Science & Mathematics

The CLARKE building and road damage assessment AI classifier was incorporated into the Catastrophic Unmanned Remote Sensing Exercise (CURSE) 2026, which trained 23 small uncrewed aerial systems (sUAS) teams from state and local agencies and non-governmental organizations in Florida and Missouri on how to collect imagery for an incident command Remote Sensing Cell. Although CLARKE has been operationally deployed to two hurricanes (Debby and Helene), CURSE'26 presented an opportunity to evaluate the machine learning model with a larger set of never-before-seen orthomosaics. The results show that building and road damage assessment together ran in 3.5 minutes on average, and that building damage assessment achieved a Micro F1 of 0.834 with no manual adjustments to correct spatial misalignment and delivered that performance in both Florida and Missouri, demonstrating good geographic transfer. It also showed that road damage assessment performed poorly due to sensitivity to misalignment between road maps and the images, and to lighting and environmental conditions in an urban setting. The analysis of CURSE'26 presents five other findings of value to the sUAS community: that the use of sUAS and machine learning is now operationally normalized and accepted, sUAS teams in the field can improve classification accuracy by timing their flights, exercises are good operational tests but their value for evaluating learning models is less clear, edge computing is needed to overcome the wireless upload barrier, and the application of sUAS and AI to disaster response has ethical and legal ramifications for pilots and data managers.

Recently, spatially selective non-linear filter (SSF) was proposed for target speaker extraction, utilizing the direction-of-arrival (DOA) of the target speaker as a spatial cue. To overcome its dependency on a fixed microphone array geometry, geometry-conditioned SSF (GCSSF) was proposed, which maps the target DOA and the microphone positions into abstract high-dimensional embeddings using positional encoding. As an alternative to these high-dimensional embeddings, in this paper we propose utilizing two spatial features incorporating acoustic propagation principles for geometry conditioning: 1) the steering vector of the target speaker, which captures inter-microphone phase differences, and 2) a spatial mask based on GCC-PHAT, which quantifies presence of the target speaker per time-frequency bin. These features are used as input to a geometry-conditioning encoder and integrated via a FiLM layer into the GCSSF system. Experimental results on circular and random arrays demonstrate that conditioning with spatial features improves target extraction performance compared to abstract high-dimensional embeddings, maintaining a clear performance advantage across a wide range of target DOA errors.

The $k$-center problem is one of the most fundamental clustering problems and has applications in a wide range of areas, including facility location, network design, data mining, and machine learning. Given a set of objects together with a distance measure, the goal is to select $k$ centers such that the maximum distance from any object to its closest center is minimized. The $k$-center problem has been studied extensively from a theoretical perspective, both in metric spaces and on graphs. While there have been experimental studies for $k$-center clustering in metric spaces, a comprehensive experimental comparison of graph-based $k$-center algorithms on large-scale graphs is still lacking. In this paper, we bridge this gap by evaluating a set of algorithms ranging from classical approaches such as Gonzalez's greedy algorithm [TCS 1985], an algorithm based on the framework established by Hochbaum and Shmoys [Math. Oper. Res. 1985], to more recent methods, including algorithms by Thorup [SICOMP 2001] and Abboud et al. [SOSA 2023]. Our study considers both running time, solution quality and the effects of parallelism on a variety of synthetic and real-world datasets. Our goal is to provide a comprehensive comparison of state-of-the-art $k$-center algorithms on graphs and identify the strengths and weaknesses of different approaches in practice.

Most contemporary artificial intelligence (AI) systems are released with frozen parameters and improve through external updates, whereas an ideal system would keep improving from its own experience, a capacity we call recursive self-improvement (RSI). Yet studying RSI requires an environment that is continuously changing, objectively verifiable, and naturally grounded, posing a fundamental challenge for agent evolution. Here we introduce EvoForesight, a sustainable evolution agent system grounded in future event prediction, in which agents continuously interact with a dynamically evolving environment and improve by internalizing and externalizing the experience gained through these interactions. A living environment synthesizes verifiable tasks, and a recursive-evolution engine turns the reasoning trajectory behind each prediction back into capability: decoupling search from prediction adapts parameter-level evolution to the contamination-prone live environment, while distilling reusable skills lets the agent evolve from the limited live trajectories such an environment affords. On the external FutureX leaderboard, which evaluates predictions against real-world outcomes and is updated weekly, our self-improved forecaster ranked first across all systems over three consecutive weeks, achieving up to a 195% improvement in rating. On our synthesized evaluation, it ranked among the top three across four base LLMs, outperforming the strongest baseline by up to 10.7 Bradley--Terry points. More broadly, these results suggest that a system built for self-improvement is best understood as an agent that co-evolves with a genuinely living, open-ended environment, pointing to a shift in AI from training static models on fixed data toward cultivating agents that grow stronger from their own experience, with such an open world as the indispensable basis of that growth.

Pingmesh monitors datacenter networks through periodic pings between servers. In reconfigurable datacenter networks, topology engineering (TPE) and traffic engineering (TE) make uniform server-pair probing inefficient: a pinglist wastes budget retesting common paths while missing low-probability interfaces. Frequent control-plane updates further require rapid pinglist refresh, and cause path ambiguity at control-plane boundaries for fault localization. We present D-Pingmesh, a probing and fault-localization system co-designed with the TPE/TE control plane. Its control-plane-aware generator builds task-efficient pinglists by greedily admitting probes with high interface coverage under the current TPE/TE snapshot. An incremental replay-and-repair mechanism updates pinglists across reconfigurations rather than recomputing from scratch. At control-plane boundaries, a Rho-EM localizer jointly infers per-task old/new-path mixture weights and the fault set through EM-style iteration. We evaluate D-Pingmesh on an 8-pod reconfigurable topology under dynamic TPE/TE. At a 0.99 coverage target, D-Pingmesh uses 17.7x fewer tasks than Pingmesh and generates each pinglist about 5.1x faster than RD-Probe. In ns-3 simulation, boundary-aware localization improves micro F1 from 0.871 to 0.953 over the deTector PLL baseline. A testbed deployment demonstrates second-scale fault localization with low host overhead. These results demonstrate that D-Pingmesh is practical for reconfigurable datacenter networks with control-plane-aware monitoring.

Object detection models rely on large-scale annotated datasets that require substantial manual annotation effort. The labeling costs further increase when annotation uncertainty is captured through soft labels, as multiple annotations must be collected for each object. We address this challenge by introducing a semi-supervised framework for the uncertainty-aware annotation of object detection datasets. Our approach leverages open-vocabulary object detectors to generate region proposals and semantically meaningful object representations. Based on these representations, we construct an object-level similarity graph and propagate probabilistic label information. Our framework assigns soft labels to previously unlabeled objects, quantifying both aleatoric and epistemic uncertainty. We evaluate our method on four object detection benchmarks with soft-label annotations and demonstrate that high-quality annotations can be obtained with only a small fraction of manually labeled objects. Furthermore, we show that detectors trained on semi-automatically annotated data achieve performance comparable to models trained on fully manually annotated data. We achieve state-of-the-art downstream detection performance for the task of semi-supervised object detection on the COCO dataset, with 40.5% AP virtually matching the performance of the model trained on fully manually constructed data (40.9%) at only 1% of labeling costs. These results highlight the potential of open-vocabulary models and graph-based label propagation for efficient, accurate and uncertainty-aware pseudo-labeling.

In likelihood-free hypothesis testing (LFHT), given samples from two unknown distributions $P,Q$ on $[n]$ with $\mathrm{d_{TV}}(P,Q)\geq\varepsilon$, and from a third distribution $D\in\{P,Q\}$, the goal is to identify whether $D=P$ or $D=Q$. We study asymmetric likelihood-free hypothesis testing, where the sample budgets $s_P$ and $s_Q$ for $P$ and $Q$ may differ, and characterize the feasible triples $(s_P,s_Q,s_D)$ up to logarithmic factors. We establish that unequal reference budgets can give a polynomial saving. For instance, for constant separation, $n$ samples from $P$ and $n^{3/4}$ from $Q$ allow testing with $\widetilde O(n^{1/4})$ samples from $D$, whereas both existing baselines---symmetric LFHT and unequal-sample closeness testing---require $Ω(\sqrt n)$ samples from $D$. We further apply our results to non-adaptive distribution 2-clustering, where the sample allocation among $k$ distributions must be fixed before observing any data. We obtain matching upper and lower bounds for this problem up to logarithmic factors in terms of the domain size, number of distributions, smaller cluster size, and separation, both with one cluster distribution known and with both unknown. Comparing with known bounds for adaptive sample allocation gives a worst-case adaptivity gap of $\widetildeΘ(k^{1/4})$.

We study the set of weights occurring in the high-order binary Reed-Muller codes $RM(m-7,m)$. For every $m\geq 14$, we prove that the weight spectrum contains all even weights in $[336,2^m-336]$, together with an explicit finite set of low weights and their complements. The proof first establishes the corresponding base-code inclusion in $RM(7,14)$. This base step uses four-block concatenation, explicit degree-six monomial witnesses, and known spectrum inputs for $RM(5,10)$ and $RM(6,12)$. The base inclusion is then lifted to all $m\geq 14$ by a concatenation-based induction. We also analyze the remaining base-code target weights $322,326,330,334$ in several restricted families, including four-block branches, affine-flat and structured inner corrections in a fixed Carlet-type Maiorana-McFarland outer shell, and sparse sums of degree-7 coordinate monomials. The residual analysis delineates several construction routes that cannot yield the four unresolved base-code weights, thereby sharpening the remaining $c=7$ spectrum problem.

We study the maximum length of nontrivial simple-root cyclic maximum distance separable (MDS) codes over ${\mathbb F}_q$. We prove $n\le q+1$ if $\gcd(n,q^2-1)>2$, or if the dimension is even and $\gcd(n,q^2-1)>1$. The bound also holds for all lengths whose prime factors lie in $\{2,3,5,7\}$. We obtain further cases involving primes $\ell\ge11$ with $(\ell-1)/2$ prime, under conditions on the exponents and the characteristic. The proofs combine Fourier supports, locality bounds and field descent with arithmetic restrictions on multiplicative orders and prime-adic valuations.

Compositional zero-shot learning (CZSL) aims to recognize novel combinations of previously observed objects and attributes. Prior work showed that large-scale vision-language models encode decomposable semantic structures that can be recomposed to enable generalization. The Geodesically Decomposable Embeddings (GDE) framework formalizes this by identifying directions in the tangent space at the Frechet mean of the embedding distribution, which encode primitive visual concepts. Unseen attribute-object compositions can be synthesized through vector addition of these primitives. We revisit the assumption that compositional generalization requires explicit attribute-object pair re-composition. By means of new compositional scoring functions and extensive evaluations, we show that directly using the decomposed primitives suffices for recognizing unseen compositions. In that, we gain insights on how the geometry of the embedding manifold (Euclidean, spherical, or hyperbolic) and dataset structure affect compositional reasoning. Across models (CLIP, DINOv2, MERU, HyCoCLIP), we show that compositional structure consistently emerges as primitive representations, enabling inference on CZSL benchmarks (MIT-States, VAW-CZSL, UT-Zappos50K, Clothing16K) and group robustness on Waterbirds and CelebA. Code: https://github.com/nis-research/Compositionality-with-Primitives.

Literature-searching language agents are becoming a standard component of scientific assistants for the life sciences, yet there is little controlled evidence on which parts of their harness matter. We run a controlled ablation on biomedical claim verification (SciFact) with the live PubMed API as the only tool. Holding the backbone and output format fixed, we compare four harnesses: closed-book answering, single-shot retrieval using the claim as the query, a ReAct-style agent that writes its own PubMed queries, and an oracle given the gold abstracts, each with two efficient frontier backbones. Naive single-shot retrieval finds the gold paper for only 6% of claims and turns overconfident errors into abstentions without improving accuracy. Letting the model formulate its own queries triples gold recall and yields the best verdicts (62% accuracy vs. 49-59% for the other realistic harnesses), although these differences are not significant at our sample size and agents use fewer than half of their search budget. With gold abstracts, the same models reach 79-86% accuracy, a significant 17-24 point gap showing that retrieval, not reasoning, is the main bottleneck. Finally, a citation audit shows that closed-book models cite PubMed IDs that almost always exist (327/335) but are almost never relevant: a cross-family LLM judge rates only 3 of 327 existing citations as relevant to the claim, versus 67% for citations produced by the search agent. Existence checks are therefore insufficient safeguards for scientific agents. We release code, prompts, and all cached tool calls at https://github.com/saqlain2204/real-pmids-wrong-papers.

Incorporating AI systems into safety systems requires monitoring to avoid silent faults and ensure safe and reliable operation. Single monitors often do not provide actionable information for human supervisors, who act as the system's adaptive controller. We present COMET, a coordinated multi-layer runtime monitoring framework for human-supervised AI-enabled safety systems in which each monitor is anchored in a different fault mode. Monitor outputs are combined in a coordination layer and mapped by a communication layer to a traffic-light display. In this work, COMET consists of three monitoring layers: signal integrity, physics consistency, and the AI system's competence envelope. The framework is evaluated on an open dataset of compressed-air leaks, where an AI system detects leaks from acoustic emissions, with held-out conditions simulating covariate shift, semantic shift, and sensor faults, events expected during runtime but unknown in their exact appearance. All monitors are frozen after development and tested zero-shot, and the framework is compared against no-monitoring and single-monitor setups. The bare AI system fails silently and confidently under all three shift types, whereas COMET reduces the trusted-wrong-answer rate to zero on covariate-shifted no-leak data and on three of four sensor faults, at a nominal false-alarm rate of $\approx 1\%$. Single monitors come close in detection but cannot attribute the fault mode. COMET converts their anonymous detection flags into diagnoses. Residual silent misses under semantic shift define the boundary of passive monitoring. Runtime monitoring frameworks for AI-enabled safety systems must therefore combine quantitative fault detection and isolation with qualitative, human-oriented feedback.

Knowledge tracing (KT) models can estimate item parameters through an item response decoder trained solely for next-response prediction, yet predictive accuracy does not guarantee parameter recovery. We study 2PL, GPCM, and NRM decoders paired with LSTM, Transformer, and DKVMN encoders. Response-model derivatives show why discrimination and option slope updates depend directly on the scale and spread of estimated ability. Affine logit identities show how item-dependent shifts and rescaling in estimated ability are absorbed into item parameters without changing response probabilities. In simulations, difficulty, mean threshold, and correct-option intercept recovery remain high under shuffled order across representation widths, whereas discrimination and keyed option slope contrast recovery require wider item representations and degrade further under fixed item order. Separating the sequence item representation (width $d_S$) from the item parameter key (width $d_M$), and pairing a wide key with a thin sequence representation ($\mathrm{SK}(64,8)$), reduces fixed-order ability distortions and improves recovery and cross-order consistency without sacrificing test accuracy or increasing model size relative to a shared width-64 baseline ($\mathrm{SH}(64)$). On TIMSS and EdNet, widening $d_M$ consistently improves agreement with an offline IRT reference, while the trade-off from narrowing $d_S$ depends on the encoder's reliance on sequence width for prediction. These findings show why item parameters, especially discrimination and option slopes, must be evaluated separately from response prediction.

A graph is called Pfaffian if it admits a Pfaffian orientation. The Pfaffian recognition problem asks whether a given graph is Pfaffian. In 1989, Vazirani and Yannakakis showed that this problem is polynomial time equivalent to testing whether a prescribed orientation is Pfaffian, and that the problem belongs to $\mathsf{coNP}$. The complexity of the problem for general graphs has remained open since then. For bipartite graphs, polynomial time recognition algorithms were obtained by Robertson, Seymour, and Thomas in 1999, and independently by McCuaig in 2004. We show that testing whether a given orientation is Pfaffian can be reduced to polynomial identity testing. This gives a randomized polynomial time algorithm for Pfaffian recognition on general graphs, placing the problem in $\mathsf{coRP}$.

Microservices give teams independent deployment and scaling, but each deployed service also reserves compute and adds operational work. This paper examines a second deployment edition of a multi-tenant .NET core-banking platform. The existing edition runs as nine AWS ECS Fargate tasks; the alternative composes the five API services by project reference into one API host, alongside a consumer host and a background host. The service code remains in its original repositories, and both editions continue to be built from the same commits. We describe the composition mechanism, module manifest, start-up checks, request handling and configuration precedence, then compare the editions in an emulation of the AWS deployment. The composed edition exposes all 844 declared operations. Across 47 financial workload cases, both editions produced the same observed outcomes and matching ledgers. In read probes across five principal types, 230 of 231 endpoint results matched; the remaining difference was associated with the emulated load balancer. Eleven cross-tenant and write-authorisation cases also produced the expected outcomes in both editions. In the emulation, the composed edition delivered 2.3 times the read throughput and used less than half the memory. Estimated service-task compute at list prices was 11-24% lower, excluding the shared gateway, load balancer, data stores and other operating costs. The consolidation also enlarges the failure domain: one API-host crash takes all five modules offline. Of 16 injected faults, the edition's checks detected 10, behavioural tests detected five additional faults, and one fault escaped both. These findings concern one platform and an emulated deployment that mirrors its topology, task sizes and configuration; they are not production-AWS measurements.