery despite pressure deviations. Sustained pressure recovery occurred in 103 of 192 scenarios, with 65 exhibiting subsequent relapse, whereas all scenarios satisfied the post-outage water-service recovery criterion. Combined mitigation eliminated unmet demand in evaluated Net1 scenarios, while outage shortening achieved the highest sustained-pressure-recovery count. These findings highlight network-dependent resilience and the distinction between service restoration and hydraulic recovery. Environmental stress scenarios represent sensitivity assumptions rather than empirically calibrated climate impacts.

Large language model (LLM) agents that serve users across many sessions must choose, for every request, the few stored memories that enter a limited context window. This choice largely decides whether the agent answers correctly, and it runs on every request at Web scale. This selection faces three challenges: a large and growing store, a per-request cost budget, and multi-session questions that need all of their evidence. Many memory systems build notes or graphs with hundreds to thousands of LLM calls per history, and each is designed on its own, so it is unclear which component yields the accuracy. These challenges mirror the properties of recommendation, so we formalize agent memory as budgeted slate recommendation, with the request as the user, stored memories as items and the context window as a slate. The formalization places memory systems in one design space of recommender cascades and factorizes accuracy into the completeness that each stage controls. Guided by it, we propose AgentMemRec, which adapts candidate generation, learning to rank and slate construction from recommender systems and selects memories without any LLM call. On multi-session questions of LongMemEval, AgentMemRec outperforms memory and context-selection methods, including A-Mem, HippoRAG 2, DynamicRAG and T-Mem, by 10 to 31 points of accuracy. It matches an LLM reranker that makes 64 calls per query at 3.2 times lower serving cost, and on LoCoMo it outperforms Mem0 and Zep without any LLM call at write time. The factorization attributes the gains to slates that more often hold all of the evidence, so recommender-system design offers an effective and efficient foundation for the long-term memory of LLM agents.

Safe household navigation requires robots to regulate clearance and speed according to what an obstacle is and how close it is, without unnecessary conservatism. Yet existing evaluations do not test whether semantic safety distinctions are reflected in continuous motion. To that end, we introduce OSMA-Bench, an Obstacle-Conditioned Safe Motion Adaptation Benchmark built on spatially matched scenario sets that differ only in obstacle type. We also propose the \emph{Semantic Sensitivity Index} (SSI), which measures whether clearance and distance-dependent speed and smoothness follow the caution ordering of obstacles. Evaluating VLM-based and vision-language-action (VLA) policies on the benchmark, we find that neither consistently adapts its motion to obstacle safety, even with explicit safety prompts. Our analysis shows that the policies can correctly distinguish obstacles by safety relevance but fail to reflect this distinction in the motion parameters or actions they produce. For VLM-based policies, which recognize the obstacle and then generate a navigation program, adding an explicit parameter-inference step between these two stages improves adaptation. These findings expose a gap in \emph{semantic-to-motion grounding}, highlighting the need to translate semantic caution into appropriately safe motion.

Large language models (LLMs) are valuable digital assets, yet their weights can be easily modified after unauthorized acquisition, making model provenance difficult to verify. Model fingerprinting provides a promising solution for model ownership protection without modifying the model or embedding ownership information. However, parameter-based fingerprints can be sensitive to model transformations, limiting their robustness, while representation-based fingerprints may assign high similarity to unrelated models, limiting their uniqueness. To address these limitations, we propose TrFP, an intrinsic fingerprinting method based on computational trajectories. Instead of comparing model parameters or representations, TrFP captures how representations evolve across network depth and constructs compact fingerprints using permutation-invariant trajectory descriptors. Experiments across multiple LLM families show that TrFP reliably distinguishes related and unrelated models. It also remains robust under fine-tuning, pruning, quantization, model merging, parameter permutation, and adaptive attacks, while requiring less than 0.3 MiB of fingerprint storage. Overall, TrFP provides stronger provenance discrimination than several existing state-of-the-art methods, demonstrating the effectiveness of computational trajectories for reliable LLM provenance verification.

Hierarchical task knowledge enables efficient problem solving by allowing previously acquired skills to be reused across tasks and problem instances. In this paper, we investigate curriculum-guided acquisition of hierarchical task knowledge in an HTN learning framework. We compare three approaches: online learning during problem solving, human-designed curricula, and an automated curriculum generation procedure that derives learning experiences from verified LLM-generated solution traces. The proposed procedure solves small training problems with an LLM, extracts task accomplishment information, and transforms it into ordered sequences of state-task learning problems for HTN method acquisition. We evaluate the three approaches across multiple planning domains. The results demonstrate that automatically generated curricula recover much of the benefit of expert-designed curricula while substantially reducing the token cost of online hierarchical knowledge acquisition.

Humans have been shown to use a "straightened" encoding to represent the natural visual world as it evolves in time (Hénaff et al. 2019). In the context of discrete video sequences, "straightened" means that changes between frames follow a more linear path in representation space at progressively deeper levels of processing. While deep convolutional networks are often proposed as models of human visual processing, many do not straighten natural videos. In this paper, we explore the relationship between network architecture, differing types of robustness, biologically-inspired filtering mechanisms, and representational straightness in response to time-varying input; we identify strengths and limitations of straightness as a useful way of evaluating neural network representations. We find that (1) adversarial training leads to straighter representations in both CNN and transformer-based architectures but (2) this effect is task-dependent, not generalizing to tasks such as segmentation and frame-prediction, where straight representations are not favorable for predictions; and nor to other types of robustness. In addition, (3) straighter representations impart temporal stability to class predictions, even for out-of-distribution data. Finally, (4) biologically-inspired elements increase straightness in the early stages of a network, but do not guarantee increased straightness in downstream layers of CNNs. We show that straightness is an easily computed measure of representational robustness and stability, as well as a hallmark of human representations with benefits for computer vision models.

Efforts to improve LLM-based agents typically focus on modifying the agent itself, yet success also depends on the environment in which an agent acts. Tools mediate this interaction: a tool's description shapes how an agent calls it, and its implementation determines the results and errors the agent receives afterwards. Many tools, such as APIs and MCP servers, are shared by many agents and maintained by providers who cannot change those agents. We introduce AT-AT (Automatic Tool Optimization from Agent Traces), an agentic tool optimizer that improves a shared toolset offline from historical agent executions while holding the agents fixed. A lightweight coding agent reads agent trajectories and their evaluation results, and selectively edits tool descriptions and, when permitted, implementations. On $τ^2$-bench-verified, AT-AT achieves the highest task completion rates in the Airline and Telecom domains at lower cost than strong prompt- and tool-optimizer baselines, and its Telecom gain is statistically significant under paired permutation tests. Optimizing the tools within a coding agent harness also improves held-out performance on TerminalBench-2 and TBLite. When applied to three production MCP servers at a large enterprise, AT-AT lowered cost or tool error rates on every server and increased held-out task passes on one. By turning accumulated execution traces into updates to shared tool environments, AT-AT provides a path toward continual improvement of agent systems through environment-side learning.

Estimating a population mean is fundamental across scientific applications, but observing every outcome can be prohibitively expensive. Often, covariates and machine learning (ML) predictions are available for the full population, even when only a small subset of outcomes can be observed. How can we leverage this information to proactively select observations to label, update ML predictions as labels arrive, and construct statistically valid confidence intervals? We develop a general, without-replacement formulation of Pro-Active Inference (PAI), extending our earlier with-replacement approach and building on Active Statistical Inference. Rather than making fixed-order label-skip decisions, PAI proactively selects which observation to label next from the remaining covariate pool while accommodating online ML updates. We establish PAI's finite-sample unbiasedness and asymptotic validity, develop an improved finite-sample variance estimator, and extend PAI to superpopulation inference without Bonferroni corrections or $α$-splitting. We characterize PAI's potential efficiency gains over Batch Active Inference in a quasi-oracle setting, highlighting the role of prediction-residual imbalance. Across eight diverse datasets, PAI achieves greater sampling efficiency than Sequential Active Inference in the vast majority of tested settings with near-nominal statistical coverage, using the same ML predictor and uncertainty score. However, comparisons with simpler baselines such as Prediction-Powered Inference and uniform sampling, as well as the closely related Prediction-Powered Active Testing, reveal when additional adaptivity helps or hurts statistical efficiency. Through interpretable, theory-motivated diagnostics and frozen-ML ablations, we investigate when and why PAI gains or loses efficiency, highlighting the initial residual-estimation problem and subsequent online learning dynamics.

Redundant context affects attention through two distinct mechanisms: repeated items add normalization mass, while score variation within correlated groups can produce extreme individual scores. We separate these effects in a planted retrieval model with exponentially many Gaussian distractors arranged in known, exponentially large clusters. The model is a two-level generalized random energy model, whose classical free energy yields exact retrieval thresholds. Gaussian interpolation further shows that increasing positive correlation at fixed token count lowers the distractor free energy, unlike adding duplicates. Our main result identifies a sharp limitation of subtracting log cluster size from pre-softmax scores: this correction removes multiplicity but retains within-cluster extremes. Even a temperature chosen after observing all scores cannot lower the required signal below $\sqrt{2α_1}$, where $α_1$ is the exponential rate of the number of clusters. Averaging scores before exponentiation instead suppresses these fluctuations and achieves $\sqrt{2ρα_1}$, where $ρ$ is the within-cluster correlation. The gap persists after adding exponentially many singleton distractors, so group size alone does not identify the target, and extends to fixed finite hierarchies of light-tailed independent-increment scores. Finite-sample, stability, and participation results complement the asymptotic analysis. A controlled retrieval experiment with frozen sentence embeddings supports this distinction while delineating its scope: averaging helps when the relevant item is a singleton, but can discard useful evidence when a relevant group contains multiple matching items.

Group fairness in retrieval-augmented generation (RAG) systems is typically evaluated using metrics aggregated across queries. However, such metrics can conceal topic-level disparities: a system may favor women on questions about human rights and men on questions about actors and filmmakers, appearing nearly neutral overall even though neither topic is fair. We therefore study topic-level fairness, which requires group fairness to hold within each topic rather than only in aggregate. To reduce the topic-level disparity, we adapt the reranker separately for each topic with a Bradley-Terry objective under direct supervision from the generated output: preference pairs built from the generator's own output carry the direction that topic requires. With preference pairs built from RoMQA alone, the adaptation reduces the mean absolute per-topic disparity by 51% on RoMQA and by 40% on TREC Fair Ranking while retaining accuracy. No baseline reaches the reduction, even when modified to treat each topic separately rather than apply one correction to every topic. Neither does our method with one reranker for all topics, which follows the direction that most topics require.

After each Web search, a search-augmented model must decide whether its evidence suffices: answer, search again, or abstain. Prevailing systems leave this to holistic model judgement, which can accept a context naming every entity yet lacking one required fact. We instead make sufficiency structural: a per-query requirement hypergraph states which facts jointly yield another fact. This raises two challenges: deriving requirements from the question alone, and linking each requirement to passages about the right entity. Certify therefore decomposes the question, searches per requirement, and accepts links only after entity checks. Horn closure over the links yields a certificate, the minimal sufficient passages and their derivation, or an abstention naming open requirements. Given correct retrieval, requirements and links, the decision is sound, and complete up to the entity check. Pooled over three corpora with unanswerable twins, Certify decides significantly more accurately than search agents, graph retrieval and sufficiency judges in default configurations. On the same evidence, it matches a proprietary model's sufficiency label without an additional model call. Over 21M Wikipedia passages, entity checks add 4 to 13 points of decision accuracy where gold evidence exists. Finally, answers from its certificate are 80% correct, against 57% from all retrieved passages.

Language-Only Vision-Language Model Selection (LOVM) aims to identify the Vision-Language Model (VLM) with the highest zero-shot accuracy on a target task using only textual information. We revisit the existing LOVM evaluation protocol and argue that ranking metrics, such as top-5 recall and Kendall's rank correlation, do not directly assess the performance of the selected model. We therefore adopt simple regret (SR), defined as the performance gap between the best and selected models, and introduce Outlier-Robust Normalized Regret (ORNR) to account for the varying performance ranges across tasks while reducing sensitivity to outliers. Building on this revised evaluation protocol, we propose Target-Conditioned Expected Source Regret (TC-ESR), a training-free method that selects the model with the lowest similarity-weighted mean regret across source tasks. Across 22 target tasks and 43 VLMs, TC-ESR achieves the lowest mean SR and ORNR among all compared methods. Separately, we identify two sources of target performance leakage in the prior ranking-based evaluation protocol and provide corrections that remove both.

The solar interior and atmosphere constitute a strongly coupled, multiscale dynamical system defined by continuous temporal evolution. Predicting the state of this system requires modeling physical interactions that span disparate spatial and temporal scales, constrained by heterogeneous and sparse data. This chapter formalizes the Sun within the mathematical framework of dynamical systems theory, defining a global state vector ($\vec{S}$) governed by thermodynamical, kinetic, and magnetic properties. Because solar observations represent projected, multi-modal, and partially observable measurements ($\vec{O}[\vec{S}]$) of this hidden state, pure physics-based deduction is fundamentally constrained. We present a unifying thesis: the primary role of machine learning in heliophysics is to act as a structured family of mathematical operators---specifically Inference ($\mathcal{I}$), Emulation ($\mathcal{E}$), Propagation ($\mathcal{P}$), and Control ($\mathcal{C}$)---that reconstruct, simulate, and couple this global state vector across disparate physical layers. By framing the vast, multi-disciplinary literature within this unified state-space taxonomy, we demonstrate how data-driven techniques bridge observational gaps to resolve hidden multiscale couplings and dramatically accelerate the simulation of non-linear processes, establishing a rigorous pathway to predict the global evolution of the Sun.

Speech interaction systems have gradually evolved from traditional cascaded systems that use text as intermediate information to end-to-end speech language models that directly process speech representations. In some applications, the speech frontend runs on the user's device and sends speech representations to a remote server for inference, while the original audio remains local. However, these representations may still contain speaker identity information. If an attacker obtains them, it may use this information to imitate the user's voice, threatening their voice privacy. To investigate this risk, we propose \stia, a Speech Token Inversion Attack that generates speech with attacker-specified text in a victim's voice using a single short leaked representation, without access to the original audio or its transcript. Specifically, \stia converts discrete, continuous, or mixed speech representations into conditioning sequences and trains a conditional flow-matching speech generation model using pairs of different utterances from the same speaker in public speech datasets. The model combines speaker characteristics contained in the leaked representation with the specified text to generate speech in the victim's voice. Experiments show that, with a Moshi representation leak corresponding to 3 seconds of speech, \stia achieves an average CAM++ speaker cosine similarity of 0.78 and a target-text word error rate of 3.80\%. These results show that speech frontend representations reveal reusable speaker identity information and highlight the need to consider privacy risks when designing speech frontend representation methods.