Algebraic Geometry
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Thursday 5 November
The multi-monopole equations are a generalisation of the Seiberg–Witten equations where the spinor is coupled to an auxiliary vector bundle. The signed count of multi-monopoles on a 3-manifold can vary according to a new type of wall-crossing phenomenon. For products $S^1\times C$, where $C$ is a closed Riemann surface, we relate the wall and chamber structure to loops in the complement of the wobbly divisor of $C$. Such a loop induces a monodromy map on the multi-vortex moduli space, and the count of its fixed points is expected to agree with the count of multi-monopoles on $S^1\times C$ in the given chamber. By studying singularities of framed multi-vortex moduli spaces, we are able to understand these monodromy maps for various loops. Counting fixed points suggests that product 3-manifolds display incredibly rich wall-crossing behaviour.
Signal Processing
13,000-Fold Frequency Down Conversion From 280 MHz to 21 kHz Using Mechanical Frequency Combs
Optical frequency combs bridge the microwave and terahertz regimes by down-converting terahertz optical energy into microwave signals. A photodetector converts the comb's equally spaced lines, generated through optical nonlinearities, into a low-frequency tone at their spacing. Here, we present the first mechanical version of this process, using mechanical frequency combs to downconvert energy over 4 orders of magnitude, from the MHz to kHz regime. Thin-film lithium niobate (LN) is used to design an acoustic resonator with a wide range of mode frequencies. We achieve phononic comb excitation from 250 MHz to 900 MHz and compare the electrical and laser Doppler vibrometer (LDV) measurements at 277 MHz. We then show the low-frequency spectrum of the devices, highlighting mechanical motion at the comb's spacing and demonstrating energy transfer from 277 MHz to 21 kHz.
Read the full abstract
Audio and Speech Processing
Conflict-Aware Meta-Learning for Generalizable Speech Deepfake Detection
Speech deepfake detectors built on large self-supervised learning front-ends achieve high in-domain accuracy, yet suffer degradation under domain shift caused by unseen synthesizers, transmission codecs, or unfamiliar acoustic conditions. Meta-learning, particularly model-agnostic meta-learning for domain generalization (MLDG), addresses this issue by simulating domain shifts. However, applying MLDG directly introduces a bi-level optimization challenge that remains unsolved. First, effective domain-shift emulation requires genuine distributional gaps between meta-splits, yet such gaps can frequently produce gradient conflicts, destabilizing the bi-level optimization. Second, the stochastic nature of mini-batch gradients causes the reference direction to fluctuate across iterations, resulting in an optimization trajectory that fails to achieve domain generalization. We propose a conflict-aware meta-learning (CAML) framework, which addresses both challenges through three complementary components. Firstly, base domain gradient surgery (base-DGS) introduces an asymmetric projection that treats the meta-train gradient as a primary anchor. Secondly, EMA-DGS replaces the unstable anchor with its exponential moving average (EMA), providing a stable optimization trajectory anchor that substantially reduces domain-shift-induced conflict detections. Thirdly, layer-wise DGS (LW-DGS) further extends the framework to individual parameter tensors, enabling fine-grained, layer-specific intervention that targets conflict-prone layers while leaving well-aligned parameters intact. Experimental results demonstrate that the proposed CAML framework achieves lower EER than the compared baselines on most cross-dataset evaluations across multiple cross-dataset benchmarks and architectures, achieving a maximum relative improvement of 45.2% in terms of EER.
Read the full abstract
Human-Computer Interaction
Goals as a Representational Layer for Human-AI Collaboration
Goals are central to how people, teams, and organizations structure their work. Yet, while generative artificial intelligence (AI) systems increasingly operate across a worker's context, information about why a person is pursuing a task, beyond local prompts and interactions, is typically missing. We argue that goals should become an explicit representational layer mediating interactions among people, AI systems, and organizations. Because human goals are hierarchical, partially latent, and dynamically evolving, such a layer must be multidimensional, both grounded in human articulation and inferred from available evidence, and continuously updated. A goal layer could connect immediate requests to broader outcomes, improve the organization and retrieval of relevant context, support reflection and coordination, and provide a foundation for training, aligning, and evaluating AI systems in terms of human-centered outcomes. We highlight these opportunities while identifying open challenges involving goal inference, socio-cognitive frictions, alignment, privacy, and governance.
Read the full abstract
Artificial Intelligence
Debate to Deceive: Poisoning LLM's Fact-Checking via Game-Theoretic Multi-Agent Debate
Knowledge attacks on Large Language Models, such as manipulating fact-checking judgments, pose significant risks to AI reliability. Existing approaches – adversarial suffix/template injection and single-agent self-reflection – lack robustness across diverse defenses and are prone to confirmation bias and costly trial-and-error, respectively. We propose Debate-to-Deceive (D2D), a multi-agent debate framework grounded in human debate structure and rhetorical style, enabling agents to generate adversarial passages that are semantically coherent, rhetorically effective, and resistant to early confirmation bias. D2D formalizes the adversarial process as an evolutionary game in which agents adaptively evolve their rhetorical strategies (Logos, Pathos, Ethos) through competition. To manage the intractably large LLM output space, we decouple agent behavior into a strategic layer, which selects a rhetorical action from a compact strategy space via evolutionary-inspired learning dynamics, and a generation layer, where the LLM produces a concrete passage conditioned on that strategy and the debate context. On four fact-checking benchmarks (FEVER, HealthVer, CLIMATE-FEVER, and SciFact), D2D substantially outperforms existing attacks and evades most defenses.
Read the full abstract
Computer Vision
AtariBench: Evaluating Spatiotemporal Reasoning of Vision-Language Models in Atari Gameplay
Vision-Language Models (VLMs) have achieved strong performance on static visual understanding benchmarks, yet their ability to act in dynamic environments remains less well understood. We introduce AtariBench, a controlled benchmark that evaluates VLMs across 20 Atari 2600 games, requiring models to make sequential decisions from high-temporal-resolution gameplay streams. By decoupling fast in-game dynamics from wall-clock inference speed, AtariBench focuses on whether models can perceive rapid visual state changes, reason over spatiotemporal dynamics, and ground actions in the current observation. Our experiments reveal substantial limitations in state-of-the-art VLMs: models often fail to infer object motion and direction, exhibit brittle sequential decision-making, and over-rely on textual interaction history despite newly appended visual feedback. We further introduce a separability analysis to identify which in-game scores provide reliable signals of model competence, and use the benchmark to study how memory design and prompting strategies affect performance. Overall, AtariBench exposes a persistent gap between passive visual perception and active decision-making, providing a lightweight and rigorous testbed for developing VLMs that can reason and act in dynamic environments.
Read the full abstract
Image and Video Processing
Measurement-Aware Multimodal LLM for No-Reference Low-Dose CT Quality Assessment
No-reference image quality assessment (IQA) of low-dose CT predicts the score that radiologists would assign to a slice without a full-dose reference, as needed to compare dose protocols and denoisers. Trained CT IQA networks agree well with radiologists but output only a score, whereas multimodal large language models (MLLMs) can explain a score but perform poorly on CT when only prompted. We propose a measurement-aware MLLM, built on Qwen3.8-27B, that rates a slice from its image together with six image-domain measurements of noise, sharpness, high-frequency content and streaks, given to the language model as percentiles in text. On the LDCT-IQA 2023 test set, it reaches an Overall score of 2.735, on par with the retrained challenge winner (2.735) and MedSigLIP-FiLM (2.713), and substantially above its base MLLM used as a zero-shot agent. Shuffling the measurements at test time reduces the score by 0.048, mostly through noise, but the same measurements also lead the model to assign higher scores to over-smoothed denoised images.
Read the full abstract
Machine Learning
Random Symbolic Boosting: A Geometric Approach to Symbolic Regression
Symbolic models offer interpretability, but their construction through generational evolution can be computationally expensive. This article presents Random Symbolic Boosting (RSB), a method that constructs explicit additive symbolic models via boosting over populations of randomly generated syntax trees, eliminating the need for generational evolution. The idea combines two observations: high-dimensional spaces can be spanned by enough linearly independent vectors, and the semantics of syntax trees can be directly manipulated as such vectors. It is shown that, for Ordinary Least Squares, the boosting process guarantees a monotonic reduction of the training error for both regression and classification. Furthermore, an adaptive geometric early stopping mechanism is introduced. It is based on the angle between the prediction and target vectors, is independent of the problems scale, and requires no adjustments across tasks. Evaluated on a standard benchmark for symbolic regression, the method is located at the Pareto-optimal point between predictive accuracy and training time, ranking first on the real-world datasets. Additionally, it was tested on a large tabular classification benchmark with 200 problems, achieving competitive results.
Read the full abstract
Artificial Intelligence
AttentionSwitch: Jailbreak Defense for LLMs via Joint Learning of Sparse Attention-Head Selection and Scaling
Large language models have achieved remarkable performance across a wide range of tasks, yet they remain susceptible to jailbreak attacks that circumvent safety mechanisms and elicit harmful or unintended outputs. Although safety can be improved through model fine-tuning, such approaches can be computationally expensive and may degrade previously acquired capabilities through catastrophic forgetting. Recent jailbreak defenses have therefore explored targeted interventions on specific model components associated with unsafe behavior. However, existing approaches typically decouple two inherently related decisions: where to intervene and how to modify the selected components. Such sequential optimization may be suboptimal because the optimal intervention location can depend on the intervention applied there, and vice versa. In this work, we introduce AttentionSwitch, the first jailbreak defense to jointly optimize intervention sites and strengths within a frozen LLM. Specifically, AttentionSwitch learns a sparse set of attention heads together with query-conditioned multiplicative scaling factors, allowing the outputs of the selected heads to be either attenuated or amplified without altering the directions of their representations. Across popular language models and jailbreak attacks, AttentionSwitch outperforms existing defenses, with particularly strong results when training data is scarce.
Read the full abstract
Differential Geometry
A new characterization of the round sphere
We provide an elementary proof of Hersch's inequality and Cheng's theorem, both characterizing the sphere. We revisit these classical results in connection with the geodesic monotonicity of heat kernels on two-dimensional spheres, which provides a third characterization.
Fluid Dynamics
Immobilisation and remobilisation of surfactant-covered drop interfaces in electric fields: Simulations
Fluid–fluid interfaces are seldom clean due to the presence of surfactants, which can strongly modify their dynamics. Interfacial flow redistributes surfactant, creating concentration gradients whose Marangoni stresses can immobilise the interface. By contrast, surface diffusion or exchange with the bulk can reduce these gradients and restore interfacial motion. We study this competition using a two-dimensional model of a surfactant-covered leaky-dielectric drop in a uniform electric field. Electric stresses drive the classical Taylor circulation, in which the drop remains non-rotating. At stronger fields, convection of interfacial charge can destabilise this state and produce the Quincke regime, characterised by spontaneous mean rotation. Using a fully nonlinear model on a fixed circular interface, we show that weak surfactant diffusion allows Marangoni stresses to suppress the Taylor circulation almost completely. Quincke rotation remobilises this nearly arrested interface and restores a substantial straining circulation, whose orientation varies with the applied field. Surfactant also shifts the critical field for rotation and can change the onset from subcritical to supercritical. Surface diffusion and exchange with a well-mixed bulk reservoir can produce similar flow responses and rotation thresholds despite different surfactant distributions. The spatial variation of surfactant concentration therefore provides additional information for distinguishing the two transport mechanisms.
Read the full abstract
Computers and Society
How are viral trends different from educational and commercial uses of LLM image generation and editing? A privacy-preserving descriptive analysis
Gemini's Nano Banana models, first introduced in late 2025, led to a shift in the capabilities and use of image generation and editing. We utilize Urania, a Google-developed framework for generating insights about LLM chatbot interactions with rigorous differential privacy (DP) guarantees, to analyze use cases and intents, techniques and ways of prompting, without ever looking at a single image or individual prompt. We select a subsample of 2.1M conversations in early 2026 which utilized Nano Banana models, into roughly 800 clusters in five categories: (1) Social & Viral Sharing, (2) Brand & Marketing Collateral, (3) Educational & Technical Information Media, (4) Product Visualizations & E-commerce, and (5) Architectural & Spatial Designs. We then compare these clusters on 21 features of use, measuring aspects of prompts' (a) editing techniques, (b) artistic style requests, (c) composition, and (d) chatbot interactions. This allows us to describe a baseline of how people are using image generation and editing tools for different use cases—(including growing and emergent use cases which were not possible before Nano Banana)—in a privacy-preserving manner.
Read the full abstract
Statistics Applications
Dynamic Smart Charging Through Integrated Agent-based Simulation and Optimization Modeling
Coordinated electric vehicle (EV) charging can improve alignment between transportation operations and power-system conditions as electrification expands. This study develops an integrated simulation-optimization framework that embeds demand-aware charging (DAC) and grid-responsive charging, including unidirectional smart charging (V1G) and vehicle-to-grid (V2G), within an agent-based model, enabling dynamic feedback between optimized vehicle (dis)charging decisions and evolving transportation-system states. Using the Austin 6-county region, the framework evaluates 15,000 shared autonomous electric vehicles (SAEVs) under DAC, V1G, fleet-only V2G, and combined fleet-plus-household V2G. Results show coordinated strategies concentrate charging toward midday and provide selective evening discharge while maintaining service performance. Under favorable pricing, V2G reduces net CO2 emissions from 45 kg/SAEV/day under DAC to below 40 kg and lowers emission intensity by up to 14% to 0.14 kg per vehicle-mile traveled. DAC and V1G offer near-term options, whereas high V2G costs may hinder adoption. Realized benefits depend on tariffs, vehicle availability, and V2G maturity.
Read the full abstract
Differential Geometry
Projective flatness of cylindrically symmetric Finsler metrics
In this paper, we study cylindrically symmetric projectively flat Finsler metrics \(F=|\bar y|\phi(x^0,r,s,z)\). We reformulate the conditions for projective flatness and strong convexity in terms of \(\Omega=\phi-s\phi_s-z\phi_z\), isolating the compatibility condition missing in earlier work. Using the method of characteristics, we give an integral formula for all smooth local solutions near points with \(r>0\) and \(s=z=0\). These solutions are determined by a characteristic potential up to the addition of an invariant exact one-form. In the reversible case, the equation for the potential reduces to a \((2+1)\)-dimensional wave equation on a half-space. We illustrate the reconstruction with a polynomial potential and compute the potentials of a known family of spherically symmetric metrics and their linear pullbacks. The convexity conditions also admit decreasing profiles and recover the spherical Beltrami metric as a particular case. Finally, suitable positive sums of linear pullbacks yield smooth reversible projectively flat Finsler metrics on \(\mathbb R^{n+1}\) whose tangent norm at the origin is non-Riemannian. Consequently, these metrics are not linear pullbacks of spherically symmetric metrics
Read the full abstract
Statistics Theory
An Asymptotic Framework for Closed-Form Estimators from Surrogate Estimating Equations
Many statistical estimators, including maximum likelihood and method of moments estimators, are defined as roots of estimating equations. In complex models, these equations often cannot be solved explicitly; although numerical methods can provide approximate solutions, closed-form approximations can reduce computation and facilitate theoretical analysis by making an estimator's dependence on the data more transparent. In this paper, we develop a general approach to constructing such approximations and a framework for quantifying their accuracy. Our theory accommodates sequences of parameters that approach the boundary of the parameter space or diverge to infinity, and provides conditions under which a closed-form approximate estimator has the same first-order asymptotic distribution as its exact counterpart. We focus on approximations to maximum likelihood estimators, although the theory applies to general estimating equations. We apply the framework to iid models from the gamma, zeta, von Mises, and generalized Pareto families, as well as to gamma regression. Together, these examples cover scalar and vector parameters, iid and non-iid settings, and normal and nonnormal asymptotic regimes. Each example is accompanied by a simulation study, and the gamma regression example also includes a real-world data application.
Read the full abstract
Cryptography and Security
I Don't Care How You Hide It, I Care What You Do: Runtime Detection of Malicious ML Models
Pre-trained machine learning (ML) models shared through public platforms can be weaponized to execute malicious payloads during loading or inference. Existing scanners prove inadequate for detecting malicious models as they rely on static analysis and function denylists, support limited serialization formats, and miss attacks that replace denylisted functions with equivalent alternatives, exploit unsupported loading paths, or abuse legitimate framework APIs. To address this problem, we present ModelShield, the first runtime behavioral detection system for malicious ML models that attributes observed system-level behavior to specific model operations. It combines application-level instrumentation of ML framework APIs with eBPF-based syscall tracing, tracking model behavior across spawned threads and processes to separate model-triggered activity from environment noise. Attributed syscall sequences are classified using 195 detection patterns mapped to MITRE ATT&CK techniques. We implemented ModelShield to support 17 ML frameworks and 36 formats. On an independent dataset, it is the only evaluated tool to achieve zero false positives, with false negatives limited to payloads that produce no malicious runtime behavior. It detected all 29 published evasion variants spanning function substitution, loading-path obfuscation, and TensorFlow API abuse, where no existing scanner achieved complete coverage. Across 145,000 models scanned from public platforms, it captured 41 malicious models on Hugging Face, including a reverse-shell payload and a payload that performed host reconnaissance and outbound exfiltration. Among these, ModelScan, PickleScan, Fickling, and ModelAudit detected only 29%, 56%, 80%, and 93%, respectively. Despite runtime overhead, ModelShield has a median analysis time of 12.3 seconds, with 90% of scans completing within 35.5 seconds, making it practical for hub-scale screening.
Read the full abstract
Quantum Physics
Random encoders and expurgation improve 1,873 entries of the Grassl qubit code tables
I report 192 explicit binary stabilizer codes $[[n,k,d]]$ with $53\le n\le 254$ and $5\le d\le 13$ whose minimum distance exceeds the lower bound listed in Grassl's code tables at the same $(n,k)$: 161 codes by one and 31 by two. Through the standard propagation rules they raise 1,873 entries of the table. For 190 of the codes, a random Clifford encoder is followed by expurgation: every logical operator of weight below the target distance is enumerated, and a few logical operators are promoted to stabilizers so that each low-weight one anticommutes with at least one of them. The other two are obtained by puncturing a code from the table and repairing it in the same way. All 192 distances are exact: the lower bound is certified by exhaustive meet-in-the-middle enumeration of up to $4\times10^{13}$ Pauli operators, and the upper bound by an explicit logical operator of weight $d$. A first-moment count explains the results. Expurgated random codes have about eight more logical qubits than a typical random code of the same length and distance, so they beat the table wherever its best known codes are no better than random. The largest gains, up to nine logical qubits, are at $d\ge9$ and $n\gtrsim170$, where the table often gives a lower bound without an explicit code. The same count predicts improvements at 619 further pairs $(n,d)$. The search was run by AI agents. The codes, their certificates and an independent checker are public.
Read the full abstract
Probability
Entropic Martingale Optimal Transport with Wasserstein Relaxed Marginal Constraints
Martingale optimal transport (MOT) is a fundamental framework in robust finance, yet its stability under marginal perturbation no longer holds in high dimensions $d\geq 2$. In this paper, we introduce Wasserstein-relaxed entropic martingale optimal transport (WEMOT), which combines Wasserstein relaxation of the second marginal with entropy regularization. This formulation ensures stability under perturbations of the marginals, thereby facilitating numerical implementation. We establish strong duality and dual attainment for WEMOT, which leads to efficient Sinkhorn-type algorithms. We further prove a convergence rate of WEMOT to MOT as regularization parameters vanish, and also the quantitative stability of WEMOT with respect to marginal perturbations.
Read the full abstract
High Energy Astrophysical Phenomena
Neutrino pair annihilation in curved spacetime: a fast moment method and the geometry of flavour conversion
Neutrino-antineutrino annihilation above black-hole accretion discs is shaped both by general-relativistic transport and by neutrino flavour conversion. Whether these effects can be treated independently depends on how gravity weights the different flavour populations. We investigate this interplay using backward ray tracing in Schwarzschild and Kerr spacetimes, combined with an exact angular-moment formulation of the annihilation kernel that we validate against the analytic spherical solution and published disc calculations. We first establish that flavour-conversion corrections are independent of the spacetime geometry when conversion is uniform and the initial heavy-lepton neutrino emission is negligible. For uniform conversion, this separability remains approximately valid when the flavour-dependent emitting regions overlap: compared with the same discs in flat space, moving with the same local velocities, gravity enhances the deposited energy by factors of 1.4-5, while changing the relative conversion effect by no more than 6% for most temperature profiles considered; conversion that varies across the disc raises this to 22% around a rapidly spinning hole. A qualitatively different regime emerges when heavy-lepton neutrinos originate from an inner ring separated from the electron-flavour emitting regions. Relativistic bending and energy shifts preferentially enhance annihilations involving inner-ring neutrinos, producing relativistic conversion factors up to 2.4 times the flat-space ones. In one configuration, flavour equipartition reduces the deposition by 54% in flat space but increases it by 11% in Kerr spacetime. These results identify the spatial arrangement of flavour-dependent neutrinospheres as a key determinant, in the models considered, of whether flavour conversion and relativistic transport can be treated separately.
Read the full abstract
Artificial Intelligence
AlterNet: Adaptive-Depth Mixture-of-Experts with Learned Halting for Elastic Inference
Most LLM architectures have fixed compute demands determined at training time. We propose a novel architecture that combines the Mixture-of-Experts architecture with shallow experts and a novel dynamic-depth system to make the extent of a forward pass dynamically adjustable. Our approach lets the model automatically determine the compute needed for a given task, paired with a system to scale it down after training without retraining when the execution environment demands faster or cheaper inference. To achieve this, we introduce an iterative computation mechanism, sequence-level depth-based routing steered by a halting mechanism within the experts, a depth encoding that lets the model track its progress in the chain, and a regularization mechanism that incentivizes the model to use allocated compute efficiently. We present initial benchmarks and discuss future research directions.
Read the full abstract
Commutative Algebra
Uniform and Relative Non-Associativity on Minimal Free Resolutions
We study unital chain-map multiplications on minimal free resolutions of quotient rings. We prove that Katthän's example admits no associative multiplication over any field, without assuming graded commutativity or compatibility with an internal grading. The proof gives a relative obstruction in which associativity on a specified finite collection of triples forces failure at another triple. The argument extends to a family that includes nonmonomial ideals. We also give criteria for uniform non-associativity, where a fixed triple fails for every multiplication, by comparing either associators or their primitives under changes of multiplication. These criteria persist from monomial elimination ideals and give explicit obstructions for Avramov's example and the known eight-vertex Bier sphere over every field. As an application, we show that the edge ideal of the eight-cycle admits an injective Taylor comparison on five selected generators, while every selection of that size fails to admit a multiplicative comparison. Four generators admit a multiplicative embedding.
Read the full abstract
Networking
Stochastic AI Model Placement for AI-and-RAN Coexistence in 6G Networks
This paper considers AI-service placement over geographically distributed RAN sites equipped with GPU resources partitioned into fixed, isolated multi-instance GPU (MIG) profiles. We propose C$^2$AIR, a chance-constrained optimization model that maximizes expected AI requests served while protecting uncertain RAN workloads and enforcing probabilistic capacity reliability requirements for admitted AI services. C$^2$AIR allows a service to run at multiple sites and accounts for discrete profile, memory, and GPU-slice constraints. We evaluate C$^2$AIR on a four-site deployment calibrated from NVIDIA Aerial RAN capacity figures and MLPerf Inference results, and on seven larger scenarios with up to 15 sites and 40 AI services, computing every violation and overload probability exactly from the demand distributions. C$^2$AIR keeps the RAN violation probability within its target at every hour of the day, whereas reserving for the mean RAN demand violates it in 24% to 47% of the intervals. The reliability condition prevents placing AI models that would be overloaded in 85% of the intervals at an event hour, and multi-site replication serves 2.7 times more AI requests than single-site deployment. The evaluation also shows that coarse MIG instances and a heavy demand tail can make strict RAN protection expensive.
Read the full abstract
Emerging Technologies
SEU Hardened Multi-Context Spintronic FPGA
Due to susceptibility to radiation-induced soft errors and resource constraints implementing large and complex designs on field programmable gate arrays (FPGAs) in harsh environments is challenging. Therefore, to mitigate these issues, this paper proposes an innovative approach through the introduction of a single event upset (SEU)-hardened spintronic FPGA architecture that incorporates multi-context capabilities. Leveraging the non-volatile properties of magnetic tunnel junction (MTJ), the proposed design can store two/four distinct configurations internally, eliminating the dependence on external non-volatile memory which significantly boosts the system performance. Additionally, through post-layout simulation, we demonstrate our design immunity to SEU, which enhances its reliability and operational stability in environments susceptible to radiation-induced soft errors. The efficiency of the proposed design was also evaluated using the ROSSETA benchmark, which serves as a representative model for real-world FPGA applications.
Read the full abstract
Machine Learning
On Differentially Private Mechanisms for Linear Regression
We study the classic least-squares linear regression problem under differential privacy (DP) in the batch and the continual observation setting. In the latter setting, the mechanism receives a stream of $n$ data points $(x_i,y_i)$ ($x_i\in\mathbb{R}^d$ and $y_i\in\mathbb{R}$) and must output an estimate of the optimal parameters at each time step $i$. Although differentially private linear regression has been extensively studied in the batch setting, the problem remains underexplored for the continual setting, with prior work being restricted to constrained linear regression and update time polynomial in $n$. In this work, we introduce ContinualAdaSSP, the first mechanism for unconstrained linear regression under continual differential privacy. Our mechanism's suboptimality gap is proportional to $\sqrt{d}\operatorname{polylog}(n)$. It requires $O(d^2\log{n})$ space and $O(d^2\log{n}+d^{\omega})$ update time, where $d$ is the dimension of the input space. We also study fundamental limits of differentially private linear regression by showing the first lower bounds for the box-constrained least-squares linear regression problem. Our lower bounds are proportional to $\min(n,\frac{\sqrt{d}}{\log^{1.5}d})$ for $(1,O(\frac{1}{n}))$-DP, and $\min(n,\frac{d}{\epsilon})$ for $\epsilon$-DP, and are derived using a reduction from a fingerprinting code problem from [Peter et al., 2024], providing new insights into the inherent hardness of this canonical task. Our experimental evaluation on real-world datasets shows that the accuracy of our continual observation mechanism clearly outperforms any other known differentially private mechanism for this problem such as rerunning the batch algorithm at each time step.
Read the full abstract
Solar and Stellar Astrophysics
Constraining the Merger Fraction and the Upper Mass Limit of Ultramassive White Dwarfs
We performed a spectroscopic survey of all known white dwarfs with $M \geq 1.3~M_\odot$ and $T_{\rm eff}\geq11,000$ K identified through Gaia DR3 astrometry from the Gentile-Fusillo et al. (2021) catalog. Of the 293 white dwarf candidates that met our criteria, 259 had no optical spectroscopy or spectral classification available in the literature. We obtained follow-up spectroscopy on 97% of this sample and performed a detailed model atmosphere analysis to constrain the atmospheric composition and mass for each object. For a selection of magnetic white dwarfs, we use offset dipole models to constrain the magnetic field strength and geometry and calculate the magnetic fraction as a function of mass. We use magnetism, unusual atmospheric composition, kinematics, and rapid rotation rates to identify merger remnants, and constrain the merger fraction to be $53_{-6}^{+7}$% for white dwarfs more massive than $1.3~M_\odot$. Hence, roughly half of the most massive white dwarfs likely form through binary mergers. We also use our sample to derive an empirical upper mass limit for white dwarfs in the solar neighborhood. Including the General Relativistic effects, the most massive white dwarf in our sample, J0318+7923, has $M=1.33-1.35~M_\odot$ depending on the unknown core composition. Furthermore, because our temperature criteria overlap with the ZZ Ceti instability strip, we find a new selection of ultramassive pulsator candidates that may provide an excellent opportunity to probe the interiors of crystallized CO or ONe-core white dwarfs.
Read the full abstract
Functional Analysis
Extensions of operators between ordered vector spaces
We prove analogues of several classical theorems about operators between vector lattices in the context of directed Archimedean ordered vector spaces. In particular, we show that every order continuous operator between pervasive Archimedean ordered vector spaces can be extended to an order continuous operator between their Dedekind completions. Additionally, we introduce the universal completion of a directed Archimedean ordered vector space, and show that every supremum-preserving operator between directed Archimedean ordered vector spaces can be extended to an order continuous homomorphism between their universal completions. We also establish some universal properties for the relative uniform completion.
Read the full abstract
Medical Physics
Medical Radioisotopes: Supply-chain Resilience, Distributed Production, and Access
Over 50 million diagnostic procedures involving medical radioisotopes are performed per year and the number of therapeutic and theranostic agents continues to climb. Yet, the global supply chain is far from stable and many regions are underserved. This is the second of a Series of two review papers providing a comprehensive overview of the landscape of medically relevant radioisotopes. This paper focuses on the supply chain. It presents a critical overview of the 2025-2026 status quo, case studies of acute stressors and disruptions they caused, and methods in place to secure a stable supply. A literature-based analysis is paired with expert interviews. We argue that the ageing fleet of nuclear reactors currently used for the centralised production of many isotopes should be complemented by regional accelerator-based centres, alleviating single-point failures and cross-border logistics complications, and increasing access to short-lived isotopes in harder-to-reach regions and low- and middle-income countries.
Read the full abstract
Optics
Deep-UV All-dielectric qBIC Metasurface for Refractive Index Sensing
All-dielectric metasurfaces supporting quasi-bound states in the continuum (qBICs) have emerged as powerful platforms for biosensing, yet their operation remains largely concentrated in the visible and near-infrared regimes. Extending these resonances into the deep-ultraviolet (DUV) to probe the intrinsic electronic absorption of biomolecules presents substantial material and nanofabrication challenges. Here, we experimentally demonstrate a hafnium dioxide (HfO$_{2}$) metasurface supporting DUV qBICs in the wavelength range from 225 nm to 340 nm with quality factors reaching 500. We further demonstrate its refractive index sensing capabilities through integration with a microfluidic cell and optical characterization in reflection mode. We show that the introduction of an aqueous buffer leads to a further quality factor enhancement. Achieving a sensitivity of up to approximately 60 nm/RIU and an operational figure of merit of 40 $\mathrm{RIU}^{-1}$, this low-loss HfO$_{2}$ metasurface provides a robust foundation for label-free DUV biosensing.
Read the full abstract
Computer Vision
FishSense Lite: A Camera-Based, Single Laser Hardware Framework with Software System for In Situ Fish Length Measurement
Fish stock assessments need fish lengths paired with species from far more places than trained surveyors can reach. Recreational divers already visit many of those places, but measuring a fish underwater has required trained observers or specialist equipment. We present FishSense Lite, which recovers fish length from a single photograph taken with a recreational dive camera, a dive laser pointer, and a 3D-printed mount. The diver aims the laser at a fish and takes a picture; afterward, the dot's position in the image gives the range to the fish, and the range turns the fish's size in the image into a length. A citizen-science program, trained by the system's builders, taught its own divers and collected lengths paired with species for 73 wild fish across seven open-water dives. On seven wild fish also measured by an underwater stereo-imaging system, the two instruments agreed without detectable bias (mean difference +0.8%). On rigid targets of known length, FishSense Lite recovered every target to within 1.3%, and 99% of 1,055 frames fell within a 15% error budget. From the error sources we measure, we derive a field protocol: calibrate against a dive slate on every dive, photograph fish within 15 degrees of perpendicular and beyond 1 m, and take at least ten frames of each.
Read the full abstract
Computer Vision
eSharp: Diffusion-based Sharp Monocular Metric Event-to-Depth Estimation
Depth perception is fundamental for autonomous navigation, industrial inspection, and mixed-reality systems. It often operates under strict constraints on power and form factor. In such cases, monocular depth estimation (MDE) is often used. Even as recent advances continue to improve monocular depth quality, conventional frame-based MDE systems remain challenged by fast motion and high-dynamic-range scenes. This work introduces eSharp, an MDE framework that addresses these problems using an event camera. eSharp repurposes pretrained diffusion priors for event-only MDE and produces accurate metric depth with unparalleled sharpness on real data. Real-world event datasets provide only sparse LiDAR depth maps for ground-truth supervision, making dense prediction challenging. We address this limitation through cross-modal distillation, using LiDAR-calibrated dense pseudo-labels generated from a spatially co-registered frame camera to train eSharp. Experiments show that eSharp outperforms prior event-to-depth baselines in both metric-depth accuracy and spatial fidelity, establishing diffusion models as a promising foundation for event-driven 3D perception.