Fluid Dynamics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 6 November
We carry out a numerical investigation of the Oberbeck–Boussinesq equations in a fluid layer inclined against gravity. Equilibria containing different numbers of co-rotating rolls are computed. Mixed-mode branches connect branches containing states with four and six identical co-rotating rolls. These branches display two contrasting routes to reducing (or increasing) the number of rolls, one in which rolls disappear (or appear) and another route in which rolls merge (or split). These results are interpreted in light of the $(2,3)$ mode interaction scenario.
Signal Processing
Sparse Array-Inspired Frame Structure Design for 5G NR ISAC
Integrated sensing and communication (ISAC) has emerged as a promising paradigm for future wireless networks. However, the extensive occupation of time-frequency resources for sensing limits the communication capacity. To address the resource conflict, this paper proposes a novel sparse sensing frame structure aligned with the 5G new radio (5G NR) resource grid at the slot and physical resource block (PRB) levels through a predefined sensing-pattern block. Specifically, inspired by the design of two-dimensional (2D) sparse antenna arrays, we map the sparse spatial geometries onto the time-frequency resource grid. By leveraging the hole-free property of the difference coarray, we eliminate the range-velocity ambiguity and cast the sensing signals into a reconstructed higher-order tensor representation. For efficient parameter estimation, we propose a novel binary-tree-based tensor train decomposition (BT-TTD) algorithm that extracts lower-dimensional tensor train cores (TT-cores) and executes a Vandermonde-prior-guided rectified alternating least squares (RecALS) algorithm. Simulation results validate that the proposed scheme achieves competitive estimation accuracy with reduced complexity. Furthermore, we investigate the topological properties of various sparse patterns, revealing the inherent trade-off among sensing accuracy, resource occupancy, and computational complexity.
Read the full abstract
Machine Learning
Marginal-Nonuniform Multiclass Learning
We study the problem of multiclass classification within the framework of marginal-nonuniform learning, where the constants in the learning rate may depend on the unknown marginal distribution generating the unlabeled data, but not on the target concept. This notion naturally lies between classical PAC learning, where the constants are independent of both the marginal distribution and the target concept, and more recent universal learning, where the constants may depend on both. We give a complete characterization of the optimal marginal-nonuniform learning rates, establishing a fundamental trichotomy: exponential, linear, and arbitrarily slow rates. This characterization is witnessed by a new combinatorial dimension, termed the uncentered DS-eluder dimension. A key ingredient in the upper-bound proofs is a set of new distributional exhaustion lemmas that exploit the sequential nature of this dimension and may be of independent interest.
Read the full abstract
Artificial Intelligence
CID: Measuring Cognitive Work Through Semantic State Change
Traditional information theory provides powerful tools for analyzing statistical properties of information but offers limited insight into meaning and semantic transformation. It has never been possible to quantify the meaning of an arbitrary piece of information. Large language models represent information as vectors embedded within high-dimensional semantic spaces, enabling meaning to be analyzed geometrically. This paper defines semantic state as the vector representation of information within a model's semantic space and defines cognitive work as the transformation between semantic states, movement through semantic space. To quantify this transformation, we introduce the Contextual Information Differential (CID), a covariance-aware metric based on Mahalanobis distance measuring semantic displacement relative to the structure of the embedding manifold. By treating meaning as a geometric property of semantic space, CID provides a practical method for measuring semantic change, cognitive transformation, and information processing. This framework establishes a foundation for analyzing cognitive processing, learning, reasoning, knowledge generation, synthetic expertise, and human/AI collaboration through implementation-independent measurement of semantic state change in both human and artificial systems.
Read the full abstract
Instrumentation and Detectors
Portable Glass-RPC Telescope for Muography and Future Applications at Laser Wake Field Accelerators
We report a 2025 beam test at ELBA, the multi-GeV laser wakefield accelerator at ELI Beamlines, using a Resistive Plate Chamber (RPC) telescope with an active area of 16 * 16 cm2. The portable design sup- ports autonomous operation in challenging environments. Previous cosmic-ray measurements established its tracking capability for muography. During the campaign at ELBA, 30 f s long, 0.5 PW laser pulses in- teracted with a plasma target at a 0.2 Hz repetition rate, where a tungsten converter upstream of the beam dump enhanced muon production. This was the first deployment of an RPC telescope in a high-power laser environment. The detector remained operational in the presence of radiation and electromagnetic interference. Although parasitic running and limited statistics prevented definitive identification of beam- induced muons, the measurements offered insight into background activity and showed evidence of re- constructed tracks. They establish the feasibility of a dedicated campaign with controlled beam conditions and improved muon sensitivity. This contribution forms part of the International Muography Workshop 2026 proceedings.
Read the full abstract
Quantum Physics
Open Quantum Time Crystals without External Thermodynamic Resources: Go and No-Go Theorems
Time crystals are phases of matter that spontaneously break time-translation symmetry. They have been ruled out in the ground and Gibbs states of isolated systems with only short-range interactions, yet they have been proposed and observed experimentally in nonequilibrium settings such as Floquet and open systems. However, existing proposals for time crystals in open quantum systems require an external supply of thermodynamic resources, such as work or energetic coherence. Whether coupling to an environment, without any such resource supply, can induce a time crystal absent in the isolated system remains open. Here, we answer this question systematically. We first construct an explicit model and prove that coupling to an environment can activate a time crystal absent in the isolated system, even without such resource supply. We consider Lindblad master equations with (i) time-independent Hamiltonian and jump operators, (ii) time-translation covariance, and (iii) detailed balance—conditions that hold in standard derivations unless thermodynamic resources are supplied externally. Our model satisfies these conditions, and the activation occurs in a stationary state of both the isolated and open systems. We give several implementations of this dynamics with a zero-temperature Gibbs environment, a time-independent, energy-conserving interaction, and only short-range Hamiltonians, one of which exactly reproduces the time-crystalline periodicity. We then prove a no-go theorem: under (i)–(iii), activation can arise only from an environment-induced Hamiltonian term, not from dissipation. Our results show that time crystals can be activated in open systems without external resource supply and identify the physical requirements on the system and environment for this activation through our general no-go theorem.
Read the full abstract
Statistical Methodology
M-Estimation in Censored Regression Model under Endogeneity
This paper develops a general M-estimation framework for censored regression models under endogeneity. A two-stage estimation is proposed to address the issue of endogeneity in the first stage and the issue of censoring in linear regression models in the second stage. The issue of endogeneity is addressed via instrumental variables using the control function approach. The issue of censoring in linear regression models is addressed via M-estimation using a suitably defined class of loss functions yielding a class of estimators. Thus, the proposed general M-estimation framework nests several existing estimators as special cases for the respective loss functions. Moreover, this provides flexibility of accommodating a broad range of robust loss functions that have not previously been studied in the context of censored regression. We establish consistency and asymptotic normality for the proposed class of estimators. An empirical study based on simulated datasets is conducted to illustrate general M-estimation for various choices of loss functions and to examine the finite-sample performance of various resulting estimators.
Read the full abstract
Quantum Physics
Twisted Compactification, Quotient Quantum Codes and Logical Gates
Many quantum codes, including balanced-product, bivariate-bicycle and genon codes and the five-qubit code, are quotient codes: a larger code folded onto itself by identifying each check with its image under a deck transformation composed with a symmetry. We show that this is the lattice form of twisted compactification, in which a theory on $X\times F$ carries a symmetry background along $F$, and we use the correspondence in both directions. For monomial quotients of Pauli stabilizer codes, a transfer theorem gives the logical qudits, logical operators and distance bounds from those of the parent, with a Tate cohomology group counting the logical operators that are new. For higher-form and cup-product symmetries the twist is an entangling diagonal circuit; when a finite cover trivializes the background, the code is a dressed quotient, so abelian Pauli codes give non-abelian topological order such as $D(D_4)$. The dressings have Clifford level at most one less than the degree of the twist, and for cubic twists the code dimension is a count of solutions of quadratic equations in cup products, which certifies that the code is not a Pauli stabilizer code when it is not a power of two; on tori and slabs the distance is linear in the size. Twists also create logical gates: three toric codes on $S^3$ have no logical qubits, but on $\mathbb{RP}^3$ the same circuit is a logical $CCZ$. A quarter-turn twist of the 3D fermionic toric code gives codes of linear distance, from $[[768,2,9]]$ up, whose $p+ip$ pump is an explicit finite-time quasi-local evolution that preserves the code space exactly; using the spin-topological response formula of Barkeshli, Hsin and Kobayashi, we identify its logical action as $T$ up to Clifford factors. Finally, twisting abelian gauge theories reach only the quantum doubles $D(G)$ with $[G,G]$ central.
Read the full abstract
Dynamical Systems
Suppression of gel formation by spontaneous fragmentation
In this work we study the discrete Smoluchowski equation with a multiplicative kernel and an additional term describing the spontaneous shattering of each particle into monomers at a constant rate $\lambda$. We show that the introduction of fragmentation suppresses gelation for $\lambda \ge 4$. We derive analytical expressions for the second moment and the monomer concentration $n_1(t)$. These expressions lead to a recurrence relation allowing the steady-state particle size distribution for $\lambda \ge 4$ to be calculated with machine precision. We validate our analytical and semi-analytical results via numerical simulations using a solver that employs the Fast Fourier Transform for computation of the right-hand side and a second-order Runge-Kutta method with adaptive time steps.
Read the full abstract
Computer Vision
Multimodal Training with Knowledge Traction for Video Anomaly Detection
The task of weakly supervised video anomaly detection is challenging. Existing approaches mainly model video frame features and determine whether anomalies are present based on their magnitudes. In this work, we propose a novel method that incorporates external knowledge into training to improve model performance, robustness, and generalization. We leverage textual prompts and a large language model (LLM) to establish prior information for the specific task, and employ dictionary learning to normalize and compress this information. Moreover, during experiments, we observed inconsistencies in the textual descriptions extracted from similar video frames by image caption models. To address this, we devise a method that leverages an LLM and auxiliary information to ensure consistency in the textual descriptions of similar frames within a video. Finally, we introduce a feature processing approach that incorporates prior knowledge and cross-modal attention mechanisms. This approach learns the relationships and characteristics among different modalities to construct more accurate and reliable feature representations, thereby improving the accuracy and stability of downstream tasks. Experimental results demonstrate that our method surpasses other state-of-the-art approaches.
Read the full abstract
Mathematical Physics
Radiation principle for discrete acoustic equation on graphs with several cylindrical outlets to infinity
A discrete analogue of the acoustic equation with a spectral parameter is studied on graphs with several cylindrical outlets to infinity. At infinity in each cylindrical outlet the coefficients of this equation converge exponentially fast to periodic functions. It is assumed that the spectral parameter is fixed and distinct from thresholds and flat bands. We construct a special basis of the continuous spectrum eigenfunctions (bounded solutions of the homogeneous equation that do not belong to l2), introduce the scattering matrix, and establish the radiation principle (a well-posed statement of the problem with radiation conditions at infinity).
Read the full abstract
Materials Science
Graph representations of dislocation networks
The mechanical response of a crystal is governed by the evolving network of dislocations it contains, yet quantitatively linking the `state of that network' to its yield stress remains challenging. Here we represent the dislocation networks obtained from discrete dislocation dynamic simulation as graphs, using five distinct graph representations, characterize each network through a compact set of topological properties, and predict the yield stress from these graph properties and the applied strain rate using a standard random forest model evaluated with strain rate stratified cross-validation. The model predicts the yield stress with $R^2\approx0.86-0.90$ across all box sizes and $R^2\approx0.91-0.94$ for larger box sizes, and this accuracy is consistent across all five representations, indicating that the topology-stress relationship is a robust feature of the graph properties rather than an artefact of any single graph representation. We propose graph topology as a compact and robust route for connecting the microstructural state of a dislocation network to its mechanical response in addition to established variables.
Read the full abstract
Discrete Mathematics
Pangenome Peeling: Decomposing Weighted Adjacencies into Path-Forest Genomes
One aspect of pangenome analysis that has received little attention is the recovery of the individual genomes that gave rise to a pangenome graph. While substantial effort has focused on constructing pangenomes from collections of genomes, the inverse problem remains largely unexplored: given a pangenome represented as a set of weighted adjacencies, can we reconstruct the constituent linear genomes that underlie it? Thus we introduce the pangenome peeling problem, which seeks to decompose a weighted adjacency graph into a collection of linear genomes whose combined adjacencies reproduce the observed edge weights. This problem is particularly relevant when considering ancestral pangenomes, where the graph structure may be known or inferred, but the underlying genomes are not. As a first step toward addressing this problem, we formally define pangenome peeling and derive a lower bound on the minimum number of genomes required to explain a given weighted adjacency graph. We then present two reconstruction methods: one based on the sequential addition of edges and a second based on extracting maximum-spanning-tree structures. Both methods produce genome collections that exactly explain the input adjacency data. To evaluate the recovered sets of genomes, we compare them to the set of input genomes using a maximum-weight matching framework based on the double-cut-and-join distance with indels. Here the maximum-spanning tree approach is superior, producing sets of constituent genomes close to the simulations while the sequential addition of edges approach yields constituent genomes increasingly divergent from the input as genomes sizes increase.
Read the full abstract
Algebraic Geometry
Fujita's Very Ampleness Conjecture for $\mathbb{P}^{n-1}$-Bundles over Curves
Let $\mathcal{E}$ be a vector bundle of rank $n\geq 2$ over a smooth projective curve $C$ over $\mathbb{C}$. We prove that $\omega_{\mathbb{P}_C(\mathcal{E})}\otimes\mathcal{L}_1\otimes\dots\otimes\mathcal{L}_k$ is very ample for arbitrary ample line bundles $\mathcal{L}_i$ on ${\mathbb{P}_C(\mathcal{E})}$ when $k\geq n+2$.
Read the full abstract
Astrophysics of Galaxies
Violent relaxation redux
Collisionless self-gravitating systems, such as merging galaxies, have long seemed an exception to the statistical-mechanical principle that chaotic many-body systems relax to maximum-entropy states: Lynden-Bell's statistical mechanics of their "violent relaxation," constrained by exact conservation of phase-space volume – equivalently, of infinitely many collisionless "Casimir" invariants – has appeared not to be a good description. We propose that violent relaxation does give rise to Lynden-Bell's maximum-entropy equilibria, but that both are underpinned by gravitational turbulence, in which large-scale gravitational fluctuations drive chaotic phase-space mixing of the particles' distribution function. We derive the universal small-scale (sub-Jeans) spectra of this turbulence and show that it leads to phase-space-volume conservation being broken on a timescale only logarithmically longer, in the particle number N, than the dynamical (violent-relaxation) time. This timescale separation lets the system relax "adiabatically" through a sequence of Lynden-Bell equilibria, each corresponding to instantaneous values of the Casimirs. At long times, it approaches an equilibrium close to its "ground state": a distribution whose Casimirs have adjusted so that the system's energy is now its minimum possible energy. Thus, the system's universal tendency is toward a minimum-energy, maximum-entropy state. We corroborate this picture with 1D N-body simulations and hypothesize that it can explain the universal equilibria of galaxies and dark-matter halos.
Read the full abstract
Optimization and Control
Data-driven representation of continuous-time linear time-invariant behaviors via Hankel integral operators
We present a data-driven representation of the finite-horizon behavior of deterministic controllable continuous-time LTI systems. We first introduce a notion of persistency of excitation for continuous-time signals that does not require the signal to be differentiable. We then show that, when the input is persistently exciting in this sense, the closure of the range of a Hankel integral operator built from a single input-output trajectory equals the behavior of the system over a finite time horizon. This result provides a nonparametric representation of system trajectories directly in terms of measured data, without requiring an explicit parametric model of the system.
Read the full abstract
Artificial Intelligence
PreForceVision: Physics-Guided Optical Estimation of Bolt Preload from Bolt-Head Deformation
Bolt preload is a decisive functional quantity in bolted joints because it determines whether external loads are transmitted primarily through frictional clamping or directly through the bolt. Existing preload measurement techniques, including strain-gauge bolts, force washers, ultrasonic methods, and bolt-length measurements, are effective in certain applications but require additional hardware, coupling, calibration, or joint modifications. This limits their scalability for large numbers of standard fasteners and motivates a non-contact optical approach. This paper proposes PreForceVision, a physics-guided optical framework for estimating bolt preload from preload-induced surface deformation of the bolt-head front face. The method treats preload estimation as an inverse deformation-to-force problem, where deformation of the optically accessible bolt-head surface is extracted from reference-loaded image pairs and related to the internal preload state. Preliminary results for an M16 strength class 8.8 bolt show an approximately \SI{12}{\micro\meter} simulated head-face depression at \SI{90}{\kilo\newton}, strain-gauge measurements of comparable magnitude, and an optical 3D measured central depression of approximately \SIrange{10}{20}{\micro\meter}. The contribution is a physics-guided optical methodology combining controlled image-pair acquisition, DIC- or optical-flow-based deformation extraction, multimodal validation, and uncertainty-aware regression for rough preload estimation under controlled conditions.
Read the full abstract
Artificial Intelligence
StockASkill: A Local-First AgentSkill for Intelligent Stock Selection
Quantitative equity selection and portfolio construction in emerging and cross-border markets frequently suffer from computational bottlenecks, remote API fragility, survivorship bias, and rigid execution constraints such as discrete board lots and investor-mandated drawdown limits. In this paper, we present StockASkill, a local-first, full-data-driven AgentSkill engineered as a specialized Quant Co-Processor for autonomous AI agents and quantitative researchers. Tailored for agile portfolios (10^5 ~ 10^7 RMB) where discrete lot execution and capital protection are mandatory, StockASkill embeds a columnar DuckDB OLAP engine over 42.44 million historical price records and 186,432 financial disclosures spanning 26 years across China A-shares, Hong Kong, US equities, and ETFs. By eliminating ad-hoc network dependencies during analysis, StockASkill reduces cross-sectional screening latency across 5,400+ equities from over 25 seconds to sub-0.5 seconds (a 95%+ reduction). We formalize an end-to-end mathematical pipeline integrating a 7+1 factor alpha model, an SLSQP convex optimizer under smooth CVaR drawdown surrogates with exact path projection, dynamic 100-share board lot integer quantization, and an autonomous thesis lifecycle finite-state machine. Furthermore, we provide a zero-disk, strongly-typed JSON streaming interface designed for high-throughput LLM tool-calling. Open-sourced at https://github.com/axjing/stockaskill, comprehensive empirical benchmarks from 2018 to 2024 demonstrate that the architecture achieves annualized Sharpe ratios exceeding 1.48 and Calmar ratios up to 2.17, while reducing max drawdown by up to 85% relative to benchmark indices.
Read the full abstract
Computer Vision
ExQuery: Explicit Structured Query Priors for Temporal 3D Object Detection
Sparse query-based 3D detectors commonly encode heterogeneous semantic, geometric, and motion information into unified latent queries, including when explicit 3D anchors are used. We introduce ExQuery, an explicit structured query framework for temporal 3D object detection. Each decoded query is organized into a shared semantic context slice and physical-state-specific slices for center, dimensions, orientation, and velocity. This structure is preserved from prior encoding to branch-specific regression and is reused to reconstruct temporal queries: predicted physical states are transformed by ego motion, classification scores form soft semantic priors, and the reconstructed query is fused with the stored latent instance feature through gated affine modulation. Experiments on the nuScenes benchmark show that our method achieves competitive detection performance, and ablation studies validate the effectiveness of our proposed modules.
Read the full abstract
Probability
Singular Vortex in the Large Deviation of Empirical Current for Symmetric Simple Exclusion Process
We prove a large deviation estimate for singular current observables of the two-dimensional symmetric simple exclusion process. In particular, the observable has a non-standard large deviation scaling $N^2(\log N)^{-1}$ due to the presence of singular vortices, whereby trajectories rotating near the origin point yield an enhanced contribution to the current observable. The proof combines the heuristics of Bodineau-Derrida-Lebowitz 2008, the averaging method of Jara-Menezes 2018, and novel cancellation and multi-scale renormalization arguments developed by this paper.
Read the full abstract
Accelerator Physics
Secondary neutron beam production at Jefferson Lab
We investigate secondary neutron production by an 11 GeV electron beam in the Hall A beam dump at Jefferson Lab and their transport to a downstream experimental area. This secondary beam could operate opportunistically with the primary electron-scattering programme. The neutron energy spectrum extends from thermal energies to several GeVs, offering a distinctive opportunity for irradiation at energy grater than 1 GeV. We considered three configurations: the current setup, the planned BDX facility, and an optimized BDX layout with a dedicated neutron extraction pipe. We also studied how radiators made of different materials and placed upstream of the beam dump increase neutron production.
Read the full abstract
Computer Vision
RadarDINO: Distilling a Visual Foundation Model into mmWave Radar for Human Mesh Reconstruction
Millimeter-wave (mmWave) radar offers a privacy-preserving and illumination-robust alternative for human mesh reconstruction (HMR). However, high-fidelity radar HMR remains challenging. Sparse point cloud-based methods suffer from information loss and missed detections, while raw radar tensor methods like RT-Mesh lack fine-grained semantics and explicit local body modeling. In this paper, we propose RadarDINO, a radar HMR framework that distills visual semantics from a DINOv3 teacher into 3D radar representations and introduces explicit geometric supervision. DINOv3 descriptors are transferred to radar voxels via calibrated RGB-radar geometry, providing voxel- and body-part-level supervision for part-aware radar representations. We further incorporate vertex- and joint-losses to directly constrain reconstructed geometry. On the M4Human dataset, our method outperforms the SOTA baseline across all three evaluation splits.
Read the full abstract
Analysis of PDEs
The BCC lattice is the optimal 3D lattice quantizer for all $p\geq1$
For every finite $p\ge1$, the body-centered cubic lattice uniquely minimizes the $p$th moment of the Voronoi cell among three-dimensional lattices of covolume one, up to orthogonal transformations. This proves the Graf–Luschgy conjecture for three-dimensional lattice quantizers (\cite{GL}).
Read the full abstract
Artificial Intelligence
Atik: A Shared-Mesh Transformer Accelerator for RISC-V
Transformer inference repeatedly invokes dense matrix multiplication and attention, including its data-dependent softmax normalization. We present Atik, an open-source RoCC accelerator for RocketChip that executes matrix multiplication and dense attention through one shared fixed-point MAC mesh. Atik accepts and returns BF16 tensors, uses a streaming online-softmax schedule that never materializes the full score matrix, and is invoked through a small descriptor-based ABI. We evaluate three meshes cycle-by-cycle on FireSim and report standalone Sky130 physical-design area. Across 137 logged FireSim test records, all tests pass their benchmark tolerance with zero reported output mismatches. Relative to the same scalar Rocket BF16 reference, the 8x8 configuration reaches 31.5x peak matmul speedup and 225.8x peak attention speedup. Hybrid projections for the single-layer workload-derived C models, using directly measured accelerator-kernel cycles, give aggregate speedups of 44.2x, 77.1x, and 32.3x for ViT, TinyBERT, and GPT-2 prefill, respectively; these are not full-model hardware replays.
Read the full abstract
Computer Vision
Embedding Predictive Adaptation and Refinement for Multi-Modal Industrial Anomaly Detection
Industrial Anomaly Detection (IAD) has evolved from 2D image-based approaches to multi-modal methods that leverage the complementary nature of RGB images and 3D point clouds. Among these, memory bank-based methods, which identify anomalies by the discrepancy between each query patch feature and its nearest normal prototype, have proven particularly effective. However, two limitations remain: (i) employing frozen pretrained encoders yields insufficient discrimination between normal and anomalous features due to the domain gap, while existing contrastive learning-based adaptation struggles to capture the spatial relationships required for precise localization; and (ii) pixel-level anomaly maps interpolated from patch-level distance scores remain coarse, lacking fine-grained detail. To address these limitations, we propose EPAR (Embedding Predictive Adaptation and Refinement). We adapt encoders by predicting the embeddings of masked regions to capture spatial relationships, together with patch-level Variance-Invariance-Covariance (VIC) regularization, resolving the poor discriminability between normal and anomalous features caused by the domain gap. We further propose a Noise Suppression Network that refines coarse maps by learning the relational discrepancy between each patch feature and its nearest normal prototype, rather than the anomaly's appearance. This relational signal remains valid for unseen real defects, despite being learned from synthetic anomalies. EPAR demonstrates strong practical applicability, excelling in few-shot settings and improving AUPRO@1% on MVTec 3D-AD and Eyecandies by +1.8% and +1.9%, respectively.
Read the full abstract
Machine Learning
Active Learning of Robust Models
In many domains from species identification to cancer diagnosis, labelling data to train predictive models remains costly. In these settings, it is advantageous to intelligently select data points to label to achieve the maximum information gain, this is the motivation behind the study of active learning. In addition to the high costs of acquiring data in such domains, there is the added complexity of data distribution shifts that can occur between training and deployment. For example, the features of cancerous skin cells taken in one country may be different in another region of the world. These two challenges - efficient labelling of data, and learning models robust to distribution shifts have been studied separately but not together, even though these challenges often co-occur in real-life settings. We identify failure modes of standard active learning approaches when the distribution of data at deployment is different from the distribution in which the data is sampled for labelling. We propose a new strategy for efficiently learning classifiers that generalise under specific forms of distribution shift, namely subpopulation shift. Drawing on work in learning from heterogeneous data, we aim to train models that perform well across subpopulations of the data, through active learning. To handle this challenging setting, we exploit the fact that many datasets include metadata about data collection that can serve as a noisy proxy for each sample's underlying subpopulation. We demonstrate how to leverage these noisy labels during active learning selection, and demonstrate empirically that our approach improves classifier accuracy across subpopulations.
Read the full abstract
Quantum Physics
Resonance-engineered single-photon routing in an asymmetric Jaynes-Cummings cavity-waveguide system
We investigate single-photon scattering and routing in an asymmetric cavity-waveguide quantum electrodynamics system consisting of a Jaynes-Cummings cavity-emitter subsystem coupled to two one-dimensional waveguides. Using a real-space scattering approach, we derive an exact three-channel scattering matrix and analytical expressions for the channel-resolved scattering amplitudes. The scattering response is governed by the interplay of asymmetric cavity-waveguide coupling, atom-cavity detuning, coherent preparation of a single-photon superposition across multiple input channels, and intrinsic dissipation. Coupling asymmetry controls the relative weights of the cavity-mediated scattering pathways, whereas atom-cavity detuning modifies their spectral response. Coherent multichannel input states provide an additional phase-sensitive control of the output probabilities through interference. In particular, we identify an analytical condition, $J_2=2J_1$ and $\alpha_1=-\alpha_2$, under which the probability in the second output channel becomes independent of photon energy; for an equal-weight normalized input, this probability is 1/2. We further show how cavity and atomic dissipation reduce the probability remaining in the guided channels. These results provide a framework for resonance-engineered single-photon routing in asymmetric Jaynes-Cummings cavity-waveguide systems.
Read the full abstract
Hardware Architecture
Memory as a Temporal Resource: Lifetime Contracts for Multi-Retention Systems
Computers move and rewrite data whose useful lifetimes range from inner-loop iterations to years, yet memory systems usually hide retention as a fixed physical property. We argue that required lifetime should become a first-class resource across algorithms, programming systems, controllers, and devices. Building on prior lifetime-aware placement, we propose an event-based validity contract that links application completion events to retention-class selection and renewal of protection. Applications would be enabled to express semantic contracts such as “until this tile completes” and the system will map data to homogeneous arrays engineered for ephemeral, epoch, or durable retention and preserve correctness through margins, ECC, refresh, migration, or recomputation. We present a trace-based simulator and FPGA prototype that exercises class migration.
Read the full abstract
Instrumentation and Detectors
The PICO-40L detector
The PICO collaboration has designed and constructed a novel bubble chamber to search for evidence of Weakly Interacting Massive Particles, a leading dark matter candidate. The PICO-40L detector is the successor to PICO-60, and uses a fully redesigned system for pressurization of the active volume after a bubble nucleation event. Pressure in PICO-40L is transferred to the target fluid through a set of two nested synthetic quartz vessels connected by bellows below the active volume. This is in contrast to the design of earlier detectors in which the pressure was transferred through a buffer fluid atop the target fluid, within a single synthetic quartz vessel connected to a bellows above the active volume. Removing the buffer fluid removes the potential for backgrounds arising from the interface between the two fluids, and moving the bellows below the target fluid suppresses potential backgrounds from particulates originating from the bellows. The newly designed detector concept of PICO-40L is referred to as the Right-Side-Up (RSU) design and requires a dedicated thermal system to keep the lower region of the detector cold while the upper region is stably superheated. The main goal of PICO-40L is to establish the performance characteristics of the RSU detector concept, which will inform the operating configuration of the final generation PICO dark matter detector, PICO-500, which is currently under construction at SNOLAB. This paper reviews the design, installation, and performance of the PICO-40L detector.
Read the full abstract
Computation and Language
Benchmarking Speaker Embedders in the Sub-Second Regime for Real-Time Voice AI Agents
Real-time Voice AI agents must separate the active user from intelligible background speech that survives conventional noise suppression. Background sources include television audio, family members, and colleagues. Production denoisers remove non-stationary environmental noise effectively. Background speech shares the spectral characteristics of the target and passes through, causing false acceptances in downstream speaker verification (SV) and automatic speech recognition (ASR). We benchmark open-source speaker embedders TitaNet-Large, TitaNet-Small, ECAPA-TDNN, CAM++, and SpeakerNet under a strict sub-second latency budget. Evaluation combines a controlled benchmark with a weighted human-annotated production set. Our controlled benchmark is built from real calls in which a second real speaker is injected into silence intervals at modulated amplitude. We report three findings. First, denoising before embedding is generally harmful and raises equal error rate (EER) across architectures. Second, truncating audio below the full voice-activity segment degrades every model. Third, roughly half of the TitaNet embedding is redundant for this task. Trailing dimensions encode population attributes such as language and gender. Such attributes carry little discriminative value in a demographically homogeneous user base. We distill these results into deployment guidance. Verification should run on raw audio with full voice-activity chunks. Thresholds should favor precision over F1. Truncated embeddings reduce the cost of similarity search.