LLM-based deep research agents are largely built on the ReAct framework. This linear design makes it difficult to revisit earlier states, branch into alternative search directions, or maintain global awareness under long contexts, often leading to local optima, redundant exploration, and inefficient search. We propose Re-TRAC, an agentic framework that performs cross-trajectory exploration by generating a structured state representation after each trajectory to summarize evidence, uncertainties, failures, and future plans, and conditioning subsequent trajectories on this state representation. This enables iterative reflection and globally informed planning, reframing research as a progressive process. Empirical results show that Re-TRAC consistently outperforms ReAct by 15-20% on BrowseComp with frontier LLMs. For smaller models, we introduce Re-TRAC-aware supervised fine-tuning, achieving state-of-the-art performance at comparable scales. Notably, Re-TRAC shows a monotonic reduction in tool calls and token usage across rounds, indicating progressively targeted exploration driven by cross-trajectory reflection rather than redundant search.

Likelihood-based policy gradient methods are the dominant approach for training robot control policies from rewards. These methods rely on differentiable action likelihoods, which constrain policy outputs to simple distributions like Gaussians. In this work, we show how flow matching policy gradients -- a recent framework that bypasses likelihood computation -- can be made effective for training and fine-tuning more expressive policies in challenging robot control settings. We introduce an improved objective that enables success in legged locomotion, humanoid motion tracking, and manipulation tasks, as well as robust sim-to-real transfer on two humanoid robots. We then present ablations and analysis on training dynamics. Results show how policies can exploit the flow representation for exploration when training from scratch, as well as improved fine-tuning robustness over baselines.

Large language models (LLMs) have demonstrated strong reasoning capabilities through step-by-step chain-of-thought (CoT) reasoning. Nevertheless, at the limits of model capability, CoT often proves insufficient, and its strictly sequential nature constrains test-time scalability. A potential alternative is divide-and-conquer (DAC) reasoning, which decomposes a complex problem into subproblems to facilitate more effective exploration of the solution. Although promising, our analysis reveals a fundamental misalignment between general-purpose post-training and DAC-style inference, which limits the model's capacity to fully leverage this potential. To bridge this gap and fully unlock LLMs' reasoning capabilities on the most challenging tasks, we propose an end-to-end reinforcement learning (RL) framework to enhance their DAC-style reasoning capacity. At each step, the policy decomposes a problem into a group of subproblems, solves them sequentially, and addresses the original one conditioned on the subproblem solutions, with both decomposition and solution integrated into RL training. Under comparable training, our DAC-style framework endows the model with a higher performance ceiling and stronger test-time scalability, surpassing CoT by 8.6% in Pass@1 and 6.3% in Pass@32 on competition-level benchmarks.

AI agents often fail in ways that are difficult to localize because executions are probabilistic, long-horizon, multi-agent, and mediated by noisy tool outputs. We address this gap by manually annotating failed agent runs and release a novel benchmark of 115 failed trajectories spanning structured API workflows, incident management, and open-ended web/file tasks. Each trajectory is annotated with a critical failure step and a category from a grounded-theory derived, cross-domain failure taxonomy. To mitigate the human cost of failure attribution, we present AGENTRX, an automated domain-agnostic diagnostic framework that pinpoints the critical failure step in a failed agent trajectory. It synthesizes constraints, evaluates them step-by-step, and produces an auditable validation log of constraint violations with associated evidence; an LLM-based judge uses this log to localize the critical step and category. Our framework improves step localization and failure attribution over existing baselines across three domains.

Large language models (LLMs) perform strongly on many language tasks but still struggle with complex multi-step reasoning across disciplines. Existing reasoning datasets often lack disciplinary breadth, reasoning depth, and diversity, as well as guiding principles for question synthesis. We propose DESIGNER: a DESIGN-logic-guidEd Reasoning data synthesis pipeline that leverages naturally available, extensive raw documents to generate multidisciplinary questions. The central insight is the notion of Design Logic, a form of reusable meta-knowledge that encapsulates the structured process human experts use to transform knowledge into complex exam questions, enabling LLMs to generate new questions with the same complex reasoning patterns from entirely different source texts with explicit control over difficulty, diversity, and question types. We use LLMs to reverse-engineer and abstract over 120,000 Design Logics from existing questions across various disciplines. By designing a two-stage retrieve-and-generate mechanism to match these Design Logics with raw corpus, we synthesized two large-scale reasoning datasets that span 75 disciplines: DLR-Book (3.04 million questions from the book corpus) and DLR-Web (1.66 million questions from the web corpus). Data analysis indicates that the questions synthesized by our method exhibit greater difficulty and diversity compared to those in the baseline datasets. Supervised fine-tuning (SFT) on Qwen3 and Llama3 with our data substantially improves multidisciplinary reasoning and outperforms baseline datasets. Notably, by applying SFT on the base versions of these models using only our data, we even surpass their official final models that have undergone the full post-training.

Enabling humanoid robots to perform agile and adaptive interactive tasks has long been a core challenge in robotics. Current approaches are bottlenecked by either the scarcity of realistic interaction data or the need for meticulous, task-specific reward engineering, which limits their scalability. To narrow this gap, we present HumanX, a full-stack framework that compiles human video into generalizable, real-world interaction skills for humanoids, without task-specific rewards. HumanX integrates two co-designed components: XGen, a data generation pipeline that synthesizes diverse and physically plausible robot interaction data from video while supporting scalable data augmentation; and XMimic, a unified imitation learning framework that learns generalizable interaction skills. Evaluated across five distinct domains--basketball, football, badminton, cargo pickup, and reactive fighting--HumanX successfully acquires 10 different skills and transfers them zero-shot to a physical Unitree G1 humanoid. The learned capabilities include complex maneuvers such as pump-fake turnaround fadeaway jumpshots without any external perception, as well as interactive tasks like sustained human-robot passing sequences over 10 consecutive cycles--learned from a single video demonstration. Our experiments show that HumanX achieves over 8 times higher generalization success than prior methods, demonstrating a scalable and task-agnostic pathway for learning versatile, real-world robot interactive skills.

Progressive Learning (PL) reduces pre-training computational overhead by gradually increasing model scale. While prior work has extensively explored depth expansion, width expansion remains significantly understudied, with the few existing methods limited to the early stages of training. However, expanding width during the mid-stage is essential for maximizing computational savings, yet it remains a formidable challenge due to severe training instabilities. Empirically, we show that naive initialization at this stage disrupts activation statistics, triggering loss spikes, while copy-based initialization introduces gradient symmetry that hinders feature diversity. To address these issues, we propose SPARKLING (balancing {S}ignal {P}reservation {A}nd symmet{R}y brea{K}ing for width-progressive {L}earn{ING}), a novel framework for mid-stage width expansion. Our method achieves signal preservation via RMS-scale consistency, stabilizing activation statistics during expansion. Symmetry breaking is ensured through asymmetric optimizer state resetting and learning rate re-warmup. Extensive experiments on Mixture-of-Experts (MoE) models demonstrate that, across multiple width axes and optimizer families, SPARKLING consistently outperforms training from scratch and reduces training cost by up to 35% under $2\times$ width expansion.

Despite advances in multimodal large language models, autonomous web agents still struggle to reliably execute long-horizon tasks on complex and dynamic web interfaces. Existing agents often suffer from inaccurate element grounding, the absence of site-specific procedural knowledge, and unstable long-term task tracking and memory, particularly when operating over complex Document Object Model structures. To address these limitations, we introduce Avenir-Web, a web agent that achieves a new open-source state of the art on the Online-Mind2Web benchmark in real-world deployment. Avenir-Web leverages a Mixture of Grounding Experts, Experience-Imitation Planning for incorporating procedural priors, and a task-tracking checklist combined with adaptive memory to enable robust and seamless interaction across diverse user interface paradigms. We evaluate Avenir-Web on Online-Mind2Web, a rigorous benchmark of live and user-centered web tasks. Our results demonstrate that Avenir-Web significantly surpasses prior open-source agents and attains performance parity with top-tier proprietary models, thereby establishing a new open-source state of the art for reliable web agents on live websites.

Rapid advancements in large language models (LLMs) have sparked the question whether these models possess some form of consciousness. To tackle this challenge, Butlin et al. (2023) introduced a list of indicators for consciousness in artificial systems based on neuroscientific theories. In this work, we evaluate a key indicator from this list, called HOT-3, which tests for agency guided by a general belief-formation and action selection system that updates beliefs based on meta-cognitive monitoring. We view beliefs as representations in the model's latent space that emerge in response to a given input, and introduce a metric to quantify their dominance during generation. Analyzing the dynamics between competing beliefs across models and tasks reveals three key findings: (1) external manipulations systematically modulate internal belief formation, (2) belief formation causally drives the model's action selection, and (3) models can monitor and report their own belief states. Together, these results provide empirical support for the existence of belief-guided agency and meta-cognitive monitoring in LLMs. More broadly, our work lays methodological groundwork for investigating the emergence of agency, beliefs, and meta-cognition in LLMs.

Activation decomposition methods in language models are tightly coupled to geometric assumptions on how concepts are realized in activation space. Existing approaches search for individual global directions, implicitly assuming linear separability, which overlooks concepts with nonlinear or multi-dimensional structure. In this work, we leverage Mixture of Factor Analyzers (MFA) as a scalable, unsupervised alternative that models the activation space as a collection of Gaussian regions with their local covariance structure. MFA decomposes activations into two compositional geometric objects: the region's centroid in activation space, and the local variation from the centroid. We train large-scale MFAs for Llama-3.1-8B and Gemma-2-2B, and show they capture complex, nonlinear structures in activation space. Moreover, evaluations on localization and steering benchmarks show that MFA outperforms unsupervised baselines, is competitive with supervised localization methods, and often achieves stronger steering performance than sparse autoencoders. Together, our findings position local geometry, expressed through subspaces, as a promising unit of analysis for scalable concept discovery and model control, accounting for complex structures that isolated directions fail to capture.

Large Language Models (LLMs) often struggle with deductive judgment in syllogistic reasoning, systematically conflating semantic plausibility with formal validity a phenomenon known as content effect. This bias persists even when models generate step-wise explanations, indicating that intermediate rationales may inherit the same semantic shortcuts that affect answers. Recent approaches propose mitigating this issue by increasing inference-time structural constraints, either by encouraging abstract intermediate representations or by intervening directly in the model's internal computations; however, reliably suppressing semantic interference remains an open challenge. To make formal deduction less sensitive to semantic content, we introduce a framework for abstraction-guided reasoning that explicitly separates structural inference from lexical semantics. We construct paired content-laden and abstract syllogisms and use the model's activations on abstract inputs to define an abstract reasoning space. We then learn lightweight Abstractors that, from content-conditioned residual-stream states, predict representations aligned with this space and integrate these predictions via multi-layer interventions during the forward pass. Using cross-lingual transfer as a test bed, we show that abstraction-aligned steering reduces content-driven errors and improves validity-sensitive performance. Our results position activation-level abstraction as a scalable mechanism for enhancing the robustness of formal reasoning in LLMs against semantic interference.

Federated learning (FL) has emerged as a promising distributed machine learning (ML) that enables collaborative model training across clients without exposing raw data, thereby preserving user privacy and reducing communication costs. Despite these benefits, traditional single-server FL suffers from high communication latency due to the aggregation of models from a large number of clients. While multi-server FL distributes workloads across edge servers, overlapping client coverage and uncoordinated selection often lead to resource contention, causing bandwidth conflicts and training failures. To address these limitations, we propose a decentralized reinforcement learning with conflict risk prediction, named RL CRP, to optimize client selection in multi-server FL systems. Specifically, each server estimates the likelihood of client selection conflicts using a categorical hidden Markov model based on its sparse historical client selection sequence. Then, a fairness-aware reward mechanism is incorporated to promote long-term client participation for minimizing training latency and resource contention. Extensive experiments demonstrate that the proposed RL-CRP framework effectively reduces inter-server conflicts and significantly improves training efficiency in terms of convergence speed and communication cost.

Large language models can generate fluent explanations, but effective tutoring requires supporting the learner's thought process, not just delivering content. Metacognitive tutoring targets this gap by prompting planning, monitoring, debugging, and evaluation, and crucially, deciding when to be active versus minimally present, based on learner signals and trajectory. We introduce MetaCLASS, a learning-science grounded framework that formulates metacognitive tutoring as move selection over 11 interpretable actions aligned to self-regulated learning processes. MetaCLASS uses a two-phase framework that first plans a pedagogical trajectory conditioned on learner profiles (calibration, help-seeking) and then generates natural dialogue consistent with that plan. This yields a dataset of 1,015 conversations (7,711 turns) annotated with turn-level metacognitive labels, and validated for pedagogical contingency and trajectory adherence. We benchmark nine LLMs on predicting the next coach move given the problem and dialogue context. The best model achieves only 43.2\% accuracy, and models exhibit compulsive intervention bias: in turns where effective metacognitive tutoring requires silent (41.7\% of cases), models predict `no intervention' only 4.2\% of the time, while severely over-predicting high-intervention moves. These results show that traditional content-based tutoring ability does not translate to metacognitive tutoring competence, positioning MetaCLASS as a testbed for developing intelligent tutors that promote self-regulated learning.

Representing and understanding 3D environments in a structured manner is crucial for autonomous agents to navigate and reason about their surroundings. While traditional Simultaneous Localization and Mapping (SLAM) methods generate metric reconstructions and can be extended to metric-semantic mapping, they lack a higher level of abstraction and relational reasoning. To address this gap, 3D scene graphs have emerged as a powerful representation for capturing hierarchical structures and object relationships. In this work, we propose an enhanced hierarchical 3D scene graph that integrates open-vocabulary features across multiple abstraction levels and supports object-relational reasoning. Our approach leverages a Vision Language Model (VLM) to infer semantic relationships. Notably, we introduce a task reasoning module that combines Large Language Models (LLM) and a VLM to interpret the scene graph's semantic and relational information, enabling agents to reason about tasks and interact with their environment more intelligently. We validate our method by deploying it on a quadruped robot in multiple environments and tasks, highlighting its ability to reason about them.

As Large Language Models transition to autonomous agents, user inputs frequently violate cooperative assumptions (e.g., implicit intent, missing parameters, false presuppositions, or ambiguous expressions), creating execution risks that text-only evaluations do not capture. Existing benchmarks typically assume well-specified instructions or restrict evaluation to text-only, single-turn clarification, and thus do not measure multi-turn disambiguation under grounded execution risk. We introduce \textbf{Drift-Bench}, the first diagnostic benchmark that evaluates agentic pragmatics under input faults through multi-turn clarification across state-oriented and service-oriented execution environments. Grounded in classical theories of communication, \textbf{Drift-Bench} provides a unified taxonomy of cooperative breakdowns and employs a persona-driven user simulator with the \textbf{Rise} evaluation protocol. Experiments show substantial performance drops under these faults, with clarification effectiveness varying across user personas and fault types. \MethodName bridges clarification research and agent safety evaluation, enabling systematic diagnosis of failures that can lead to unsafe executions.

Robot learning from interacting with the physical world is fundamentally bottlenecked by the cost of physical interaction. The two alternatives, supervised finetuning (SFT) from expert demonstrations and reinforcement learning (RL) in a software-based simulator, are limited by the amount of expert data available and the sim-to-real gap for manipulation. With the recent emergence of world models learned from real-world video-action data, we ask the question of whether training a policy in a world model can be more effective than supervised learning or software simulation in achieving better real-robot performance. We propose World-Gymnast, which performs RL finetuning of a vision-language-action (VLA) policy by rolling out the policy in an action-conditioned video world model and rewarding the rollouts with a vision-language model (VLM). On the Bridge robot setup, World-Gymnast outperforms SFT by as much as 18x and outperforms software simulator by as much as 2x. More importantly, World-Gymnast demonstrates intriguing capabilities of RL with a world model, including training on diverse language instructions and novel scenes from the world model, test-time training in a novel scene, and online iterative world model and policy improvement. Our results suggest learning a world model and training robot policies in the cloud could be the key to bridging the gap between robots that work in demonstrations and robots that can work in anyone's household.

We present a robust safety-critical control framework tailored for networked susceptible-infected-recovered (SIR) epidemic dynamics, leveraging control barrier functions (CBFs) and robust control barrier functions to address the challenges of epidemic spread and mitigation. In our networked SIR model, each node must keep its infection level below a critical threshold, despite dynamic interactions with neighboring nodes and inherent uncertainties in the epidemic parameters and measurement errors, to ensure public health safety. We first derive a CBF-based controller that guarantees infection thresholds are not exceeded in the nominal case. We enhance the framework to handle realistic epidemic scenarios under uncertainties by incorporating compensation terms that reinforce safety against uncertainties: an independent method with constant bounds for uniform uncertainty, and a novel approach that scales with the state to capture increased relative noise in early or suppressed outbreak stages. Simulation results on a networked SIR system illustrate that the nominal CBF controller maintains safety under low uncertainty, while the robust approaches provide formal safety guarantees under higher uncertainties; in particular, the novel method employs more conservative control efforts to provide larger safety margins, whereas the independent approach optimizes resource allocation by allowing infection levels to approach the boundaries in steady epidemic regimes.

Discovering causal relationships requires controlled experiments, but experimentalists face a sequential decision problem: each intervention reveals information that should inform what to try next. Traditional approaches such as random sampling, greedy information maximization, and round-robin coverage treat each decision in isolation, unable to learn adaptive strategies from experience. We propose Active Causal Experimentalist (ACE), which learns experimental design as a sequential policy. Our key insight is that while absolute information gains diminish as knowledge accumulates (making value-based RL unstable), relative comparisons between candidate interventions remain meaningful throughout. ACE exploits this via Direct Preference Optimization, learning from pairwise intervention comparisons rather than non-stationary reward magnitudes. Across synthetic benchmarks, physics simulations, and economic data, ACE achieves 70-71% improvement over baselines at equal intervention budgets (p < 0.001, Cohen's d ~ 2). Notably, the learned policy autonomously discovers that collider mechanisms require concentrated interventions on parent variables, a theoretically-grounded strategy that emerges purely from experience. This suggests preference-based learning can recover principled experimental strategies, complementing theory with learned domain adaptation.

A central decision problem in Petri net theory is reachability asking whether a given marking can be reached from the initial marking. Related is the covering problem (or sub-marking reachbility), which decides whether there is a reachable marking covering at least the tokens in the given marking. For live and bounded free-choice nets as well as for sound free-choice workflow nets, both problems are polynomial in their computational complexity. This paper refines this complexity for the class of sound acyclic free-choice workflow nets to a quadratic polynomial, more specifically to $O(P^2 + T^2)$. Furthermore, this paper shows the feasibility of accurately explaining why a given marking is or is not reachable. This can be achieved by three new concepts: admissibility, maximum admissibility, and diverging transitions. Admissibility requires that all places in a given marking are pairwise concurrent. Maximum admissibility states that adding a marked place to an admissible marking would make it inadmissible. A diverging transition is a transition which originally "produces" the concurrent tokens that lead to a given marking. In this paper, we provide algorithms for all these concepts and explain their computation in detail by basing them on the concepts of concurrency and post-dominance frontiers - a well known concept from compiler construction. In doing this, we present straight-forward implementations for solving (sub-marking) reachability.

This paper derives non-asymptotic error bounds for nonlinear stochastic approximation algorithms in the Wasserstein-$p$ distance. To obtain explicit finite-sample guarantees for the last iterate, we develop a coupling argument that compares the discrete-time process to a limiting Ornstein-Uhlenbeck process. Our analysis applies to algorithms driven by general noise conditions, including martingale differences and functions of ergodic Markov chains. Complementing this result, we handle the convergence rate of the Polyak-Ruppert average through a direct analysis that applies under the same general setting. Assuming the driving noise satisfies a non-asymptotic central limit theorem, we show that the normalized last iterates converge to a Gaussian distribution in the $p$-Wasserstein distance at a rate of order $γ_n^{1/6}$, where $γ_n$ is the step size. Similarly, the Polyak-Ruppert average is shown to converge in the Wasserstein distance at a rate of order $n^{-1/6}$. These distributional guarantees imply high-probability concentration inequalities that improve upon those derived from moment bounds and Markov's inequality. We demonstrate the utility of this approach by considering two applications: (1) linear stochastic approximation, where we explicitly quantify the transition from heavy-tailed to Gaussian behavior of the iterates, thereby bridging the gap between recent finite-sample analyses and asymptotic theory and (2) stochastic gradient descent, where we establish rate of convergence to the central limit theorem.

Large Language Models (LLMs) have remarkable capabilities across NLP tasks. However, their performance in multilingual contexts, especially within the mental health domain, has not been thoroughly explored. In this paper, we evaluate proprietary and open-source LLMs on eight mental health datasets in various languages, as well as their machine-translated (MT) counterparts. We compare LLM performance in zero-shot, few-shot, and fine-tuned settings against conventional NLP baselines that do not employ LLMs. In addition, we assess translation quality across language families and typologies to understand its influence on LLM performance. Proprietary LLMs and fine-tuned open-source LLMs achieve competitive F1 scores on several datasets, often surpassing state-of-the-art results. However, performance on MT data is generally lower, and the extent of this decline varies by language and typology. This variation highlights both the strengths of LLMs in handling mental health tasks in languages other than English and their limitations when translation quality introduces structural or lexical mismatches.

Edge AI applications increasingly require ultra-low-power, low-latency inference. Neuromorphic computing based on event-driven spiking neural networks (SNNs) offers an attractive path, but practical deployment on resource-constrained devices is limited by training difficulty, hardware-mapping overheads, and sensitivity to temporal dynamics. We present NeuEdge, a framework that combines adaptive SNN models with hardware-aware optimization for edge deployment. NeuEdge uses a temporal coding scheme that blends rate and spike-timing patterns to reduce spike activity while preserving accuracy, and a hardware-aware training procedure that co-optimizes network structure and on-chip placement to improve utilization on neuromorphic processors. An adaptive threshold mechanism adjusts neuron excitability from input statistics, reducing energy consumption without degrading performance. Across standard vision and audio benchmarks, NeuEdge achieves 91-96% accuracy with up to 2.3 ms inference latency on edge hardware and an estimated 847 GOp/s/W energy efficiency. A case study on an autonomous-drone workload shows up to 312x energy savings relative to conventional deep neural networks while maintaining real-time operation.

We propose sVIRGO, a scalable virtual tree hierarchical framework for large-scale distributed systems. sVIRGO constructs virtual hierarchical trees directly on physical nodes, allowing each node to assume multiple hierarchical roles without overlay networks. The hierarchy preserves locality and is organized into configurable layers within regions. Coordination across thousands of regions is achieved via virtual upper-layer roles dynamically mapped onto nodes up to the top layer. Each region maintains multiple active coordinators that monitor local health and perform dynamic re-selection if failures occur. Temporary drops below the minimum threshold do not compromise coordination, ensuring near-zero recovery latency, bounded communication overhead, and exponentially reduced failure probability while maintaining safety, liveness, and robustness under mobile, interference-prone, or adversarial conditions. Communication is decoupled from the hierarchy and may use multi-frequency wireless links. Two message hop strategies are supported: (i) with long-distance infrastructure-assisted channels, coordinators exploit the virtual tree to minimize hops; (ii) without such channels, messages propagate via adjacent regions. sVIRGO also supports Layer-Scoped Command Execution. Commands and coordination actions are executed within the scope of each hierarchical layer, enabling efficient local and regional decision-making while limiting unnecessary global propagation.

We consider a time-slotted job-assignment system consisting of a central server, $N$ task-specific networks of machines, and multiple users. Each network specializes in executing a distinct type of task. Users stochastically generate jobs of various types and forward them to the central server, which routes each job to the appropriate network of machines. Due to resource constraints, the server cannot serve all users' jobs simultaneously, which motivates the design of scheduling policies with possible preemption. To evaluate scheduling performance, we introduce a novel timeliness metric, the age of job, inspired by the well-known metric, the age of information. We study the problem of minimizing the long-term weighted average age of job. We first propose a max-weight policy by minimizing the one-step Lyapunov drift and then derive the Whittle index (WI) policy when the job completion times of the networks of machines follow geometric distributions. For general job completion time distributions, we introduce a Whittle index with max-weight fallback (WIMWF) policy. We also investigate the Net-gain maximization (NGM) policy. Numerically, we show that the proposed WIMWF policy achieves the best performance in the general job completion time setting. We also observe a scaling trend: two different max-weight policies can outperform the NGM policy in small systems, whereas the NGM policy improves as we scale the system size and becomes asymptotically better than max-weight policies. For geometric service times, the WI policy yields the lowest age across all considered system sizes.

Bayesian optimization (BO) is a popular, sample-efficient technique for expensive, black-box optimization. One such problem arising in manufacturing is that of maximizing the reliability, or equivalently minimizing the probability of a failure, of a design which is subject to random perturbations - a problem that can involve extremely rare failures ($P_\mathrm{fail} = 10^{-6}-10^{-8}$). In this work, we propose two BO methods based on Thompson sampling and knowledge gradient, the latter approximating the one-step Bayes-optimal policy for minimizing the logarithm of the failure probability. Both methods incorporate importance sampling to target extremely small failure probabilities. Empirical results show the proposed methods outperform existing methods in both extreme and non-extreme regimes.

Decentralized Collaborative Simultaneous Localization And Mapping (C-SLAM) techniques often struggle to identify map overlaps due to significant viewpoint variations among robots. Motivated by recent advancements in 3D foundation models, which can register images despite large viewpoint differences, we propose a robust loop closing approach that leverages these models to establish inter-robot measurements. In contrast to resource-intensive methods requiring full 3D reconstruction within a centralized map, our approach integrates foundation models into existing SLAM pipelines, yielding scalable and robust multi-robot mapping. Our contributions include: (1) integrating 3D foundation models to reliably estimate relative poses from monocular image pairs within decentralized C-SLAM; (2) introducing robust outlier mitigation techniques critical to the use of these relative poses; and (3) developing specialized pose graph optimization formulations that efficiently resolve scale ambiguities. We evaluate our method against state-of-the-art approaches, demonstrating improvements in localization and mapping accuracy, alongside significant gains in computational and memory efficiency. These results highlight the potential of our approach for deployment in large-scale multi-robot scenarios.

Tropical forests harbor most of the planet's tree biodiversity and are critical to global ecological balance. Canopy trees in particular play a disproportionate role in carbon storage and functioning of these ecosystems. Studying canopy trees at scale requires accurate delineation of individual tree crowns, typically performed using high-resolution aerial imagery. Despite advances in transformer-based models for individual tree crown segmentation, performance remains low in most forests, especially tropical ones. To this end, we introduce SelvaMask, a new tropical dataset containing over 8,800 manually delineated tree crowns across three Neotropical forest sites in Panama, Brazil, and Ecuador. SelvaMask features comprehensive annotations, including an inter-annotator agreement evaluation, capturing the dense structure of tropical forests and highlighting the difficulty of the task. Leveraging this benchmark, we propose a modular detection-segmentation pipeline that adapts vision foundation models (VFMs), using domain-specific detection-prompter. Our approach reaches state-of-the-art performance, outperforming both zero-shot generalist models and fully supervised end-to-end methods in dense tropical forests. We validate these gains on external tropical and temperate datasets, demonstrating that SelvaMask serves as both a challenging benchmark and a key enabler for generalized forest monitoring. Our code and dataset will be released publicly.

Protein fitness optimization is challenged by a vast combinatorial landscape where high-fitness variants are extremely sparse. Many current methods either underperform or require computationally expensive gradient-based sampling. We present CHASE, a framework that repurposes the evolutionary knowledge of pretrained protein language models by compressing their embeddings into a compact latent space. By training a conditional flow-matching model with classifier-free guidance, we enable the direct generation of high-fitness variants without predictor-based guidance during the ODE sampling steps. CHASE achieves state-of-the-art performance on AAV and GFP protein design benchmarks. Finally, we show that bootstrapping with synthetic data can further enhance performance in data-constrained settings.

The self-attention mechanism, at the heart of the Transformer model, is able to effectively model pairwise interactions between tokens. However, numerous recent works have shown that it is unable to perform basic tasks involving detecting triples of correlated tokens, or compositional tasks where multiple input tokens need to be referenced to generate a result. Some higher-dimensional alternatives to self-attention have been proposed to address this, including higher-order attention and Strassen attention, which can perform some of these polyadic tasks in exchange for slower, superquadratic running times. In this work, we define a vast class of generalizations of self-attention, which we call poly-attention mechanisms. Our mechanisms can incorporate arbitrary higher-order (tensor) computations as well as arbitrary relationship structures between the input tokens, and they include the aforementioned alternatives as special cases. We then systematically study their computational complexity and representational strength, including giving new algorithms and matching complexity-theoretic lower bounds on the time complexity of computing the attention matrix exactly as well as approximately, and tightly determining which polyadic tasks they can each perform. Our results give interesting trade-offs between different desiderata for these mechanisms, including a tight relationship between how expressive a mechanism is, and how large the coefficients in the model may be so that the mechanism can be approximated in almost-linear time. Notably, we give a new attention mechanism which can be computed exactly in quadratic time, and which can perform function composition for any fixed number of functions. Prior mechanisms, even for just composing two functions, could only be computed in superquadratic time, and our new lower bounds show that faster algorithms for them are not possible.

Modern machine learning has achieved remarkable success on many problems, but this success often depends on the existence of large, labeled datasets. While active learning can dramatically reduce labeling cost when annotations are expensive, early performance is frequently dominated by the initial seed set, typically chosen at random. In many applications, however, related or approximate datasets are readily available and can be leveraged to construct a better seed set. We introduce a new method for selecting the seed data set for active learning, Active-Transfer Bagging (ATBagging). ATBagging estimates the informativeness of candidate data point from a Bayesian interpretation of bagged ensemble models by comparing in-bag and out-of-bag predictive distributions from the labeled dataset, yielding an information-gain proxy. To avoid redundant selections, we impose feature-space diversity by sampling a determinantal point process (DPP) whose kernel uses Random Fourier Features and a quality-diversity factorization that incorporates the informativeness scores. This same blended method is used for selection of new data points to collect during the active learning phase. We evaluate ATBagging on four real-world datasets covering both target-transfer and feature-shift scenarios (QM9, ERA5, Forbes 2000, and Beijing PM2.5). Across seed sizes nseed = 10-100, ATBagging improves or ties early active learning and increases area under the learning-curve relative to alternative seed subset selection methodologies in almost all cases, with strongest benefits in low-data regimes. Thus, ATBagging provides a low-cost, high reward means to initiating active learning-based data collection.