Science & Mathematics

Skills provide agents with reusable experience to guide future decisions without modifying the underlying model. However, coding trajectories are often entangled with task- and repository-specific contexts, making it challenging to distill procedural skills that generalize across diverse coding environments. To address this challenge, we present Hierarchical Skill Evolution (HSE), which progressively distills procedural knowledge from coding trajectories through hierarchical evidence aggregation. HSE first extracts task-level evidence from individual trajectories, then aggregates recurring evidence within the same repository to form repository-specific skill candidates, and finally consolidates consistent patterns across repositories into transferable skills across repositories. The evolution is driven by an external writer-evaluator loop that generates and assesses skills while keeping the underlying coding model frozen. The loop externally updates the writer and evaluator policies to improve subsequent skill generation and assessment. This hierarchy progressively filters task-specific details while preserving procedures that generalize across diverse coding environments. Successful trajectories yield reusable guidance for future tasks, while failed trajectories reveal recurring failure modes that guide avoidance and recovery. We evaluate HSE on SWE-bench Verified and DeepSWE, using GLM-5.2 across Pi and Claude Code harnesses and GPT-5.5 and Opus 4.8 with Pi. Across both benchmarks, 10 skills evolved solely from SWE-Gym improve coding performance across all evaluated settings, demonstrating transfer across benchmarks, coding models, and agent harnesses. We further evolve the skill library from target-benchmark trajectories without verifier feedback, relying solely on the learned evaluator for candidate assessment and achieving additional gains through benchmark-tailored skills.

The problem of predicting when a multi-planet system will become dynamically unstable has defied analytical progress over the past three hundred years. Recent machine learning advances achieve high predictive performance of instability dynamics, but as black-box systems cannot be interpreted to connect to existing scientific understanding. Here we show analytic equations that predict when a multi-planet system will undergo instability. We discover these equations by first training a neural network on a dataset of N-body simulations, and then distilling its latent knowledge into closed-form algebraic expressions. Our two-stage pipeline first learns an interpretable variable basis, then searches for equations using symbolic regression. The resulting equations outperform existing analytic estimates of instability and apply across wide ranges of mass ratios, spacing, and orbital eccentricities. They capture structure in the prediction landscape with a small number of variables amenable to dynamical interpretation. These results demonstrate how symbolic learning can be combined with neural networks to extract scientific insight from complex dynamical systems.

Evaluating deformable 3D reconstruction requires ground truth linking camera motion, surface deformation, material correspondence and image appearance. We introduce the Synthetic Deformable Tubular Benchmark, generated with LumenMorph, an open-source toolkit for creating, annotating and evaluating tubular scenes. The release contains 5,580 RGB images. Its baseline crosses three surface-motion conditions with moving and stationary cameras in six scene families; matched appearance variants cover only moving-camera finite-element actions in four families. Documented public and evaluation interfaces expose camera poses, geometry, depth, deformation fields, persistent material correspondences and discrete finite-element records. Persistent triangle identities and material weights let users track the same surface location across frames and query its three-dimensional trajectory. Separately implemented geometric and mechanical checks assess annotation consistency with the generating models. Structure-from-motion, optical-flow and supplied-geometry displacement-interpolation studies demonstrate how the resource distinguishes registration coverage from geometric accuracy, exposes camera- and surface-motion effects, and tests the benefits and mismatch of mechanical priors. Public readers, evaluation contracts and the configurable generator support reuse and extension. Data integrity, fixed-environment repeatability and retained-prediction score replay have distinct verification scopes. The benchmark provides a controlled synthetic reference without claiming anatomical or clinical realism.

An adversary brain is an LLM-based model of a threat actor's behavior, held as an evolving logical signature: a hypothesis with enough structure that data can test it continuously, without a person re-interpreting it each time. Rather than any single observation, such as an IoC (Indicator of Compromise), it is the arrangement that becomes a detection, or intelligence, mechanism. Where YARA matches a file and Sigma matches a log line, an adversary brain matches the activity pattern of a campaign or threat actor. This is based on an intelligence analysis technique called Analysis of Competing Hypotheses (ACH), which the design lifts from the limitation of a few hypotheses at a time, to thousands, and then also operationalizes it for continuous detection purposes. Language models make the concept, and scale, practical: they turn prose reports into brains, judge matches, and drive the loop that keeps brains current. A partial pattern in a brain can also match a new value, such as a fresh C2 (command and control) domain name, assisting with the discovery of new indicators, and turning unknown unknowns into known unknowns, and potentially into knowns. The author would like to thank Ryan Moon for the collaboration, and Sounil Yu, Chris Inglis, Thomas Roccia, Marius Benthin, Israel Grimberg, Michal Kamensky, Justin Borland, Brandon Levene, John Hultquist, and John Bambenek for their feedback and assistance.

Efficient robotic data collection remains a major bottleneck for vision-language-action (VLA) policy learning, especially for tasks requiring accurate end-effector positioning and camera-action alignment. This challenge is particularly severe in electronic-waste disassembly, where robots operate in cluttered assemblies of densely packed, visually similar components. Full digital-twin simulation is difficult because many parts are thin, reflective, or mechanically complex, making accurate object-level meshes and contact models difficult to obtain. In this work, we focus on pre-contact reaching and propose a 3D Gaussian Splatting (3DGS)-based data generation pipeline for wrist-camera VLA adaptation. Our method reconstructs the real workspace with 3DGS and renders diverse wrist-camera observations from robot-aligned poses. These observations are paired with smooth, structured reaching trajectories to provide robot-aligned visual-action supervision for VLA fine-tuning. We further augment the VLA policy with an instruction-conditioned reach verifier that estimates target proximity and provides reliable termination before physical interaction. Experiments in motherboard disassembly demonstrate precise VLA-based reaching using wrist-camera observations alone, without an external side-view camera. With $π$0.5, 1,500 generated trajectories achieve 80.0\% precise-reaching success across three targets, compared with 58.7\% using 150 teleoperated demonstrations, while requiring less than 10 minutes of active human capture. Project website: https://001-wang.github.io/PrecisionReach-GS/

Networks have been used to model the interactions among the elements that underlie a narrative. In this paper, we investigate whether the structure of these networks can predict which links will emerge as a storyline evolves. We construct these networks from news articles using a Semantic Edge Parser (SEP) built on a Large Language Model (LLM), and evaluate the networks on three disparate real-world storylines, the 2022 Canadian Freedom Convoy, Hurricane Melissa (2025), and the 2025-26 Iranian protests. To test the narrative network's predictive capability, we construct a rolling window of the narrative and predict the emergent links between existing narrative elements on the day following the rolling window period. We identify candidate links with seven standard link predictors, including the Adamic-Adar index, Common Neighbors, and Preferential Attachment. Forecasting next-day links across the three storylines, we find Preferential Attachment to be the strongest standard link predictor, achieving a mean ROC-AUC of 0.76-0.79. It exceeds EdgeBank, a memorization baseline, by 11.3-14.5 points on every event. To summarize, narrative-network structure ranks next-day emergent relations above chance, though this advantage is concentrated among the most frequently mentioned, highly connected entities.

This work explores a multimodal approach to the alignment of the behavior of open-ended learning autonomous robots with human purposes through the use of large language models as cultural enablers. While autonomous robots can generate their own skills and goals through interaction with the environment, they often lack mechanisms to ensure that these remain relevant and aligned to those of human users. The proposed approach makes use of the general and cultural knowledge contained in LLMs and VLMs to establish a multimodal interactive alignment process in which human purposes, expressed in natural language, are transformed into motivational components of a robot architecture in the form of robot missions and drives that guide the behavior of the robot. The system is structured into three layers for semantic alignment, mission generation, and drive modeling. This design enables a bidirectional interaction loop: human inputs define high-level purposes, robots expand them into executable missions making use of multimodal information relayed by LLMs and VLMs, and feedback maintains alignment with human intent. Experiments in simulation and with a physical robot demonstrate that the process produces valid missions and behaviors, supporting effective human-agent alignment in interactive scenarios with autonomous robots.

In autonomous driving, robot teleoperation serves as a critical safety fallback mechanism, requiring operators to assume direct control in the event of system failure. While recent research has pivoted toward high-level supervisory control or simulation-based studies, these approaches often fail to capture the physical dynamics and sensory demands of driving a full-scale vehicle. To address this gap, we evaluated the impact of spatial data representation on operator performance using a real car-sized vehicle in a field experiment with $N = 18$ participants. We compared three visualization interfaces: (i) a standard camera feed baseline, with two bird's-eye views (BEV) from (ii) a raw LiDAR point cloud, and (iii) a sensor fusion of LiDAR and camera data streams. Contrary to prior passive observation studies suggesting that sensor-fusion BEV may hinder awareness, our active-control results demonstrate that such representation significantly outperforms the standard LiDAR BEV in perceived usability ($p \leq 0.0001$) and cognitive task load ($p = 0.01$, NASA-TLX score: 34.58/100) while also being significantly better than the Camera-only view ($p = 0.043$). The fused interface also led to statistically significant reductions in task completion time ($p = 0.0461$) and superior usability (SUS: 77.9/100) compared to those of the Camera-only and LiDAR-only views. By synthesizing objective performance data with comprehensive subjective assessments, we demonstrate that BEV fusion maximizes safety margins while minimizing the operator's mental demand. These findings challenge existing teleoperation conventions and provide an empirically validated approach for designing robust remote piloting systems.

Quadratic programs (QPs) are a core building block in many applications. In practice, constraints are often softened with slack variables to guarantee feasibility. The standard approach tackles this by introducing additional decision variables, growing the resulting optimization problem. In this paper, we propose an Alternating Direction Method of Multipliers (ADMM) scheme for a broad class of convex slack penalty functions, including all $\ell_k$-norms, without increasing the problem size. The only change to the ADMM scheme is a modified slack projection function, while the remaining algorithm stays the same. We demonstrate our approach on numerical examples, showing significant speedups, and show its flexibility on an energy optimization problem.

Generative AI is putting the law of reverse engineering (RE) under severe strain. Existing IP and technology laws assume that RE is expensive and time-consuming, and that it requires specialized expertise. They generally allow it, but they rely on its difficulty as a filter to separate (permissible) RE from (impermissible) copying and theft. Generative AI is changing this equation by making \RE much faster, cheaper, and easier. AI systems can identify hardware components, extract firmware, decompile object code, and clone entire apps. The resulting struggle between technology vendors and reverse engineers will challenge a wide range of laws, including trade secret, copyright, anti-circumvention, computer misuse, and contract. In this article, we describe the ways in which AI is changing \RE, map these new practices onto existing laws, and sketch ways that the legal system could respond and their associated trade-offs.

An offline language-based support tool is presented to assist pilot training and tactical decision-making in air combat operations. The system, named MATE (Mission Assistant for Training and Evaluation), applies retrieval-augmented generation (RAG) to answer natural language questions using content extracted from doctrinal and tactical manuals. Built entirely with open-source components, it performs text preprocessing, embedding generation, semantic search, and language generation locally, ensuring full offline functionality. The tool was evaluated by three Brazilian Air Force fighter pilots, who rated the answers produced with three open-weight language models (Llama 3.1 8B, Qwen 2.5 14B, and Mistral NeMo 12B) to five tactical questions derived from a publicly available air combat manual. Average ratings ranged from 4.00 to 4.40 on a five-point scale, with Mistral NeMo 12B obtaining the highest score, and the pilots rated the response time of the tool as fully adequate compared to manual search. The relevance of the retrieved content was identified as the main point for improvement. Potential applications are discussed in onboard mission systems, flight simulators, and training environments.

Latent world models are typically pre-trained offline and kept frozen at test time, so planning with them can degrade under distribution shift. Test-time adaptation (TTA) can mitigate this degradation by updating the model during deployment. Unlike pre-training on fixed data, TTA for planning is closed-loop: the planner uses model rollouts to select actions, and the resulting transitions drive the next update. This raises the question: which objective should guide TTA for planning? Under a prediction-only objective, the model is fit only to the outcomes of its own actions, so it can remain wrong about the actions that would reach the goal, and low prediction error need not translate into better planning. Inspired by reward-biased maximum likelihood from adaptive control, we introduce Optimistic Test-Time Adaptation (OpTTA). Its objective has two terms, the prediction loss on observed transitions and the planner's goal-reaching cost on its candidate rollouts, both minimized over the model parameters. The planning term biases adaptation toward models that are optimistic about reaching the goal, without any additional environment interaction. Across visual and environment shifts on manipulation, reaching and navigation tasks, OpTTA raises final planning success over the frozen model by 13.5 percentage points on average, compared with 6.9 for the prediction-only objective (AdaJEPA), when only the predictor is adapted. When the encoder is adapted as well, the gains are 13.5 and 7.5 points. These experiments span three JEPA world models with different encoders and training objectives and, for temporal straightening, both CEM and gradient-based planners. Our results suggest that test-time adaptation of world models for planning benefits from an optimistic, task-aware term in addition to the prediction loss.

Due to the inherent discrepancies in imaging viewpoints and field-of-view mismatches among multi-modal sensors, existing fusion methods have long struggled to achieve complete scene-level information integration beyond the limitation in sensor fields of view. Consequently, the generated fusion results often contain modality-specific regions and fusion artifacts, limiting the comprehensive representation of scene structures and semantic information. To address this challenge, we propose a generative registration-fusion framework, termed SeeRF. Specifically, it first introduces a deformation generation strategy based on the diffeomorphic stationary velocity field, combined with visual foundation model-guided feature extraction and a coarse-to-fine progressive multi-modal image registration module, to achieve accurate cross-modal flow estimation and generate valid field-of-view masks. Subsequently, the registration module is tightly coupled with the generative diffusion module for multi-modal completion and fusion, where the cross-modal control encoding module and the control residual feature encoder-decoder mechanism provide effective guidance. By leveraging the powerful prior modeling capability of generative diffusion models, the proposed framework overcomes the limitations imposed by shared fields of view and enables effective completion and natural fusion of non-overlapping field-of-view regions. Extensive experiments on multiple public datasets and real-world scenarios demonstrate that the proposed method achieves superior performance, validating its effectiveness, superiority, and robustness. The code will be available at https://github.com/XunpengYi/SeeRF.

With the increasing integration of inverter-based resources (IBRs) into power systems, assessing small-signal interaction stability has become essential. Impedance/admittance models enable frequency-domain characterization of IBR dynamics without requiring access to proprietary internal control details, but measurement-based identification is time-consuming and must be repeated across operating points. This paper proposes FedAvgIm, a privacy-aware federated-learning framework for multi-site dq-admittance identification under severe local data scarcity and raw-data-locality constraints. Each IBR site trains a local feed-forward neural network using a small set of operating-point-frequency samples, while only model parameters are shared with a cloud aggregator that performs sample-size-weighted FedAvg to update a global model and broadcast it back to all sites. To demonstrate effectiveness, EMT measurement-based admittance datasets are constructed for nine grid-following IBRs with different parameters, architectures, measurement-noise levels, and non-IID settings, where each client has only 40-390 training samples. Results show that FedAvgIm achieves performance close to centralized training and consistently outperforms local-only learning and raw-data-locality-preserving transfer-learning baselines. Additional studies show that FedAvgIm maintains stable learning behavior under measurement noise and remains beneficial under moderate non-IID conditions. The estimated dq-admittance models are further applied to generalized Nyquist stability analysis and validated against EMT time-domain simulations under multiple operating points and grid-strength conditions.

Passive acoustic monitoring in tropical waters such as those in Southeast Asia is hampered by the persistent, high-level impulsive noise of snapping shrimp. This is a particularly daunting challenge for detecting broadband transient vocalizations of dolphins - clicks, buzzes and burst-pulses. These vocalizations, unlike frequency-modulated whistles, share the broadband, impulsive character of the shrimp snaps and cannot be easily separated from them by frequency-localized or energy-based means. We develop TRIDENT, a lightweight waveform-based detector for these vocalizations, and apply it to 54.2 recorder-months of data from ten island sites around Singapore. We show that the training-data pipeline and the architecture are both important: controlled signal-to-noise-ratio based mixing to generate training data, appropriate normalization and hard negative mining, together steer the detector toward learning transient temporal structure rather than energy shortcuts. A single joint detector handles all three signal types better than specialized models for each, and outperforms established baselines. Applied to long-term recordings, it reveals tightly clustered, predominantly daytime activity, motivating days-with-detections as a robust indicator of cetacean presence.