Papers

A region can overlap a target object yet contribute more to another class. We test regions selected by Cluster-based Concept Importance (CCI) in frozen CLIP. Across COCO and VOC with two checkpoints, 41.08-64.78% of regions that pass overlap and mean-contrast checks fail against the strongest competing class. Removing competitors annotated in the image leaves 39.69-63.64% failing. We then test all eight candidate regions per image. An alternative passes the test for 6.25-7.84% of failures on COCO and 27.40-31.15% on VOC. Requiring it to preserve the original target-score drop within $\epsilon = 0.02$ reduces these rates to 0.16-0.98%. Available regions and target-drop tolerance constrain repair; relaxing the tolerance increases repair opportunities.

Unlearning ensures LLM compliance by removing the influence of private or copyrighted training data. However, since LLM models typically undergo post-training compression, like quantization, in practical deployment, it has been observed that the unlearning effect can be substantially weakened, with the forgetting behavior degrading more severely than that of model utility. This paper proposes a quantization-robust unlearning framework that makes forgetting robust to quantization while maintaining overall model utility. We analyze this gap through the lens of loss landscape. Specifically, our analysis reveals a curvature-based criteria that pinpoints sensitive weights in the unlearned model that leads to both non-robust forgetting and reduced utility. We therefore propose sensitivity-guided noisy regularization, which is applied on the sensitive parameters to steer the model convergence towards a smoother minima of uniformly low forget and retain losses. Balancing unlearning and utility, we further propose forget-critical optimization, which updates only forget-critical layers, preserving most of the network to retain useful knowledge. Extensive experiments on the MUSE and TOFU benchmarks across multiple LLM unlearning algorithms show that our approach achieves substantially more quantization-resilient forgetting while maintaining utility. Code is available at https://github.com/jialuw0830/QRUnlearn.

Generative AI systems are increasingly deployed to address domain problems. These systems operate under technical, regulatory, institutional, and normative constraints that define acceptable AI behaviour and outcomes within their domains. We observe a recurring pattern in our industry engagement: partners often arrive with a functioning but relatively generic AI solution. The challenge is no longer to build an AI system from scratch, but to improve the quality and domain appropriateness of an AI-generated solution. In these settings, the limiting factor is often the quality, scope, and structure of the context available to the system. Yet, existing context engineering approaches primarily focus on supplying domain knowledge through retrieval, memory, and tools, with limited support for systematically identifying and operationalising the constraints that govern AI systems in their operational environments. This paper proposes Constraint-Driven Context Engineering (CDCE), a design approach for engineering domain interfaces for AI systems. Drawing on software architecture design and Domain-Driven Design (DDD), CDCE treats domain constraints as first-class design drivers. It identifies and characterises constraints, determines the required context assets, and designs representations through which these assets are made available to AI systems. We conducted a comparative multiple-case study with industry and public-sector partners across educational assessment, healthcare decision support, and financial-distress prediction. Depending on their characteristics, constraints can guide AI behaviour, enforce permissible boundaries, or support verification of AI-generated outcomes. The cases demonstrate CDCE's applicability across contrasting domains and show how constraint characteristics shape the resulting domain interfaces.

Web agents are usually evaluated in live environments, where environment state and judge models drift between runs, so the same checkpoint rarely reproduces the same score, making controlled studies of training phenomena impractical. We present WebMRE, an offline benchmark of 541 tasks and 5,293 steps derived from successful WebArena trajectories, with fully audited test labels and a deterministic protocol that scores a checkpoint identically on every run without any environment. Each step pairs a human oriented guide sentence with a grounded action, enabling the first study of the mutual reinforcement effect between them in web agents. Averaged over three seeds the effect holds for both models in both decoding orders and grows with scale: jointly decoding a guide lifts element selection over an action only reference by 0.9 and 0.2 points for Qwen3.5-4B and by 1.7 and 2.2 points for Qwen3.5-9B. A mediation analysis shows that the guide is a causal channel rather than commentary: forcing the gold guide as a decoding prefix lifts action accuracy from .422 to .684, another step's guide collapses it to .055, and a paraphrase that renames the target still recovers half of the gain, so the channel carries instruction meaning and not only the label string. The same channel yields an offline reward that only a replayable protocol makes computable, though optimizing it from a strong checkpoint brings no gain yet. Our fine tuned models outperform GPT-5.5, Claude Opus 4.8, and Gemini 3.5 Flash, run zero shot, on every offline metric.

We investigated the upper critical field of a large single grain of the Nb-substituted van der Waals layered quasicrystal (Ta$_{0.7}$Nb$_{0.3}$)$_{1.6}$Te. The sample exhibits a sharp superconducting transition at $T_{\mathrm{c}}$ = 1.35 K, the highest value reported to date among quasicrystal superconductors. The angular dependence of the critical field exhibits a pronounced criterion dependence: the field determined using the 10% $R_{\mathrm{N}}$ ($R_{\mathrm{N}}$: normal-state resistance) criterion is well described by the anisotropic Ginzburg-Landau model, whereas those determined using the 65% and 90% $R_{\mathrm{N}}$ criteria exhibit Tinkham-like angular dependence characteristic of two-dimensional superconductivity. The high-field part of the resistive transition is well described by a surface-superconductivity model and exhibits a pronounced excitation-current dependence for magnetic fields close to the $ab$ plane, supporting the presence of surface superconductivity on the quasiperiodic $ab$-plane surfaces. The bulk $H_{\mathrm{c2}}$ is strongly anisotropic, with the in-plane $H_{\mathrm{c2}}$ exceeding the weak-coupling Pauli limit by a factor of approximately 2.5. For both field orientations, $H_{\mathrm{c2}}(T)$ deviates upward from the conventional dirty-limit Werthamer-Helfand-Hohenberg prediction at low temperatures. A phenomenologically modified Ginzburg-Landau-Abrikosov-Gorkov model incorporating a spatial distribution of the electronic diffusivity substantially improves the description of $H_{\mathrm{c2}}(T)$, suggesting that spatial variations in electronic transport properties may contribute to its anomalous temperature dependence.

2PBC J0800.5$-$4306 (Swift J0800.7$-$4309; optical counterpart Gaia DR3 5533020382479508736) is a hard-X-ray-selected cataclysmic variable whose magnetic subtype had remained uncertain. We analyse TESS photometry from Sectors 88 and 89 and archival XMM-Newton timing data. The TESS data show signals at the orbital frequency $\Omega$, its first harmonic $2\Omega$, the white-dwarf spin frequency $\omega$, and the spin–orbit beat frequency $\omega-\Omega$. We measure $P_{\rm orb}=5.38540^{+0.00027}_{-0.00028}$ h and $P_{\rm spin}=0.7015570^{+0.0000059}_{-0.0000062}$ h. The XMM-Newton EPIC-pn, MOS1, and MOS2 periodograms show broad peaks near the 2525.6-s (0.701557-h) optical spin period. Together with the previously published cataclysmic variable classification and Swift/BAT hard-X-ray detection, these timing results identify 2PBC J0800.5$-$4306 as an intermediate polar. We report the first published measurements of its orbital and white-dwarf spin periods, while SPOC difference-image diagnostics and pixel-level timing tests support that the measured TESS modulations are associated with the Gaia counterpart despite the crowded field.

Andrews and Dastidar introduced the statistic $\SOME(n)$, defined as the difference between the sum of all odd parts and the sum of all even parts occurring in the partitions of $n$, and investigated its arithmetic properties. Motivated by their work, Baruah and Gogoi defined two analogous statistics for overpartitions, namely $\OSOMEo(n)$ and $\OSOMEe(n)$, which record the sums of all odd and all even overlined parts, respectively. They established several congruences for these statistics and their difference, and proposed a conjecture on congruences modulo $7$. In this paper, we prove their conjecture by elementary $q$-series methods. Our proof combines two specializations of Watson's quintuple product identity, a $(p,k)$-parameterization of divisor-sum series, and termwise logarithmic differentiation

Real-world molecular design remains challenging for large language model (LLM)-based agents. It requires them to interpret design contexts, satisfy multiple constraints, identify infeasible specifications, and reason over multi-step tool outputs. Existing benchmarks do not capture this complexity, focusing instead on explicit and narrow constraints, only feasible problems, and single-path solutions. To address this gap, we propose MolDesignBench, a scenario-grounded benchmark that more closely reflects real-world molecular design for evaluating tool-augmented LLM agents. MolDesignBench comprises 2K generation and optimization instances that combine implicit requirements embedded in design narratives with explicit property and functional-group constraints, including infeasible cases, and require the effective use of 17 specialized chemistry tools. Experiments across diverse frontier LLMs reveal low success rates–with the best achieving only $\sim43$%–and frequent failures in implicit-constraint reasoning, infeasibility detection, and tool reasoning. The corresponding fine-grained failure-mode analysis identifies implicit constraint interpretation and infeasibility detection as the primary bottlenecks, establishing MolDesignBench as a rigorous testbed to guide future research on chemical agents. The benchmark, tool interface, and evaluation code are publicly available.

We study trust-region minimization of a smooth, possibly nonconvex functional plus a convex functional composed with a bounded linear operator. A relative primal–dual gap condition controls both the error in an approximate proximal-gradient step and its linear-model decrease. Together with a computable absolute stationarity test, it yields a finite Cauchy search, convergence of the proximal stationarity measure to zero, and an $O(\varepsilon^{-2})$ bound on outer trials. The outer analysis allows the linear operator to take values in a Banach space and does not require dual attainment. When the operator takes values in a Hilbert space and the regularizer is finite and Lipschitz, the dual proximal-gradient method produces finite gaps tending to zero, provided the required proximal maps and functional values can be evaluated. We prove $O(j^{-1})$ gap bounds for both recovered and averaged primal candidates and give a sharper bound on the primal error for exactly recovered points. A semilinear elliptic control problem with unsmoothed total-variation regularization and an $L^2$ control cost illustrates the method in the full $H^1$ metric. Across five meshes, outer and state Newton counts remain constant, while interior-point iteration counts vary mildly.

Magnetic moments carried by rotating ionic motion in crystals are becoming recognized as an important contribution to magnetism, angular momentum transport, and optical activity. However, efficient approaches to their dynamical control are lacking. Here, we report THz-driven generation and optical detection of quantum ionic magnetism in quantum paraelectric SrTiO3. We observe an oscillatory ionic magnetization without a corresponding oscillatory polarization, contradicting from the classical relation M~PXdP/dt while its suppression above the quantum paraelectric regime points to a quantum ionic origin. Analysis shows that this effect results from the beating between quantum ionic eigenstates whose degeneracy is lifted due to the directional symmetry breaking by the THz pulse. The presented approach provides a pathway for the ultrafast control of ionic magnetization in quantum materials for spintronic and thermotronic applications.

Reliable localization of non-line-of-sight (NLOS) pedestrians is critical for safe urban autonomous driving, yet it remains highly challenging in ego-dynamic outdoor environments, where ego-vehicle motion makes radar multipath propagation complex and noisy. In this paper, we present a reflection-aware framework for NLOS pedestrian localization with a moving ego-vehicle in outdoor testbed scenarios. Our framework fuses front-view camera images and 2D radar point clouds to infer reflection orders and reflective surface distributions in bird's-eye-view space. It then uses physics-guided ray tracing to reconstruct distorted reflection paths and localize the hidden pedestrian. We validate the framework in outdoor testbed scenarios under ego-dynamic conditions. The results demonstrate the effectiveness of the proposed framework for NLOS pedestrian localization with a moving ego-vehicle.

In this paper, we study sampled-data stabilization of general nonlinear stochastic systems under a local exponential-type Lyapunov condition for the continuous-time closed-loop system. The coefficients are assumed only locally Lipschitz, with neither linear-growth nor Khasminskii-type conditions. Two counterexamples show that almost-sure stabilization is generally unattainable. Instead, given an initial state, we establish exponential stability with arbitrarily high probability under sufficiently fast sampling. Separate methods are developed for \(p\ge 2\) and \(0<p<2\), where \(p\) is the growth order in the Lyapunov condition. Numerical examples illustrate the results.

Absorption spectroscopy identifies molecules by detecting frequency-dependent optical loss, and its fundamental limit is a quantum channel-discrimination problem. In this work, we develop a general framework for characterizing the ultimate quantum limit of molecular absorption spectroscopy without collective measurements: we formulate it as an asymmetric quantum hypothesis testing task between different absorption channels. Using a semidefinite-programming formulation, we construct optimal probe states and compare them with coherent-state photon counting, Fock states, and two-mode squeezed vacuum probes. We introduce a random-loss channel model for imperfectly known background transmission, reflecting the fact that a blank experiment need not return all incident photons due to scattering, solvent absorption, or collection losses. Our results show that optimized non-classical probes can substantially outperform conventional laser spectroscopy, and that the advantage is robust to realistic baseline random loss. % In multi-frequency spectroscopy, the optimization also determines how photons should be allocated across absorption lines, revealing an additional resource-allocation advantage beyond single-mode state engineering. We illustrate our framework using infrared detection of 4-nitrophenol against chemically relevant comparison molecules, showing how quantum channel-discrimination methods provide constructive performance limits for molecular spectroscopy under realistic photon constraints.

We propose a sensor pairing method for online target tracking based on time-difference-of-arrival (TDOA) measurements through a sensor network. Existing pairing designs fix active sensor pairs in advance. In online tracking, however, adapting the pairing to the current target estimate maintains a favorable sensor-target geometry around it. We select $K$ pairs at each time step to maximize a Fisher information matrix (FIM) criterion under a communication-degree constraint. The original combinatorial problem is cast as a mixed-integer second-order cone program. We solve it at every time step to obtain a globally optimal pairing. Experiments under various noise settings show that the proposed method obtains the lowest tracking error among the methods under the same communication-degree constraint.

Cutting changes both the shape and topology of deformable objects, making accurate simulation challenging for robotic manipulation. A simulator must track the cutting tool as a cut develops, preserve the resulting discontinuities after tool withdrawal, and enable newly exposed surfaces to interact with the tool and with each other. Existing formulations often prescribe cut surfaces in advance or couple material separation to auxiliary geometric fields. We introduce BladeMaster, a GPU-accelerated cutting framework based on the total Lagrangian material point method (TLMPM). Our key idea is to encode the cutting history directly on material points through persistent side labels generated online from the blade geometry. These labels govern particle-grid coupling, preserving connectivity within intact material while preventing spurious coupling across cut faces after tool withdrawal. Our formulation supports progressive and intersecting cuts without predefined cut surfaces or particle duplication. Material-material contact enables cut surfaces to recontact and slide against each other without reconnecting, while two-way tool-material coupling allows material reaction forces to influence tool motion. Experiments demonstrate tool-driven cutting followed by manipulation, with faster-than-real-time performance on representative tasks. Project page: https://jango6324.github.io/blademaster/.

We study the uniqueness and boundary behavior of nonzero convex Aleksandrov solutions to $\det D^2 u=M|u|^p\nu$ with zero boundary values on bounded convex domains in $\mathbb{R}^n (n \geq 2)$. For $0<p<n$, we prove the uniqueness of nonzero convex solutions in the finite-energy class when $\nu$ is a locally finite Borel measure with positive mass and $\int_\Omega \text{dist }(\cdot,\partial\Omega)\,d\nu<\infty$. For $p>n$, we construct an explicit two-shell measure on the unit ball for which the problem has at least three radial solutions that are globally Lipschitz and have finite energy. In the case of $\nu=\text{dist }(\cdot,\partial\Omega)^{-\alpha}\,dL^n$, $0\leq\alpha<2$, we prove global Lipschitz continuity when $p-\alpha>n-2$ and obtain sharp upper and lower estimates on domains with a flat boundary part when $n(\alpha-1)-2<p-\alpha\leq n-2$. When $\alpha=0$ and $p=n-2$, our log-Lipschitz lower estimate has the same exponent as the known upper estimate. This answers the question raised by Le (Global Lipschitz and Sobolev estimates for the Monge-Ampère eigenfunctions of general bounded convex domains. Ann. Fac. Sci. Toulouse Math. (6) 35 (2026)). We also give a sufficient condition for finite Monge-Ampère energy on every bounded convex domain, prove its necessity when the boundary contains a flat part, and apply it to prove the uniqueness of the Monge-Ampère eigenvalue among all nonzero convex solutions.

Large language models (LLMs) are increasingly used as natural-language interfaces for robotic systems, yet their integration with Robot Operating System (ROS)-based navigation remains limited by two gaps. First, navigation data such as occupancy grids are represented as raw geometric messages that are difficult for LLMs to use directly as spatial or semantic context. Second, adding LLM-driven capabilities often requires custom wrappers or robot-specific interfaces, limiting reuse across systems. To address these challenges, we propose a non-invasive framework that connects LLM reasoning with ROS-based navigation through a navigation-oriented representation layer, exposed through the Model Context Protocol (MCP) as standardized, reusable tools so that any MCP-compatible LLM can access them without robot-specific wrappers. The visual map modules transform occupancy grids into metric, pose-aware images for goal reasoning, while the semantic annotation modules record waypoint-level observations with robot poses. We evaluate the framework on three tasks: autonomous mapping, spatial reasoning-based navigation, and semantic reasoning-based navigation. The results show that the evaluated LLM backends use these representations to achieve over 97% map coverage and select spatial or semantic navigation targets from natural-language instructions in a simulated indoor environment. This demonstrates representation-mediated LLM navigation without modifying the existing ROS navigation stack.

A novel efficient method for controlling photonic hook (PH) with an anisotropic dielectric microparticle dimer in the snowman-particle configuration is introduced and numerically justified. In contrast to existing schemes that rely on structured illumination, partial shielding, or complex morphologies, the proposed anisometric geometry utilizes a simple and easily manufacturable form composed of two partially overlapping cylinders. The finite element method was employed to study how the degree of asymmetry, spatial rotation, and optical properties of the upper particle (the head of the snowman-particle) affect the bending angle and maximum intensity of PH. For the first time, it has been demonstrated that adjusting these parameters allows for the deliberate switching between single and double PH modes, as well as controlling the direction and curvature of the hook, including the fragmentation of the focal area. This approach offers high efficiency in utilizing incident radiation, simplicity in fabrication, and broad tunability of PH characteristics. The findings pave the way for the development of compact near-field optical devices, such as optical traps, high-resolution microscopes, sensors, and tools for surface microstructuring.

Learned local navigation in crowded indoor environments is sensitive to how dynamic obstacle motion is represented, while collision-prone behaviour may persist after nominal policy training. We present a risk-aware reinforcement-learning framework that addresses these two issues through an uncertainty-aware Dynamic Uncertainty Grid Map (DUGM) and a modular post-training recovery mechanism. DUGM combines local occupancy, estimated obstacle motion, and motion-estimation uncertainty in a robot-centric representation. After the nominal policy is frozen, a finite-horizon Risk Value Function (RVF) is trained from nominal rollouts and used to trigger a dedicated recovery policy when continued nominal execution is predicted to be collision-prone. Experiments in a held-out NVIDIA Isaac Sim clinical-logistics benchmark show that uncertainty-aware dynamic representation improves nominal navigation over static and deterministic alternatives, while the recovery mechanism further mitigates residual collision-prone behaviour. The complete framework is also deployed directly on a TurtleBot3 without policy fine-tuning, retraining, or site-specific adaptation, retaining the performance trend observed in simulation. These results indicate that uncertainty-aware dynamic representation and post-training recovery provide complementary mechanisms for improving learned local navigation.

Open RAN (O-RAN) slicing xApps must adapt resource allocations to changing channel conditions and traffic demands while meeting service-level agreements (SLAs). Deep reinforcement learning can produce adaptive policies, but their allocation rules remain encoded in neural-network parameters. Our goal is to retain this adaptability while making the controller's decision logic directly inspectable and editable by operators. We use a large language model (LLM) to evolve slicing controllers as compact Python programs whose decision logic remains readable and editable after optimization. The LLM proposes and revises candidates offline, while a calibrated simulator scores them, and the selected decision module runs unchanged in the O-RAN control path. On the NSF POWDER 5G testbed, the evolved controller releases resources from a guaranteed slice whose throughput target becomes unattainable under a sustained channel fade, improving best-effort throughput from 158.2 to 228.6 Mbps, a 44.5% gain over the best static allocation. Since the controllers are readable source code, their behavior can be predicted from their equations, defects can be diagnosed by reading the code, and calibration errors can be corrected with one-line edits, reducing SLA misses from 79.9% to 2.2% in one case and more than doubling fitness in another. In a four-slice trace-driven simulation calibrated to the same testbed, evolutionary search achieves higher average evaluation scores than independent prompting at a matched proposal budget, with mean normalized gains on held-out traces of 16.3% for prompting alone, 32.1% for evolution from scratch, and 51.0% for evolution from a starting program.

Automated red teaming often replays a fixed set of prompts, which measures known risks but cannot learn from failures found during testing. We present CART (Closed-Loop Adaptive Red Teaming), a framework that uses each result to guide what it tests next. CART begins with broad risk coverage, follows weaknesses that emerge, keeps new probes diverse, and records the evidence and source of every finding. It separates the Challenger that creates tests, the Target being tested, which may be a text-only model or a bounded tool-using agent, and the Judge that evaluates the results, allowing these roles to be studied independently. Across three evaluation families (Frontier, JAH, and Agentic), CART discovers more failures and higher average risk than static seed replay for every Target with an available baseline. The gains extend to tool-mediated agent tests, suggesting that contextual adaptation can reveal weaknesses that direct prompt replay does not exercise. These results describe what the test policies discover, not how often failures occur in real deployments. We also find that Challenger-Judge choices affect the evidence uncovered, highlighting the need for role separation and independent review. Overall, CART turns red teaming from a one-time checklist into a continuous, adaptive, and auditable search for model and agent weaknesses.

Let $P(t)$ be the transition function of an absorbing continuous-time Markov chain on a countable state space. We study asymptotic relations of the form $p_{ij}(t)\sim a_{ij} L(t)$, $t\to\infty$, where $L$ is independent of $i$ and $j$. We obtain general conditions under which the coefficient matrix $A=(a_{ij})$ has rank one and describe the resulting consequences for survival probabilities and conditional distributions. The analysis applies to both reducible and irreducible chains. We also construct an irreducible counterexample, based on a killed random walk on a homogeneous tree, for which a common asymptotic scale exists but the coefficient matrix has rank greater than one. This shows that irreducibility alone does not imply rank-one asymptotics. The results identify conditions under which fixed-state transition asymptotics determine the asymptotic behaviour of the entire transition function and clarify the limitations of such conclusions in the absence of additional structure.

Agentic memory systems reuse past experience to improve future performance, yet most existing designs curate memory at write time: once a task is completed, its trajectory is distilled into a fixed artifact, such as a reflection, workflow, skill, or reasoning strategy, that is later retrieved by similarity. This forces the system to decide what is worth remembering before the future query is known, irreversibly discarding information and producing a query-independent summary that must serve many possible downstream tasks. Learning such a write-time curator is also difficult because the value of a storage decision may only become apparent when a relevant query arrives, potentially many tasks later, creating a long-horizon credit-assignment problem. We instead retain raw trajectories and defer curation until read time, when the current task is known. Given the retrieved traces and the new task, a memory curator synthesizes a compact, task-adaptive payload tailored to the immediate need. Because this payload is consumed on the same task, the curator can be trained directly from immediate task success, avoiding delayed utility signals and the need to artificially group related tasks. Across ALFWorld, WebShop, and $\tau^2$-bench, our Just-in-Time Memory (JitMem) consistently outperforms no-memory agents as well as heuristic and learned write-time memory methods, improving over the strongest baseline by 16.2, 16.3, and 3.9 absolute success-rate points, respectively. Notably, even an untrained curator is already competitive with or surpasses these baselines, showing that task-adaptive read-time curation itself is a major source of the gain; training the curator further compounds the improvement.

Platform operators increasingly rely on system prompts and fine-tuning to govern model behavior, yet it remains unclear how reliably these interventions override behavior inherited from prior training. We propose Override Success Rate (OSR) and alignment inertia to measure when operator interventions succeed or fail to change prior behavior. We evaluate zero-shot prompting and LoRA fine-tuning across Llama and Mistral in medical misinformation and hate speech. Alignment inertia persists across both models but varies by model, domain, and policy direction. Notably, in Mistral's restrictive hate-speech condition, LoRA increased inertia by 46.5 percentage points, showing that fine-tuning can reinforce rather than override prior behavior. We also use TRAK to test whether inertia is associated with weaker adaptation signals. TRAK achieves AUC of at least 0.85 in 7 of 8 conditions and outperforms model confidence, TF-IDF similarity, and embedding similarity as a predictor of inertia. These results provide an operator-facing audit of where prior training constrains downstream model governance.

Long horizon agents accumulate growing interaction histories that increase context and inference costs. We find that geometric redundancy alone is an insufficient criterion for safe compression. Although agent histories exhibit strong low dimensional structure, similar global geometry can preserve very different amounts of task evidence. At identical retained block counts, evidence aware selection raises next action Top 3 retention from 0.31 to 0.69, while centroid similarity remains 0.98. Controlled replacement further shows that action related information can be substantially altered while global geometric measures remain nearly unchanged. Motivated by this gap between geometry and evidence, we introduce Geometry Guided Evidence Preserving Memory (GEM), a training free compressor that protects task and execution evidence before using geometric residuals to complete coverage. GEM reduces mean combined token usage from 2.69M to 2.11M per task, a 21.4% reduction, while maintaining comparable task reward. Our results show that efficient agent history compression should optimize for preserved task evidence rather than geometric coverage alone.

We present JEV-Star, a StarCraft II controller that defeats the strongest non-cheating built-in AI, Lv7, by combining fast JEV action selection with persistent GPT-6 planning. The combined system wins four full games at Lv5–Lv7, including two Lv7 victories with different seeds, while retaining a median JEV response time of 0.422 seconds. Mean estimated model cost across these games is USD~3.71 per game: USD~0.15 for JEV and USD~3.56 for GPT-6. We compare this system with an initial JEV-only controller in full-game macro control and multi-unit micromanagement. The standalone controller reaches a 20-minute limit against Lv2 without expanding. Across 35 battle maps with three episodes per map and controller, the combined system raises mean enemy elimination from 16.50% to 37.69% and wins from 3 to 7 out of 105. Replay frames and decision logs document resource reservation, persistent economic goals, and stable army objectives in successful full games. The results demonstrate a practical division between inexpensive, subsecond decisions and longer-horizon planning. The comparison evaluates complete systems; concurrent interface improvements mean that planning's contribution is not isolated by a controlled ablation. Code is available at https://github.com/sc2musa/Jev_Star

We develop a sample-based framework for constructing information-driven hierarchical multi-resolution representations of probabilistic occupancy grids. Recent methods compute information-optimal abstractions via dynamic-programming-based exhaustive recursions, which become computationally prohibitive for large-scale grids and are ill-suited to robotics applications. To address this limitation, we introduce a sample-based strategy inspired by Monte Carlo Tree Search (MCTS) that incrementally constructs hierarchical abstractions through statistical estimation rather than exhaustive enumeration. The proposed method is anytime in nature, allowing computation to be terminated at any stage to produce a valid compressed representation. We compare our approach with the information-optimal Q-tree search algorithm and demonstrate its effectiveness in rapidly generating abstractions of large real-world probabilistic occupancy grids.

In this paper, we propose and analyze an augmented Lagrangian framework for solving stochastic nonconvex optimization problems with expectation-based equality constraints over a closed and convex constraint set. The framework generates a sequence of nonconvex primal subproblems, which are solved inexactly using stochastic second-order methods. We establish iteration complexity results for obtaining approximate second-order stationary points, both in expectation and with prescribed probability, under corresponding conditions on the accuracy of the primal subproblem solutions. We further establish sample complexity results for the framework with general stochastic second-order subproblem solvers under reasonable assumptions on their theoretical guarantees. Moreover, we incorporate three existing stochastic second-order solvers and derive the corresponding deterministic and probabilistic sample complexity results for obtaining approximate second-order stationary points. To the best of our knowledge, such sample complexity guarantees for obtaining approximate second-order stationary points in nonconvex expectation-constrained optimization have not been established previously. Finally, we demonstrate the empirical performance of the proposed framework on two nonconvex machine learning problems.

Multi-Fidelity optimization (MFO) has received extensive attentions in engineering design, which resorts to augmenting the small number of expensive high-fidelity (HF) samples by a large number of low-fidelity (LF) but cheap samples to improve the optimization performance. A key factor that influences the effectiveness of MFO is how to adaptively assign samples for HF and LF simulations in the iteration process. To address such sample assignment issue in MFO, we propose a new infill criterion named as Filter-GEI, which imposes an adaptive filter function on top of the generalized expected improvement (GEI) acquisition function. In particular, by taking the correlations between HF and LF models into account, the Filter-GEI can efficiently allocate HF and LF samples to achieve a good balance in between the local and global search. Furthermore, considering parallel computing, the Filter-GEI infills multiple HF and LF samples in each iteration, which can further improve its efficiency as computing power increases. Through tests on five mathematical toy problems and one engineering problem for the turbine blade design, the effectiveness of the proposed algorithm has been well demonstrated.

Video provides a rich record of human behavior, interaction, and situated contexts, offering important evidence for understanding people and conducting human-centered research. As vision-language models (VLMs) become increasingly capable of analyzing video, they offer opportunities to automate this traditionally human-intensive process. Yet a central question remains: when can VLMs analyze human-centered video independently, and when does reliable analysis still require human involvement? To address this question, we first characterize video analysis practices in human-centered research. We systematically analyze all 1,702 CHI 2026 full papers and identify 125 that annotate videos. Through iterative coding, we derive a five-dimensional taxonomy spanning analytic purpose, viewpoint, phenomenon, reasoning requirement, and annotation authority. Grounded in recurring annotation tasks captured by this taxonomy, we construct a benchmark of 15 representative tasks from open datasets to map the capabilities and limitations of a general-purpose VLM. We examine the division of labor between humans and VLMs by comparing three annotation workflows: VLM alone, human alone, and human verification of VLM outputs. Across tasks, VLM-alone annotation approaches human accuracy on average (HNS = 97.0, where 100 denotes human-alone performance), demonstrating substantial potential to automate human-centered video analysis. Human verification achieves the highest accuracy (HNS = 121.5) while reducing human annotation time by 48.9% and monetary cost by 31.3%-44.5% relative to human-alone annotation. Our findings connect real-world human-centered video analysis tasks and current VLM capabilities, and clarify how human-AI collaboration can make VLM-assisted analysis reliable and efficient.