Science & Mathematics

Long contexts increase the memory cost of the key-value (KV) cache in large language model (LLM) inference. Sparse attention reduces attention computation and memory traffic by selecting historical tokens, but the retained KV cache must remain available for later decoding steps. We present UlaCache, a KV cache management system built around a unified logical address space spanning GPU KV pools and optional CPU memory. Stable logical references let the inference engine allocate and select KV independently of physical placement. Common allocation and sparse-access interfaces support data-parallel (DP) and tensor-parallel (TP) attention. For distributed storage, local GPU buffers retain selected KV across decoding steps, so only missing entries require transfers. When CPU storage is enabled, Parallel Bucketed LRU (PB-LRU) manages a GPU pool shared across requests. It updates recency metadata and selects eviction victims in parallel. UlaCache transfers KV and then updates the address mappings. With speculative decoding enabled under matched CPU and GPU memory budgets, UlaCache achieves peak decoding throughput speedups of 1.16 $\times$ (DP) and 1.70 $\times$ (TP) on H200, and 1.73 $\times$ (DP) and 2.51 $\times$ (TP) on B300, compared with the state-of-the-art baseline.

Magnetic resonance imaging (MRI) is an acquisition technique used for medical diagnosis. It is notably used for structural and functional MRI (or fMRI) image analysis. MRI data is known to be complex and noisy, so pipelines are created to extract and enhance relevant information in retrieved images. The NARPS study highlighted the significant variability that can emerge from different pipelines set up by independent teams while trying to answer the same research question about brain functioning. Having different implementations realizing the same task is a typical playground for software product lines (SPLs), which may be out of the sight of neuroimage practitioners. Thus, proposing an SPL out of these different pipelines falls back on a typical reverse-engineering problem with its usual difficulties. In this document, we advocate for the conceptual adoption of SPL principles as a promising direction to improve reproducibility and consistency in neuroimaging research.

This paper presents a novel strategy for UAV detection through visible (RGB) and infrared (IR) fusion, in order to improve accuracy and precision in harsh environments and weather conditions. In this research, two opposite approaches have been studied and implemented: the first concerns image-level (early) fusion, which merges two different spectra into a single image to which one detector is applied. On the other hand, the second approach, called decision-level (late) fusion, focuses on blending together the output bounding boxes of two different detection pipelines, aiming to create a common result that can be applied to both spectra simultaneously. Both strategies have been extensively tested in a dedicated environment, which included alternating low and high brightness scenes, different backgrounds (sky and ground), distances (5, 25 and 60 m), altitudes and zoom levels (1X, 2X and 4X). These tests allowed for the assessment of actual detection capabilities and system reliability in real-world applications. Furthermore, the two fusion strategies were evaluated and compared over a set of aligned images in order to highlight the better approach of the two. The results indicate that early fusion achieves better raw performance, since it incorporates the thermal information from the beginning of the training process. However, late fusion presents an innovative and useful detection workflow that can be applied in cases of sensor or frame misalignment. Both positive and negative aspects have also been pointed out and discussed, in order to give a more comprehensive overview. Finally, the paper outlines implementation considerations and practical recommendations based on operational priorities. Overall, the findings contribute to the research in RGB-IR fusion modalities, with the goal of enhancing UAV detection capabilities.

General-purpose robots must handle tasks, objects and environments they were never trained on, yet zero-shot generalization remains one of the least solved problems in physical AI. In digital AI, zero-shot capability grew with the volume and, above all, the diversity of training data, much of it sourced from the crowd. Robotics instead relies on paid collection by a few well-funded teams, which yields volume without diversity. We formulate zero-shot generalization in physical AI as coverage of the deployment distribution, and show that more samples from the same collectors shrink the unseen region only slowly, while each new contributor can extend it. Building on this and on zero-shot learning, we propose a shared latent space for robot actions that combines multi-modal alignment, generative augmentation, and noisy-label learning. As a scalable source of diverse data, we propose a community of developers building applications on one standard, approachable robot, Origami. We close with a testable hypothesis and an evaluation protocol relating real-robot zero-shot success to the diversity of contributed data.

Deep learning models for chest X-ray anomaly detection remain vulnerable to subtle distributional shifts (e.g., acquisition technique, patient-related factors, and preprocessing). Traditional error analysis often relies on semantic metadata or model embeddings, which can mask low-level signal variations that degrade performance. In this work, we propose a data-centric framework for automated failure mode discovery using spectral analysis. We project images into the frequency domain and extract a compact profile summarizing the distribution of signal energy across frequency bands. By performing unsupervised clustering on these spectral profiles, we demonstrate that model failures are not randomly distributed, but are strongly concentrated within specific spectral clusters. This method effectively isolates "blind spots", enabling the prediction of model reliability and the discovery of performance-degrading data slices without requiring ground-truth failure annotations.

Calibrating large language models (LLMs) in open-ended generation is uniquely challenging because many distinct token sequences can express the same underlying meaning. Prior calibration techniques target either fixed-set classification confidence or the binary correctness of individual strings, making them inherently ill-suited for the dynamic, multi-string nature of open-ended semantic outcomes. To bridge this gap, we introduce S-EDL, a novel evidential method that natively optimizes LLMs for semantic calibration. By eliciting evidence from the sequence likelihoods of the model itself, S-EDL constructs a differentiable Dirichlet prior over prompt-specific semantic classes. Optimizing the evidential loss under this prior yields a calibration-aware training signal over variable semantic classes, completely bypassing the need for fixed multiple-choice options. Evaluations across four open-ended benchmarks and three models demonstrate that S-EDL substantially reduces semantic calibration error while improving accuracy. By aligning generative likelihoods with semantic correctness, this work establishes a framework for deploying trustworthy LLMs in complex, open-ended applications. Code is available at: https://github.com/sandylaker/s-edl.

In this paper, we investigate the Linear Complementary Pairs (LCPs) of Maximum Distance Separable (MDS) codes that arise from elliptic curves. We establish geometric conditions for LCPs of elliptic codes and prove that they cannot consist entirely of MDS codes when evaluated over all rational points of the curve. To address this, we construct LCPs of MDS elliptic codes by restricting evaluation divisors to subgroup cosets. We apply this method to obtain explicit infinite families of LCPs of MDS codes over Mordell and supersingular elliptic curves, including large-length constructions over quadratic extensions.

Finite-field inversion is a basic nonlinear building block in the design of S-boxes for symmetric cryptography. In inverse-based S-boxes, finite-field inversion is often combined with linear or affine layers that may preserve or destroy algebraic symmetries of the inverse function. We study how often nontrivial linear self-symmetries remain as the outer invertible linear transformation varies over \(GL_n(\mathbb F_2)\). Using the classification of two-sided linear equivalences of inversion, we reduce the problem to semilinear conjugacy and centralizer estimates. We obtain matching upper and lower bounds for the probability that a nontrivial linear self-symmetry remains, and these bounds determine its leading quadratic exponent. This exponent is governed by the smallest prime divisor of the extension degree. Thus the decay rate depends on the arithmetic of the field extension and not only on its dimension. As cryptographic applications, we provide a practical rigidity criterion, prove that the full AES S-box has no nontrivial linear self-symmetry, and illustrate the broader relevance of the framework through ARIA and Vision.

Movement primitives commonly represent robot mo- tions through basis functions indexed by time or phase. Inspired by spacetime representations employed in physics, we propose a spline-based formulation in which spatial geometry and temporal progression are represented within a common set of continuous variables. The proposed method jointly estimates a shared Bezier curve, while allowing spatial and temporal model complexity to be adapted independently (e.g., demonstration-specific phase modulation). As use case, we consider an alternating optimization process that simultaneously estimates path intervals observed in partial demonstrations. Simulation experiments demonstrate accurate joint recovery, refinement through modifications in model complexity and continuity, and improved robustness to variance in the demonstration set. Finally, we demonstrate online retiming and synchronization capabilities in a patient-turning task assisted by a robot, evaluated with a physical manikin.

The mid-wave infrared (MWIR, 3-5 microns) is indispensable for thermal sensing, gas detection, long-range imaging, and free-space optical communications. Meta-optics offer a compact, lightweight, and potentially low-cost alternative to conventional refractive optics, making them attractive for size, weight, and power (SWaP)-constrained systems. However, their strong chromatic dispersion has largely restricted imaging to narrow spectral bands. Here, we demonstrate broadband imaging across the full 3-5 micron MWIR band using a single-layer silicon metasurface coupled with computational reconstruction. Remarkably, the reconstruction model is adapted using only four image captures, overcoming large training data requirements and enabling broadband image recovery despite the strongly wavelength-dependent optical response. In contrast to approaches that engineer complex, dispersion-optimized metalens profiles for broadband operation, our method leverages the wavelength-dependent information encoded by a simple 2cm diameter hyperboloid metalens. We show that this metasurface captures sufficient broadband intensity information for computational reconstruction across the entire MWIR band. These results establish a simple route toward broadband computational meta-optics without requiring broadband achromatic metalens designs, offering a promising platform for compact MWIR imaging systems.

Autonomous mobile robots operating in dense crowds and cluttered sites require $360^\circ$ perception to react to surrounding hazards. While panoramic LiDAR offers full-surround coverage, learning representations within reinforcement learning (RL) is sample-inefficient under sparse rewards, and existing 3D pretrained point-cloud encoders are too memory-heavy for massively parallel RL simulation. We present RangeNav, a compact (183k parameters) LiDAR range-image encoder pretrained entirely on synthetic data. RangeNav uses anisotropic convolutions and circular padding to handle the extreme aspect ratio and azimuthal seam of range images. We pretrain the network using a navigation-aligned objective combining range reconstruction, semantic segmentation, and bird's-eye-view (BEV) prediction, then freeze it for policy learning. In simulation, adding semantic and BEV supervision increases success rates over a reconstruction-only baseline by an average of 32.03 and 23.73 percentage points (pp) in procedurally generated training and out-of-distribution realistic test environments, respectively. Zero-shot real-world deployments on a Unitree B2-W quadruped demonstrate robust navigation through dense crowds, low-light settings, and construction zones where limited-field-of-view depth policies struggle. Code and models will be open-sourced. Project website: https://rangenav.github.io/.

Topological semimetals are of increasing interest to replace conventional metals for interconnects in future semiconductor chips, due to their potential to exhibit decreasing resistivity with reducing wire dimensions and a lower sensitivity to electron scattering at defects. First-principles evaluation of promising topological semimetals for interconnects typically employ non-equilibrium Green's functions (NEGF) techniques to estimate defect scattering, but such calculations for specific defect geometries are difficult to systematize and compare between disparate materials. Here, we develop an NEGF-based approach to systematically and rapidly evaluate the defect sensitivity of electron transport in materials without depending on material-specific defect geometries. Specifically, we analyze the eigen-spectrum of the transmission matrix for an incoherent ensemble of all-possible single-atom vacancy defects, which clearly reveals eigenvalues close to 1 for defect-insensitive transport modes. Applied to the leading candidates, we show that this robustness to defect scattering is extremely sensitive to surface orientation and Fermi level for each material. We find nearly-perfect insensitivity to defects only for wavevectors with single transport modes within specific Fermi level ranges in CoSi(001) and PtAl(001). Importantly, we also find high insensitivity for certain non-isolated modes in NbAs that emerge specifically due to spin-orbit coupling; these modes could not be identified previously as they were buried in the total transmission. These non-isolated modes have been buried in previous studies of total transmission, and can now be directly compared to other materials due to our standardization of defect-creation. This technique improves the state-of-the-art NEGF method and enables unbiased evaluation of future topological semimetal candidates for nanoscale interconnects.

Geometric structure is essential for robotic manipulation and understanding the 3D world. Most VLA models rely on 2D representations of current observations, while geometry-aware approaches often require additional 3D information. However, geometry inferred independently from individual observations is often discarded rather than accumulated into a persistent task representation. Our key insight is that action generation needs a compact, task-relevant geometric abstraction that can be continuously updated, rather than a complete, renderable 3D reconstruction at every step. We introduce GeoMem-VLA, a VLA framework that internalizes explicit 3D supervision into dedicated geometric tokens and maintains them as temporal memory for action generation, without directly aligning the VLM's native visual embeddings with the teacher feature space. During training, a frozen geometry foundation model supervises learnable GeoPlan tokens through a geometric flow-matching objective over the current scene point map and future end-effector tracks. These tokens are integrated with the memory of observation representations, where historical information is retrieved, adaptively fused with current observations, and consolidated to retain event frames as the task progresses. At inference, the policy retrieves historical representations and fuses them with the current working representation to predict actions, without invoking the 3D teacher or explicitly reconstructing the scene. After benchmark-specific downstream adaptation, GeoMem-VLA achieves 98.9% average success on LIBERO and 91.10% / 91.24% on RoboTwin's clean/randomized settings. Across five real-world tasks, GeoMem-VLA achieves 78.0% average success. These results suggest that internalizing and remembering 3D scene understanding is a promising approach to robot control.

Galaxy morphology out to $z\sim 6$ provides insight into galaxy formation and evolution, but identifying rare or structurally complex systems is difficult due to observational limits and the scale of modern imaging surveys. Here, we apply AdaCLIP, a Zero-Shot Anomaly Detection (ZSAD) model built upon the Contrastive Language-Image Pretraining (CLIP) model and adapted using non-astronomical anomaly-detection datasets, to identify structurally unusual galaxies without galaxy-specific anomaly-detection training or morphological labels. We extract anomaly scores and 768-dimensional feature embeddings from 38,853 galaxies in the Cosmic Evolution Early Release Science (CEERS) and Next Generation Deep Extragalactic Exploratory Public (NGDEEP) surveys, and compare the resulting anomaly scores with parametric GALFIT and nonparametric statmorph measurements. The feature embeddings reveal structure associated with galaxy morphology and observational properties, while anomaly scores show modest trends with structural measures, including asymmetry, concentration, $M_{20}$, and size, but weaker or less consistent trends with Sérsic index, axis ratio, smoothness, and Gini coefficient. These results highlight the potential of ZSAD as a scalable tool for identifying unusual galaxy morphologies in large astronomical imaging surveys, including those anticipated from the Nancy Grace Roman Space Telescope.

Coding agents that run long enough must compact their context, summarizing the conversation, discarding the original, and carrying the summary forward as memory. We show that in popular open-source coding agents this step is an exploitable trust boundary. Harnesses that flatten the typed conversation into a single plain string, marking each message with a text label such as [User]:, destroy message provenance in code, before the summarizer runs. An attacker who controls anything the agent reads can forge a user turn that survives compaction as a genuine instruction. Across five flattening harnesses (pi, opencode, Cline, Aider, OpenHands) the forged approval launders into the agent's memory 87-100% of the time, up to a ~908,000-token context window; across four role-typed harnesses (Codex, Gemini CLI, Qwen Code, Continue) the identical attack succeeds 0% of the time. We follow the attack to executed consequences. A laundered standing approval drives real tool-running agents to execute attacker-controlled curl | sh in a sandbox, and end to end in the live pi TUI at ~50% against a 0% paired clean arm. The same channel reliably steers an agent's plan away from a file an attacker designates off-limits, which can blind a code review to a planted backdoor. Because the provenance is destroyed deterministically in harness code, no prompt-level defense can restore it; we measure a code-level provenance check that catches 77% of laundering while demoting 2% of legitimate summary content. The defect also has a straightforward fix at the serializer, escaping the role labels before flattening; the role-typed harnesses avoid it entirely by keeping messages role-typed.