Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Friday 6 November

Computer Vision

TriNetra: Thermal-Reliability-Guided RGB-Thermal Fusion and Tracking for Surveillance

Visible cameras fail at night and thermal sensors fail under saturation, blur or loss of contrast, yet most RGB-thermal (RGB-T) detectors mix the two streams with fixed weights and cannot tell when one modality has become unreliable. TriNetra is a reliability-aware surveillance framework built around this problem. The proposed solution computes a frame-level Thermal Reliability Coefficient (TRC) from five label-free image cues, uses it to gate the thermal features of a dual-branch YOLOv8m detector whose visible branch, neck and head are frozen and COCO-pretrained, and passes the same score to a ByteTrack-based tracker that feeds a rule-based behaviour analyser. On the M3FD benchmark (2,940/630/630 split) the detector reaches 87.6% precision, 79.4% recall, 70.0% mAP@50 and 47.8% mAP@50:95 on the test split, with 17.0 M of its 42.9 M parameters trained. TRC stays near 0.78 on clean frames (standard deviation 0.012), decreases with the severity of thermal saturation and blur, reaching 0.53 and 0.69 at full severity, and is insensitive to RGB-thermal misalignment. With identical weights, the adaptive gate matches a fixed gate on clean data (70.0% versus 69.8% mAP@50) but keeps 43.6% mAP@50 when the thermal stream is corrupted, against 38.4%, 29.2% and 25.0% for a fixed gate, full thermal admission and a closed gate; the 5.1-point gain over the fixed gate has a 95% bootstrap confidence interval of +3.6 to +6.5 points. In tracking, the TRC-gated association performs on par with ByteTrack on a synthetic benchmark with exact ground truth and on 14 M3FD clips, and is slightly less stable under thermal saturation, so its measured benefit is confined to the detector. The limitations of the current evaluation are reported together with the results.
Read the full abstract

All authors Sadique Khatib, Aditya Ahirrao, Vedant Sonawane, Disha Kapse, Y. D. Bhise

Computer Vision

Evaluating Anterior Cruciate Ligament Risk of Injury Using a Graph Convolutional Autoencoder

Current solutions for assessing the risk of anterior cruciate ligament (ACL) injury typically rely on expensive motion-capture labs that require substantial resource investment or subjective evaluations of specific athletic stress tests by experts. The Drop Vertical Jump (DVJ) and the Single Leg Hop (SLH) are two examples of such tests typically used to track progress in ACL rehabilitation. In this paper, we develop an unsupervised learning model that runs entirely on a smartphone for two tasks, based on an autoencoder architecture with multiple layers of graph convolutional networks, which utilizes pose estimation data to determine a person's injury risk status. Our model accurately predicts athletes at high risk of injury for both athletic stress tests, with average sensitivities and specificities of 78% and 78% for the DVJ, and 75% and 74% for the SLH. Our findings support the development of a widely accessible smartphone solution for real-time screening of ACL injury risk.
Read the full abstract

Machine Learning

Fixed-point acceleration on Neural Wasserstein manifolds

We propose a fixed-point acceleration method for optimization over probability measures, applicable to a variety of machine learning problems. We represent probability measures as pushforwards of a reference distribution through a neural network map and equip the parameter space with the induced pullback Wasserstein metric. This geometric structure allows natural gradient descent to be interpreted as a fixed-point iteration on a Riemannian manifold. We accelerate this iteration using Riemannian Anderson mixing method, which exploits historical residuals to capture approximate second-order information without forming Hessians. In the present parametric setting, the manifold is represented by Euclidean coordinates endowed with a point-dependent metric, simplifying retraction and vector transport. We evaluate the method on multimodal sampling, PDE-based inverse problems, aggregation dynamics, Bayesian logistic regression, and generative modeling. Across the tested tasks, Anderson mixing reduces the number of natural-gradient iterations and achieves competitive performance relative to SGD and Adam in terms of both loss-gradient evaluations and wall-clock time. These results suggest that fixed-point acceleration is an efficient tool for improving the computational efficiency of training on parametric Wasserstein methods.
Read the full abstract

Artificial Intelligence

FOSCAR: Accelerating Decode Attention over 2- and 4-Bit KV Caches through Affine Folding

Large-language-model decode attention streams the key-value (KV) cache for every generated token. Packing that cache into 2-bit or 4-bit quantized integers reduces memory traffic but does not automatically translate into latency saving. Existing 2-bit fused kernels reach under 31%/15% of peak HBM throughput on H100/B200. Converting each value and applying its scale and offset before feeding Tensor Cores is a major bottleneck because instruction-issue throughput is scarce on modern GPUs. Instead of applying scale and offset to each KV value, FOSCAR uses affine folding to apply them to the other operand or the output of decode-attention matrix products, depending on the quantization scheme (token-wise or channel-wise). This reduces scale-and-offset operations by up to two orders of magnitude while preserving the dequantized attention output in exact arithmetic. FOSCAR also uses bit manipulation to feed KV operands to Tensor Cores efficiently, requiring only 1 and 0.19 ALU instructions per INT2 element on H100 and B200, respectively. Finally, FOSCAR builds warp-specialized GPU pipelines for Hopper (H100) and Blackwell (B200) using wgmma and tcgen05. On H100, FOSCAR INT2 and INT4 kernels achieve geo-mean speedups of $3.36\times$ and $2.83\times$ over FlashInfer BF16, and $2.03\times$ and $1.71\times$ over TensorRT-LLM FP8. On B200, the corresponding INT2 and INT4 speedups are $2.67\times$ and $2.30\times$ over FlashInfer BF16, $1.48\times$ and $1.28\times$ over TensorRT-LLM FP8, and $1.16\times$ and $1.00\times$ over TensorRT-LLM NVFP4. We integrate FOSCAR into SGLang to benchmark serving efficiency. On long-context serving workloads, FOSCAR increases peak aggregate throughput by up to 170% and 341% over 2-bit and 16-bit baselines, respectively. We open-source FOSCAR as a kernel library at https://github.com/FutureMLS-Lab/FOSCAR.
Read the full abstract

All authors Donglin Zhuang, Pragaash Ponnusamy, Xiaoxia Wu, Shuaiwen Leon Song, Zhongzhu Zhou

Cryptography and Security

MemoryDojo: Benchmarking Poisoning Through LLM Agent Memory

LLM agents benefit from using memory systems to improve performance on long-horizon, multi-session tasks. However, memory introduces a critical attack surface. Malicious instructions encountered in one session may be stored and propagated through memory before being retrieved and acted upon in later sessions. Because memory systems differ in how they process and retrieve information, the downstream effects of poisoning can vary significantly across systems. To quantify the effects of poisoning across different memory architectures, we introduce MemoryDojo, a benchmark for tracing the lifecycle of poisoned information through agent memory and measuring its influence on subsequent agent behavior. MemoryDojo is easily extensible, consists of 155 multi-session long-horizon trajectories covering four domains, and supports five different memory systems. Through our evaluation, we find that all memory systems are susceptible to memory poisoning, with substantial variation in downstream effects. We further propose a memory-aware poison optimization process that increases the poisoning efficacy, highlighting differences between memory architectures. We hope that MemoryDojo will support research on understanding and mitigating agentic memory risks.
Read the full abstract

All authors David H. Yang, Blake Bullwinkel, Amanda Minnich, Mohammad Mohammadi Amiri, Giorgio Severi

High Energy Astrophysical Phenomena

Forecasting Fundamental-Plane Tests of Intermediate-mass Black Holes in Radio-selected Dwarf Galaxies

A radio/X-ray Fundamental Plane (FP) test of an intermediate-mass black hole is informative only when the compact radio emission, hard X-rays, accretion state, and host association refer to the same accretor. We used published VLA, VLBA, Chandra, and HST observations of a radio-selected dwarf-galaxy sample to determine when present nondetections are constraining and what observations are required for a stronger test. The VLBA follow-up reveals that compactness alone does not establish association with the target dwarf, while for the best nuclear candidates the radio component relevant to the FP has not yet been isolated on the intermediate physical scales between the VLA and VLBA measurements. We therefore recast the FP as a forward question: for an assumed black-hole mass, how likely is a given radio observation to detect the compact core? The calculation includes intrinsic FP scatter, X-ray uncertainties, the radio spectral index, the accretion state, and source size relative to the beam. The existing VLBA nondetections have low completeness across the adopted intermediate-mass range and do not provide useful black-hole mass upper limits. Meaningful tests require deeper resolution-matched radio imaging, recovery of intermediate-scale structure, and contemporaneous measurements of the intrinsic hard-X-ray component. Because the calculation returns a mass-dependent detection function for each source, the same framework can be applied to larger dwarf-galaxy samples to quantify survey completeness and build selection functions for population studies of low-mass black holes.
Read the full abstract

Optimization and Control

Row-sparse ah-symmetric reflexive generalized inverses with a few extra rows: local search with approximation guarantees

For a rank-$r$ matrix $A\in\mathbb{R}^{m\times n}$, ah-symmetric reflexive generalized inverses $H$ with few nonzero rows give least-squares solutions that are fast to evaluate and use few variables. A known determinant-based local search finds one with exactly $r$ nonzero rows and 2,1-norm within an essentially tight factor $r$ of the minimum at an exact local optimum. We let the nonzero rows of $H$ lie in a set $K$ of $k\ge r$ indices and study how to choose $K$. We bound the loss of this restriction (restricted over unrestricted minimum 2,1-norm) by a simple geometric certificate and establish that a local search maximizing the squared rank-$r$ volume of the corresponding columns of $A$ gives loss at most $\sqrt{k(r+\epsilon k)/(k-r+1)}$, where $1+\epsilon$ is the swap tolerance. With $\epsilon=0$ this is $r$ at $k=r$ and less than $\sqrt{2r}$ for $k\ge2r$; with $\epsilon=1/k$ the search runs in polynomial time on rational data. The same support yields an ah-symmetric reflexive generalized inverse with spectral norm within a factor $\sqrt{1+(n-k)(r+\epsilon k)/(k-r+1)}$ of that of the Moore-Penrose pseudoinverse. The loss guarantees hold for every norm on the rows. In general, our certificate bound is tight up to a factor depending only on $k/r$, and on some instances, for a suitable row norm, so is the loss bound; for the 2,1-norm, however, the certificate can overestimate the loss by a factor $\sqrt r$. The local search is the Fedorov exchange algorithm of 0/1 D-optimal design; this yields a fast implementation and, via the Kiefer-Wolfowitz theorem, supports of at most $r(r+1)/2+1$ rows with certificate at most $\sqrt r$, best possible in general. On instances from earlier work, the observed loss is far below the guarantee, and the method matches two ADMM heuristics when few rows are removed and otherwise gives 2,1-norms up to 27% smaller, one to two orders of magnitude faster.
Read the full abstract

Algebraic Topology

How many points does it take to see the homotopy of a space? Minimal acyclic categorical models

Every connected homotopy type is the classifying space of a small category, but this general result gives little control over the categories involved. We study finite acyclic categorical models and measure their complexity by the number of objects. Our main tool is the sink lemma, which describes the addition of a sink as the attachment of a cone along an explicit map. Using it, we determine the minimum number of objects required to model spheres, bouquets of spheres, Moore spaces of cyclic groups, real projective spaces, and lens spaces. We also exhibit infinitely many pairwise non-isomorphic rigid minimal models of the sphere. To include spaces admitting no finite model, we introduce a truncated invariant measuring the minimum number of objects required for a finite acyclic category to approximate a space through degree n. Spreading a finite monoid over n+1 objects produces such a category with an n-connected map to the monoid's classifying space. For every nontrivial finite group, this bound is optimal, although its classifying space admits no finite acyclic model.
Read the full abstract

Computer Vision

Direct Regression and Geometric Recovery for Event-based Spacecraft Pose Estimation

Event cameras offer high temporal resolution and dynamic range for uncooperative spacecraft pose estimation, but models trained on synthetic event streams must generalize to the noise, background activity, and contrast variation of physical sensors. We present a comparative systems study of two independently developed approaches to synthetic-to-real, event-based 6-DoF spacecraft pose estimation. The first is a modular pipeline that combines learned keypoint detection and unlabeled target-domain adaptation with geometry-based pose recovery and temporal smoothing. The second directly regresses translation and rotation from global and cropped event representations using an apparent-scale cue, physics-motivated synthetic corruptions, and feature-consistency regularization. We evaluate both systems under the same SPARK 2026 hidden-test protocol on the real-event split of SPADES, with predictions for all test frames. The results reveal a component-wise reversal concealed by the aggregate challenge score: direct regression reduces orientation error by 22.7% (1.1767 to 0.9101 rad) and obtains a 13.4% lower composite score, whereas the modular pipeline reduces relative translation error by 78.7% (0.1335 to 0.0285). Archived configurations and a qualitative failure case identify plausible explanations for these complementary error profiles, without isolating their causes. These findings motivate a testable hybrid design that combines direct orientation estimation with geometry-constrained translation recovery.
Read the full abstract

Computational Complexity

Tensor value and the 9/4 bound for matrix multiplication

We give an exposition of OpenAI's recent result that the exponent of matrix multiplication is at most 9/4 valid over every field. Our main tool is a fractional symmetrized variant of the tensor value introduced by Coppersmith and Winograd, used in place of spectral points (tensor characters) in the original proof. We prove a superadditivity inequality for a power of this value on partially direct sums, whose summands share one tensor factor and occupy independent subspaces in the other two. Together with the degenerations of polynomial multiplication tensors from the OpenAI proof, this inequality yields the bound.
Read the full abstract

All authors Mahmut Levent Doğan, Vladimir Lysikov, Keiya Sakabe, Michael Walter

Artificial Intelligence

Intent-Aware Caching for Efficient LLM Serving

Modern LLM serving systems use prefix caching to accelerate multi-turn conversational workloads. They cache the key-value states of previous conversation prefixes and reuse them when subsequent requests share the same prefix, to avoid redundant prefill computation and reduce serving latency. However, achieving a high cache hit rate remains challenging because cache reuse depends on whether a conversation continues and how soon the next turn arrives. Our analysis of real workloads shows that user intent is a useful signal for this behavior, as different intents exhibit different continuation probabilities and inter-turn delays. Based on this key insight, we propose Intent-Aware Caching (IAC), an online eviction policy that classifies requests by intent, learns lightweight reuse statistics from recent serving logs and leverages these statistics to guide an oracle-inspired eviction strategy. Extensive evaluation on real workloads (WildChat and ShareChat) shows that IAC reaches a 34.6% cache hit rate. This substantially outperforms both the previous learning-based approach (18.8%) and traditional methods such as least recently used (LRU, 12.2%), while reducing average serving latency by up to 46.1%.
Read the full abstract

All authors Kazi Hasan Ibn Arif, JinYi Yoon, Dimitrios S. Nikolopoulos, Hans Vandierendonck, Deepu John, Bo Ji

Image and Video Processing

Beyond Segmentation: Enriching ImageCAS with Coronary Ostia Landmarks, Rooted Centerline Graphs, and Open-Weight Localization Models

Coronary ostia and centerlines are key reference points for automated analysis of coronary computed tomography angiography (CCTA), from surgical planning to intraoperative vessel tracking. However, the only public ostium annotations cover 77 non-gated aortic scans, not CCTA, while CAT08 and ASOCA (32 and 40 CCTA scans) mark the ostia only implicitly, as centerline start points. This work enriches ImageCAS, the largest public dataset for coronary artery segmentation, with ostium annotations for its 1000 patients and centerline graphs rooted at these new ostia, making it, to the authors' knowledge, the largest public dataset for ostium localization and the only one with centerlines following the official ImageCAS masks and starting at annotated ostia. Three annotators marked each ostium in the first 500 volumes; in the next 500, the candidates of a model trained on the first batch were corrected (human-in-the-loop). Six 3D localization models, trained on ImageCAS, CAT08 and ASOCA, were benchmarked on 58 held-out volumes (50 ImageCAS, 4 CAT08, 4 ASOCA). SwinUNETR-V2 and nnResUNet localized all ostia without failure (>5 mm error), with median errors of 0.70-0.87 mm. Trained on CAT08 and ASOCA alone, SwinUNETR-V2 fails on 34% of right and 53% of left ostia, whereas adding 250 ImageCAS volumes removes all failures. The centerlines were extracted by a deterministic, quality-controlled minimal-path procedure; on ASOCA, they cover a median of 99.5% of the reference length, with a median per-patient mean distance of 0.22 mm between matched reference points and centerlines. Annotations, centerlines, code, and weights are available at https://hal9000-lab.github.io/.
Read the full abstract

All authors Matteo Leccardi, Andrea Moglia, Marco Morandini, Michele Brembilla, Alessandro Nicosia, Andrea Penserini, Christian Brandini, Ender Konukoglu, Marco Marcon, Luca Mainardi, Pietro Cerveri

Quantum Physics

Shaping Frequency-Bin Biphoton Correlations via Programmable Optical Frequency Combs

Programmable shaping of photon-pair spectral correlations is important for frequency-bin quantum information processing. However, achieving this control remains challenging in nonlinear optical systems without reliable forward models. Here, we demonstrate model-free, hardware-in-the-loop shaping of the measured joint spectral intensity of a cascaded photon-pair source driven by a programmable optical frequency comb. A blockwise simultaneous perturbation stochastic approximation controller optimizes up to 22 pump, signal, and idler parameters. Each iteration requires only two accepted hardware evaluations, independent of the parameter count. The controller converges to diverse target patterns on 3$\times$3, 4$\times$4, and 5$\times$5 frequency-bin grids. The optimized distributions achieve Pearson correlations between 0.66 and 0.98 and classical overlap fidelities between 0.32 and 0.62. Cross-target similarity analysis confirms that the optimized patterns remain target-specific. Source characterization yields a maximum coincidence-to-accidental ratio of approximately 585. Accuracy is limited by pump-phase transfer, parasitic nonlinear background, waveshaper constraints, amplitude-only filtering, and measurement noise. Joint spectral intensity contains no spectral-phase information. Therefore, the results demonstrate control of measured frequency-bin populations and correlations, not complete reconstruction of the two-photon state. Phase-sensitive characterization is required to verify inter-bin coherence and quantify the entanglement dimensionality of the output state. Our results establish direct measurement feedback as a practical approach to reconfigurable frequency-bin correlation engineering when an accurate end-to-end model is unavailable.
Read the full abstract

All authors Usman Khan, Daniel Lopez, Venkat Aneesh Padmasola, Nirmal Dangi, Santosh Kumar, Yuping Huang

Machine Learning

ICare-DT: a digital twin for continuous forecasting of intensive care patients

Digital twins of intensive care patients should update forecasts from new data and predict treatment responses, but few studies test whether routine records support either. We propose ICare-DT, a Transformer-based twin that updates a latent patient state hourly from new measurements without retraining. We trained it on 140,473 eICU-CRD stays to forecast vital signs crossing their training-set means. On test data, later forecasts beat the twin's own hour-6 forecasts of the same windows in all 12 settings (+0.073 to +0.153 AUROC). On validation data, controls that withheld or swapped the new hours traced the gain to the patient's own measurements. Without refitting, the model kept its gain on a second database (8 of 9 settings, +0.025 to +0.114 AUROC) in a prespecified evaluation of the previously analyzed MIMIC-IV test split. Refitted within MIMIC-IV, a twin state fine-tuned on clinical endpoints gave gradient boosting a significant gain for acute kidney injury (+0.0035 AUROC; hypotension +0.0010, not significant). To test response prediction, we built a benchmark that replaces the treatments in real records with simulated doses and injects responses of known size, visible only at measurement times. Ridge models supervised with injected effects recovered them from real records only through frequently measured variables (heart rate) read at their measurement times. We asked standard learners to predict these effects for a dosing regime no patient had received. No standard learner outperformed predicting zero effect at an out-of-support query. When told how the simulated responses were generated, a separate learner recovered the effects from the same records without seeing their true values (R2 0.30 to 0.40 at one effect size). Routine records support forecasts that improve with each patient's new data, and our benchmark is a testbed for models that learn how responses are generated.
Read the full abstract

All authors Yanke Li, Yahang Qi, Yumi Kim, Diego Paez-Granados

Number Theory

Euler trace congruences over local fields

Let $K$ be a finite extension of $\QQ_p$ with uniformizer $\pi$ and residue field $\FF_q$. For each $r\ge1$, we determine the largest integer $E_r$ such that $\Tr(A^{q^r})\equiv\Tr(A^{q^{r-1}})\pmod{\pi^{E_r}}$ for every integral matrix $A$ with a prescribed residual similarity class. We give formulas for both nilpotent and nonnilpotent residual classes and construct integral lifts attaining the resulting exponents. We also exhibit local fields with the same ramification index and residue field for which the optimal exponents differ, even when $r$ and the residual similarity class are fixed.
Read the full abstract

Physics and Society

Slow structures, fast degrees of freedom, and the morphology of chess openings

Chess openings naturally separate into slow and fast degrees of freedom: pawns evolve through irreversible moves, whereas pieces can be rapidly reorganized within the resulting structure. We exploit this separation to construct a quantitative morphology of opening space. From a catalogue of 3810 named variations, 1749 reach a common depth of $N=10$ plies, corresponding to 619 distinct positions and only 435 joint pawn structures. We characterize these positions through observables measuring pawn commitment and spatial organization, blockade and tension, and piece activity and centralization. Their partial independence reveals a multidimensional opening space and in order to explore its structural organization, we construct a network whose nodes are distinct White pawn structures, connected when a single pawn move or capture transforms one into another. Remarkably, 139 of the 149 observed structures belong to a single connected component containing $98\%$ of all positions. This network exhibits a statistically significant modular organization, with eight communities whose representative structures correspond to familiar pawn formations. Rather than forming isolated categories, these formations emerge as distinct regions of a connected structural landscape. Our approach thus complements the traditional opening taxonomy by revealing similarities between differently named openings and providing a quantitative basis for comparing the positions they produce.
Read the full abstract

Machine Learning

RidgeFill: Training-Free Late-Layer Cache Fill for Faster Prefill

RidgeFill shortens prefill by not computing most of a model's upper layers for the prompt. The lower layers run exactly up to a split; from the split layer's state a per-layer map predicts every later layer's input, and each late layer writes its cache from that prediction. The design follows YOCO and DeepSeek-V4.1-Flash, where the projection is trained into the model. RidgeFill retrofits it onto models that were never trained for it: the map is a ridge regression fitted once in closed form from one pass over calibration text, the layer's own projections run unchanged, and a chosen set of rows, the last few thousand tokens, and every decoding step run the full model. Nothing is trained and no token is dropped. Most of the work was in measurement, finding a repeatable methodology. Loss averaged over all tokens hides RidgeFill's error. Needle retrieval in the filled region, loss on tokens that depend on distant context, and splice tests that swap filled keys or values for exact ones show which component carries it: the keys on Llama-3.1-8B, half of it in their scale, and the values and the recurrent state on Qwen3.5-9B. On Qwen3.8-27B in a llama.cpp fork on an A100, time to first token is 1.25 to 1.31x faster at 16K to 120K tokens with the maps in VRAM; RULER changes by -1.2 and -1.4 points and LongBench v2 by -1.4; the 95 % intervals include zero but do not rule out losses of a few points. On Qwen3.8-Flash-Next in vLLM it is 1.50 to 1.60x faster than exact at 9K to 32K tokens, with RULER's needle tasks at 99.8 against 100.0 and perplexity on novel tokens 2.65 % higher [-0.02 %, +4.93 %]. On three 8 to 9B reference models, at the same split and tail as the identity fill, the fitted map retrieves 49.1 and 49.6 more filled-region RULER points on the two dense models (at that split Llama's loss is worse), and its long-document loss is +/-0.03 nats of the released SwiftKV checkpoint's.
Read the full abstract

Robotics

A wearable input/output interface learns personalized models of haptic perception

Wearable haptic interfaces deliver physical stimuli to the skin, yet perceived locations vary widely across users and diverge from actuator locations, particularly for emergent tactile-feedback techniques. We developed a wearable haptic input/output (I/O) device integrating a 36-channel electrotactile array with capacitive touch sensing, enabling stimulation and user-reported percept locations to be registered within the same body-referenced coordinate frame. In 24 participants, the device revealed electrotactile and referred-thermal perceptual maps with notable interparticipant variability. Despite this variability, both modalities retained electrode-pair-specific spatial structure across participants. Tactile and thermal percepts also showed spatial correspondence, whose strength varied across participants. These findings motivated our human-in-the-loop calibration framework, which adapts a forward-prediction strategy to individual users and generates personalized perceptual models. These models enabled inverse rendering by selecting stimulation patterns predicted to generate a target percept for a specific user. Using these models, we demonstrate personalized haptic rendering in haptic-guided robot teleoperation, visual-to-haptic sensory substitution, interpersonal haptic communication, and virtual reality (VR). Together, this platform provides a route toward adaptive haptic interfaces that measure, model, and render according to each user's perception.
Read the full abstract

All authors Wen Li, E. Magnus Adams, Zijie Xu, Jiawen Yao, Aybala Esmer, Xiaoyue Ni, Bozhi Tian, Pedro Lopes

Cryptography and Security

Empirical Assessment of Post-Quantum Cryptography Adoption Status in TLS

Anticipating Q-day security concerns, major CDNs and cloud providers are migrating to post-quantum cryptography (PQC). While existing literature benchmarks PQC performance, empirical research evaluating real-world adoption data remains scarce. We present an empirical assessment of PQC deployment across seven critical protocols (HTTPS, LDAPS, SMTPS, IMAP-SSL, POP3-SSL, TELNETS, and RDP) using 10,000 target IP addresses per protocol. At the origin-server level, PQC algorithm support reaches 9.1% for HTTPS and 7.39% for LDAPS, while falling below 1% for all other evaluated protocols. At the entity level, PQC support is 11% for HTTPS and 5.4% for LDAPS, while the remaining protocols have less than 5% of entities supporting PQC on the hosted servers. We further correlate these adoption trends against organizational digital and financial footprint, and industry sectors, yielding a multi-dimensional empirical assessment of PQC deployment. Finally, at the domain level, we perform a comparative analysis of PQC adoption in Cloudflare-hosted domains against baseline Cloudflare datasets. Collectively, our findings provide critical insights into the current state of PQC readiness over Transport Layer Security (TLS).
Read the full abstract

All authors Md Ariful Haque, Sander Vinberg, Ben Edwards, João Cruz

Numerical Analysis

Finite-time Cholesky Factorization

We propose an algorithm for estimating the Cholesky factorization of positive, semi-definite matrices without pivoting. Our method has the same data access pattern as the canonical Cholesky method, and therefore is amenable to familiar hardware optimization techniques. Our method interprets the factorization task as a variational problem and in turn solves it using an induced dynamical system. We study the rate of convergence of this dynamical system to its stationary state for both positive definite and semi-definite matrices. Numerical experiments validate that our method is competitive with pivoted Cholesky on several practical examples.
Read the full abstract

Mesoscale and Nanoscale Physics

Cavity-enhanced second harmonic generation from Rydberg excitons in Cu$_2$O

Rydberg excitons in Cu$_2$O provide a solid-state setting for creating giant optical nonlinearities. Here, we use second-harmonic generation (SHG) to study Rydberg excitons in a doubly resonant optical microcavity designed to enhance both the pump and second-harmonic fields. We observe even-parity exciton states in an optical cavity and extend cavity-based measurements of Rydberg excitons up to $n=10$. The SHG intensity depends strongly on pump cavity detuning, showing the dominant role of the microcavity is the enhancement of the intra-cavity pump field. Analysis reveals an additional enhancement from the resonant second-harmonic cavity giving an overall SHG enhancement factor of $21 \pm 2$ inside the microcavity. These results establish cavity-enhanced SHG spectroscopy as a valuable tool for studying Rydberg excitons in microcavities and provide a pathway towards exploiting optical nonlinearities in Rydberg exciton polariton systems.
Read the full abstract

All authors Horatio Q. X Wong, Krzysztof Sawicki, Madeleine Fisher, Stephen A. Lynch, Wolfgang Langbein, Sai Kiran Rajendran, Hamid Ohadi, Matthew P. A. Jones, Liam A. P. Gallagher

Materials Science

Theory of Optical and Magneto-Optical Properties of [111]-Oriented Site-Controlled Quantum Dots

Site-controlled [111]-oriented quantum dots (QDs) are promising sources of non-classical light, yet their optical spectra exhibit apparently conflicting symmetry signatures. At zero magnetic field, excitonic spectra are commonly interpreted as $C_{2v}$-like, whereas measurements in Faraday geometry display selection rules characteristic of $C_{3v}$-like symmetry. Here, we show that both regimes can be described consistently without invoking any change in the underlying structural symmetry. Combining 8-band $\mathbf{k}\cdot\mathbf{p}$ theory, a self-consistent Hartree-informed configuration interaction scheme, and group theoretical analysis, we introduce a controlled symmetry reducing perturbation that lowers the QD symmetry to $C_s$. At zero field, the optical response remains predominantly $C_{2v}$-like because the additional low-symmetry transitions carry only weak oscillator strengths. In Faraday geometry, the same reduced-symmetry structure produces four dominant, nearly circularly polarised trion transitions, reproducing the experimentally observed $C_{3v}$-like response. By contrast, an exact $C_{2v}$ description inferred from the zero-field spectrum permits only two such trion transitions and therefore fails at finite field. Our results thus provide a unified description of the zero and finite magnetic field optical response without any restoration or enhancement of the underlying structural symmetry. More generally, we demonstrate that a higher-symmetry-like optical spectrum does not uniquely identify the exact symmetry of a QD, therefore also providing useful guidance for other QD and nanostructure systems that lie beyond the ones studied here.
Read the full abstract

All authors N. D. O'Connor, G. Juska, E. Pelucchi, S. Schulz

Cryptography and Security

Empirical Study of Post-Quantum Cryptography Adoption Status in SSH Infrastructure

With the anticipated arrival of Q-day, quantum computers threaten current cryptographic schemes for encryption and digital signatures. In response, major content delivery networks and cloud service providers are already adopting Post-Quantum Cryptography (PQC). The current state-of-the-art focuses on benchmarking the operational performance of implementing PQC, but the real-world adoption status remains under-researched. This work utilizes large-scale SSH fingerprint data from a proprietary scanning tool to analyze PQC adoption across 1.52 million hosts and 27,012 entities. Our study shows that the PQC transition in the SSH protocol is underway but highly polarized: 45.2% of SSH hosts on port 22 support PQC, while 54.8% remain classical-only. At the entity level, 39.9% have enabled PQC on at least one host, whereas 60.1% have no PQC support. By correlating PQC adoption with industry, geography, and organizational size (digital and financial), this study provides a comprehensive empirical assessment of PQC adoption status in SSH infrastructure.
Read the full abstract

All authors Md Ariful Haque, Sander Vinberg, Ben Edwards, João Cruz

Systems and Control

Adaptive PD-like Fault Tolerant Controller for a Class of Nonlinear Systems

The article presents a nonlinear adaptive control strategy for a class of feedback-linearizable systems subject to process faults and disturbance. The controller is derived through Lyapunov analysis and incorporates an adaptive proportional gain within a PD-like structure, preserving simplicity while providing nonlinear adaptive behavior. A residual-based gating mechanism activates the adaptation only when the tracking error exceeds a prescribed threshold, preventing unnecessary gain updates during nominal operation and improving robustness to noise. The proposed design yields uniform global ultimate boundedness of the closed-loop signals, with explicit expressions for the steady-state tracking bound. Practical tuning guidelines for the adaptive parameters and Lyapunov matrix selection are provided to ensure consistent gating behavior. Simulations on a precision positioning system illustrate effective compensation of process faults, rapid gain escalation during degraded conditions, and smooth reversion to nominal performance, demonstrating that the method enhances fault tolerance while retaining the structural simplicity of PD control.
Read the full abstract

All authors Alessio Dallabona, Vicenç Puig, Mogens Blanke, Dimitrios Papageorgiou

Machine Learning (Statistics)

Where Should Learning Live? Regret Analysis for Text, Parameter, and Joint Adaptation

Language-model systems can select prompts and parameter adapters repeatedly during deployment, using feedback to guide later choices. Earlier choices may change the context or environment inherited by later configurations. We study selection among preconstructed candidates and compare cumulative loss with the best fixed prompt–parameter pair used from the beginning. This excess loss separates into three terms: selection regret within the available family, the gap from restricting that family, and the benefit or harm of the actual deployment history. Selection is evaluated using the history each candidate would have produced if used throughout, which separates candidate quality from the effects of switching. When loss functions are fixed independently of selector randomness, past choices affect loss for only a bounded number of rounds, and every candidate's own-history loss is revealed, limiting switches yields sublinear expected regret within the selected family. We also quantify the costs and additional assumptions of independent probing and deployment-only feedback. Controlled experiments show that fewer switches can either lower or raise deployed loss, depending on whether inherited histories are harmful or helpful. With Qwen3.5-9B on Spider, text-only selection has lower within-family regret than joint selection, yet produces more failures because its candidate family excludes stronger configurations.
Read the full abstract

All authors Jungtaek Kim, Jaden Park, Jongwon Jeong, Emmanouil-Vasileios Vlatakis-Gkaragkounis, Kangwook Lee

Artificial Intelligence

Attend at the Right Resolution: Head-Adaptive Sparse Attention via Query-Key Interaction Structure

Attention is a major scalability bottleneck in large language models (LLMs), with its computational cost and key-value (KV) memory traffic becoming increasingly prohibitive as context length grows. Sparse attention addresses this bottleneck by restricting each query to a subset of the context, but its effectiveness hinges on identifying relevant regions both cheaply and reliably. Existing training-free methods primarily rely on runtime relevance estimates, with selection dimensionality and granularity typically governed by fixed heuristics rather than the intrinsic interaction structure of pretrained attention heads. We take a different perspective: the query-key operator of each pretrained head defines a characteristic response structure that determines the resolution needed for sparse selection, both across feature directions and across context tokens. Building on this insight, we propose Head-Adaptive Sparse Attention (HASA), a training-free sparse attention framework that jointly adapts selection resolution along these two dimensions. Along the feature dimension, the singular spectrum of each head's query-key operator determines an effective rank, yielding a compact, head-specific response space. Along the sequence dimension, neighboring regions are progressively merged according to the directional coherence of their token responses in this same space, while heterogeneous regions remain separate. Exact attention is then computed using the original full-dimensional KV states of the selected regions. HASA captures 88.1% of dense-attention mass and, compared with UNIQUE, a strong recent sparse-attention baseline, improves accuracy by up to 3.47 and 2.57 points on RULER and LongBench-Pro, respectively, and accelerates self-attention by up to 1.32x, while achieving up to 14.07x speedup over dense attention.
Read the full abstract

All authors Wajih Hassan Raza, Ahsan Bilal, Yidan Shen, Renjie Hu, Xin Fu, Yu Wen

Machine Learning

Verifier Noise Is a Temperature: Testing Tilt Predictions in Finite-Budget LLM-Guided Search

LLM-guided search loops propose a modified candidate, score it, and keep it if it scores better. When both candidates are re-scored at every comparison, as in the minibatch acceptance gates of prompt optimizers, keep-if-better is a reversible Markov chain if the score difference is logistic and the mutation operator is reversible, and its stationary law is the mutation equilibrium tilted by exp(beta f). For any continuous symmetric noise difference the same tilt is the stationary law of a small-step diffusion limit, with beta = 4F'(0); for a minibatch gate beta grows with the square root of the minibatch size. Caching the incumbent's score instead inflates the recorded score. We test the tilt as a prediction for LLM-guided program search, committing each prediction from calibration data before the search runs. The level after 150 steps rises with the minibatch size in six of seven settings, but its predicted value fails: on bin packing, search attains 62-88% of the predicted gain when the proposer edits the parent and 109-119% when it writes new heuristics. The attained fraction falls with the parent-child quality correlation of unselected mutation; a registered test on two new settings confirmed the direction, but not the quantitative agreement after a correction. Further registered tests found that runs started from a good program kept losing quality over 300 steps, more so from starts that search had selected, by a margin that passed its registered rule narrowly. Runs from different starts had not reached a common level, so whether the tilt describes the stationary law of LLM editing remains open. A registered test in prompt optimisation on frozen task-model scores failed: search exceeded the tilt, and the bin-packing relation did not transfer.
Read the full abstract

Statistical Computation

A Generalized Bayesian Approach to Clustering with Dimension Reduction

Generalized Bayes product partition models provide a probabilistic counterpart to loss-based clustering. They update a prior over partitions through an exponentiated clustering loss rather than a likelihood, so allocation uncertainty can be quantified without a full generative model for the data. However, their performance depends on the space in which the loss is evaluated. For high-dimensional observations, a squared Euclidean loss on the raw features can be dominated by irrelevant coordinates or fail to reflect cluster structure lying near a nonlinear low-dimensional manifold. We propose an extension in which an autoencoder and the partition are learned together, so that the space in which the clustering loss is evaluated is shaped by the clustering itself. Our generalized posterior combines a reconstruction loss, which keeps the low-dimensional representation faithful to the observed data, with a within-cluster sum of squares in the latent representation, which encourages cluster separation. Posterior computation alternates between Gibbs sampling over partitions and gradient-based updates of the autoencoder network, with the learning rate estimated jointly. We illustrate the approach through simulation studies and applications to four benchmark datasets.
Read the full abstract

All authors Shen Jia, Weipeng Huang, Geng Yang, Eoin Gallagher, Tin Lok James Ng

Complex Variables

Geometric folding and maximal asymptotic variance

Based on the recent breakthroughs by OpenAI and consequently by Binder-Ivrii-Prause-Smirnov, we develop a finite-scale folding method for the nonlinearity of a schlicht function, based on a signed critical-point count for superharmonic potentials and a quantitative target cutoff. Furthermore, we use a cylindrical root transform to reduce the logarithmic cost of access to the point at infinity. Combined with a Littlewood-Paley formula as well as dilation, the method yields the maximal asymptotic variance bound $\Sigma_{\max}^2\le1$ for the unrestricted schlicht class of normalized univalent functions.
Read the full abstract

Soft Condensed Matter

Influenza A Virus as a Multivalent Active Particle: Statistical Mechanics of Binding, Motility, and Antiviral Design

In this contribution, we review existing multivalent models of Influenza A Virus (IAV) systems to study virion dynamics and design of multivalent entry blockers. We review existing biochemical estimations of the dissociation constants and catalytic activities of the two virion ligands [hemagglutinin (HA) and neuraminidase (NA)], which can be used to develop quantitative theories and experiments. Multivalent models are statistical mechanics frameworks for studying the simultaneous interaction of multiple reversible ligand-receptor bonds. They quantify the effects of both combinatorial entropy terms and single-molecule ligand-receptor affinities. Multivalent frameworks led to the rationale of superselectivity, the ability of probes to sense specific concentrations of conjugated receptors. Multivalent models guided the design of antiviral nanomedicines constituted by polymeric scaffolds conjugated with small molecules such as sialic acid receptors targeting the ligands on the surface of the virion. Importantly, neuraminidase is an enzyme that limits the number of host-guest bonds through receptor desialylation. Like active particles, neuraminidase activity enhances virion motility in the extracellular medium. Modeling the dynamics of Influenza A virions requires developing out-of-equilibrium multivalent models.
Read the full abstract

All authors Sumati Bhatia, Andreas Herrmann, Thomas Kolbe, Lalan Kumar, Daniel Lauster, Susanne Liese, Bortolo Matteo Mognetti