Peer Revenue Past citations are no guarantee of future returns

Papers

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

All of arXiv

Sunday 13 September

Audio and Speech Processing

Exploring Multimodal Turn-Taking Cues in Face-to-Face Conversation using Voice Activity Projection

Turn-taking is a fundamental component of spoken interaction, and while humans naturally rely on both verbal and non-verbal signals, dialogue systems usually depend on audio cues alone. This paper investigates whether visual features from face-to-face conversations can enhance turn-taking prediction beyond what is achievable from audio-only. We extend the Voice Activity Projection (VAP) model, a self-supervised transformer-based model for predicting future voice activity, by incorporating visual features extracted from the large-scale Meta Seamless Interaction dataset of dyadic face-to-face conversations. The visual features include gaze direction, head movement, body and hand pose, and facial action units (FAU). For incorporating the visual features, we explore concatenation, cross-attention fusion, delta features, and trainable gating mechanisms. Results show that visual information improves performance over the audio-only baseline, with FAU being significantly more informative than other feature groups. Body and gaze features nevertheless contribute complementary information, as the model combining all features performs best. Furthermore, results indicate that performance on specific tasks varies depending on whether training and test data come from improvised (acted) or naturalistic (non-acted) conversations.
Read the full abstract

All authors Willem Berner, Julio Cesar Cavalcanti, Kalle Åström, Gabriel Skantze

Statistical Mechanics

Thermodynamic Geometry of an Ideal Quon Gas

We investigate the equilibrium thermodynamics and thermodynamic Riemannian geometry of an ideal quon gas within the grand canonical ensemble. By incorporating the algebraic deformation parameter $q$, the system generalizes standard bosonic behavior while recovering the conventional ideal Bose gas in the undeformed limit. A rigorous examination of the ground-state occupation reveals two disconnected mathematical domains of the fugacity. By enforcing thermodynamic continuity, single-valuedness, and consistency with the high-temperature classical limit, we exclude the second mathematical branch and establish the lower interval as the unique physically admissible state space. This identifies the deformation parameter as an intrinsic, generalized Bose–Einstein condensation threshold. Employing the Fisher-Rao metric on the equilibrium parameter manifold, we probe the thermodynamic scalar curvature across all temperature regimes. The scalar curvature remains strictly positive throughout the physical domain, confirming that the deformation preserves an effectively attractive statistical interaction without inducing fermionic tendencies. Near the critical condensation threshold, the curvature increases sharply and exhibits a definitive divergence, providing an unambiguous geometric signature of macroscopic coherence and critical fluctuations. Below the transition temperature, the pinning of fugacity eliminates a fluctuating degree of freedom, collapsing the scalar curvature to zero.
Read the full abstract

All authors Leila Yousefzadeh, Hosein Mohammadzadeh, Habib Esmaili, Zahra Ebadi, Morteza Nattagh Najafi

Representation Theory

Breaker Chains and Splitting of Interval Submodules in Zigzag Modules

Finite-dimensional zigzag modules decompose into interval modules. For an ordinary persistence module, every fully supported interval submodule of a restriction splits as a direct summand, whereas this need not hold for a zigzag module. We study this failure of splitting. A breaker pair is a local configuration of two overlapping interval summands with a backward arrow at the left boundary of the second interval and a forward arrow at the right boundary of the first. A single breaker pair gives a local nonsplitting interval submodule, but it need not control a prescribed larger restriction. We therefore introduce breaker chains. Our main result shows that a restriction $V[p,q]$ contains a nonsplitting submodule isomorphic to $I[p,q]$ if and only if its interval decomposition contains a breaker chain spanning $[p,q]$. Equivalently, the proper interval summands $J$ for which $\operatorname{Hom}(I[p,q],J)\ne0$ have supports covering $[p,q]$. This cover formulation gives a constructive description of nonsplitting embeddings, a trichotomy for their splitting behavior, and an $O(m\log m)$ algorithm from an oriented barcode with $m$ intervals.
Read the full abstract

Machine Learning

Symmetries and Singularities

Deep neural networks are highly over-parameterized, and different parameter values represent the same predictive function. This makes their effective complexity difficult to measure using only the number of parameters or the rank of the Hessian. Singular Learning Theory addresses this issue through the local learning coefficient (LLC), which characterizes the effective complexity of a model near a given solution. Existing methods for estimating the LLC often rely on posterior sampling, which can be computationally expensive for large neural networks. This makes accurate LLC estimation difficult at scale. In this work, we use known structures in the model to simplify the analysis and make LLC estimation more tractable. Specifically, we study the LLC of a graph attention model by exploiting symmetries in both the graph structure and the attention parameters. An analytic framework through a teacher–student setting, and explicit LLC estimates after considering the symmetry–induced degeneracies are developed.
Read the full abstract

All authors Vishnu Varadarajan, Mihir More, Aritra Das, Debayan Gupta

Functional Analysis

Wold-type decompositions for norm-increasing $m$-concave weighted shifts on directed trees

In this article, we show that every norm-increasing $3$-concave weighted shift on a directed tree admits Wold-type decomposition. We then classify the $4$-isometries within a class of weighted shifts on a rootless directed tree introduced by S. Chavan and S. Trivedi, and use this classification to construct analytic norm-increasing strict $4$-isometries without the wandering subspace property. This answers a question of S. Shimorin in the negative for every $m\in \mathbb{Z}_{\geqslant 4}$.
Read the full abstract

Instrumentation and Detectors

Constraint-Traceable Liquid-Nitrogen Dewar Design for Low-Noise HTS-SQUID Magnetometry

Compact high-temperature superconducting quantum interference device (HTS-SQUID) magnetometers require a liquid-nitrogen Dewar that balances operating duration against instrument-envelope and mass constraints. Here we introduced a constraint-traceable design workflow in which body radius and neck-length fraction were scanned under fixed outer-envelope volume and geometry-based mass limits. The longest-duration feasible grid point lay adjacent to the 3.85 kg and 550 mm constraint intersection, predicting 224 h hold time from a body-only initial fill to a residual liquid depth of 10 mm. A reference Dewar was experimentally monitored for 137.3 h, consuming 1.22 L liquid-nitrogen of its initial inventory. A thermal model calibrated over the first 40 h reproduced the remaining 97.3 h. Device measurements at liquid-nitrogen temperature showed a superconducting transition and Fraunhofer-like junction response, a maximum voltage-modulation depth of 36.2 $\mu$V, and a median noise level of 41.5 fT Hz$^{-1/2}$ from 100 to 1000 Hz. These results establish a reproducible method for selecting Dewar geometry under coupled constraints and demonstrate compatibility with long-endurance and low-noise HTS-SQUID magnetometer systems.
Read the full abstract

All authors Xinmin Shi, Bingke Xiang, Lingtong Hou, Wanjuan Tang, Geming Zhang, Shiqun Liu, Ruonan Wang, Yibo Wang, Zhiqiang Cao, Xueshen Wang, Xueying Zhang, Xiaoyang Lin

Distributed Computing

MANE: A Multi-Path Adaptive Network for Edge Onloading of Deep Neural Networks

Split computing constitutes a widely used distributed inference approach, where a lightweight head model is onloaded onto the device and a heavier tail model resides on an edge server, leveraging the growing computational capabilities of modern System-on-Chips while alleviating server load. As intelligent indoor environments such as smart offices grow increasingly populated with diverse IoT devices, a single edge server must simultaneously assist multiple devices, each competing for the same shared inference resources. Without a principled mechanism to manage this shared load, the server is quickly overwhelmed, causing latency SLO violations and rendering server-assisted inference ineffective. In this work, we present MANE, a distributed inference framework that equips the server with a multi-path tail architecture, enabling a dynamic accuracy–throughput trade-off at runtime. By introducing a novel multi-path model architecture, a three-stage training scheme featuring a Joint Head Network Distillation loss and a hysteresis-based scheduler with an equitable device-fallback policy, MANE maintains over 80% SLO satisfaction rate where state-of-the-art onloading methods fail completely, while preserving accuracy 6pp higher than on-device alternatives, across up to 40 concurrent devices.
Read the full abstract

All authors Sokratis Nikolaidis, Stylianos I. Venieris, Leonidas Malachias, Iakovos S. Venieris

Analysis of PDEs

On a Multispecies Electrodiffusion Model in Higher-Dimensional Porous Media

We consider the Nernst–Planck–Darcy system describing the electrodiffusion of $N$ ionic species with arbitrary valences and different diffusivities, transported by a Darcy flow driven by the electric force, in the whole space $\mathbb{R}^d$, $d \ge 3$. We prove the existence of global solutions for initial ionic concentrations that are small in the critical Lebesgue space $L^{d/2}(\mathbb{R}^d)$. Moreover, we establish uniqueness for initial data in $L^{d+\epsilon}(\mathbb{R}^d)$ when $3 \le d \le 4$ and in $L^{2d-1}(\mathbb{R}^d)$ when $d \ge 5$, improving on results in the literature, which require Sobolev regularity of the data. The proofs rely on energy estimates exploiting cancellations of the highest-order nonlinear terms, and on the resulting dissipative structures, which yield an instantaneous gain of regularity used to control the doubly nonlinear velocity term in the uniqueness argument.
Read the full abstract

Quantum Physics

Exact Asymptotic Rates and an Exponential Strong Converse for quantum SMP and One-Way Communication

Computing many instances jointly can reduce communication per instance. We ask whether such savings occur in the simultaneous-message-passing (SMP) model and how they depend on quantum messages and shared resources. For every finite total function $f$ and fixed error $0\le\varepsilon<1$, we prove that the optimal worst-case communication per instance for $f^n$ converges to \[ D^\parallel(f)=\log|\mathrm{Row}(f)|+\log|\mathrm{Col}(f)|, \] where $\mathrm{Row}(f)$ and $\mathrm{Col}(f)$ are the sets of distinct rows and columns of the table $(f(x,y))_{x,y}$. This holds for classical and quantum SMP without shared entanglement, with or without shared randomness. Thus, transmitting row and column indices is asymptotically optimal, even with bounded error, joint computation, and quantum messages. In particular, the exponential advantage of quantum fingerprinting for a single instance of equality without shared randomness disappears in the asymptotic rate. We also prove that the same threshold governs an exponential strong converse: every fixed worst-case communication rate below it forces the probability of computing all $n$ outputs correctly to decay as $2^{-\Omega(n)}$, even on average under a fixed product input distribution. With shared randomness, the optimal expected communication rate is $(1-\varepsilon)D^\parallel(f)$. Entanglement shared between each sender and the referee halves both rates; arbitrary tripartite entanglement gives no further reduction. The results extend to a class of total relations and yield one-way characterizations with $D^\parallel(f)$ replaced by $\log|\mathrm{Row}(f)|$. Our proof reduces SMP to one-way communication and relates successful computation to sampling based on quantum min-entropy; we also give a simpler proof of the key lemma using the pretty-good measurement.
Read the full abstract

Computer Vision

Compositional SVG Generation via VLM-Driven Hierarchical Semantic Parsing

While Vision-Language Models (VLMs) excel at visual reasoning, generating structured, editable Scalable Vector Graphics (SVG) remains a fundamental challenge. Existing pipelines predominantly yield flat, semantically agnostic collections of paths, where editing a single object requires manually identifying its constituent paths. To address this, we propose a VLM-driven agentic framework for semantic compositional SVG generation. Our pipeline recursively parses visual scenes into semantic and geometric hierarchies via top-down decomposition, visual grounding, and prompt-driven amodal occlusion recovery, ensuring each component is geometrically complete. Furthermore, we introduce the Semantic SVG Benchmark with human-annotated semantic groups and novel sub-component metrics (Semantic Recall/Precision, PERE) to explicitly evaluate structural compositionality and functional editability. Experiments show that our natively predicted structures surpass the upper bounds of existing flat-generation methods in both grouping quality and editability, while maintaining state-of-the-art visual fidelity.
Read the full abstract

All authors Sehwan Park, Taehoon Kim, Geonhee Han, Dohyun Kim, Seung Wook Kim, Paul Hongsuck Seo

Dynamical Systems

The conjugacy and flip conjugacy problem for Cantor minimal systems

We prove that topological conjugacy and flip conjugacy of minimal homeomorphisms of a Cantor space are complete analytic relations and hence are not Borel. We also prove that mutual reducibility by injective continuous graph homomorphisms and topological graph isomorphism are complete analytic relations on the space of nonempty compact graphs on a fixed Cantor vertex space with continuous chromatic number two. Both relations remain complete analytic on the larger space of nonempty compact graphs on the same Cantor vertex space with continuous chromatic number two or three.
Read the full abstract

All authors Konrad Deka, Kosma Kasprzak, Philipp Kunde, Dominik Kwietniak, Felipe García-Ramos

Combinatorics

Oriented Paths with Few Direction Flips Are Tournament Anti-Sidorenko

An oriented graph $H$ is said to be tournament anti-Sidorenko (TAS) if a uniformly random tournament asymptotically maximizes the homomorphism density of $H$ among all tournaments. For an oriented path $P$, a direction flip is a non-leaf source or sink. Sah, Sawhney and Zhao proved that consistently directed paths (paths with no direction flips) are TAS. He, Mani, Nie, Tung and Wei proved that for $3\le k\le 7$, every oriented path of length $k$ with exactly one direction flip is TAS, and Chen, Clemen and Noel recently extended this to every $k\ge 3$. In this paper, we extend these results by proving that for every integer $r\ge 0$, every oriented path of length at least $1665r+1454$ with $r$ direction flips is tournament anti-Sidorenko.
Read the full abstract

Computer Vision

Evaluating Contextual Bias in CNN Image Classification: Evidence from Agricultural Benchmark Datasets

Convolutional neural networks (CNNs) are typically evaluated using held-out classification accuracy, an approach that presupposes predictions are based primarily on the intended object of interest rather than incidental surrounding context. We test this assumption in CNN-based agricultural image classification by comparing model performance on original images with performance on background-dominated patches extracted from the same images across eight publicly available agricultural benchmark datasets and four widely used CNN architectures. Background-dominated patches were classified above dataset-specific random chance for six of the eight datasets, and substantially above chance for four of them, indicating that contextual information contributes to model predictions for the majority of datasets evaluated. For these four datasets, we further evaluated whether this behavior reflected genuine class-discriminative information or was primarily attributable to class imbalance using macro-averaged precision, recall, and F1 together with class-balanced test subsets. The results show that contextual reliance does not admit a single explanation: class imbalance accounts for a substantial portion of the observed signal for some datasets and architectures, whereas above-chance contextual classification persists after balancing for others. Together with previous evidence from curated object recognition and cancer pathology imaging, these findings support the growing view that contextual bias is a recurring characteristic of CNN-based image classification rather than a phenomenon confined to a single application domain. More broadly, this work provides a systematic framework for quantifying contextual bias across heterogeneous image datasets by combining dataset-specific random-chance baselines, contextual bias categorization, macro-averaged evaluation, and class-balanced robustness analysis.
Read the full abstract

All authors Abhilekha Dalal, Michael Okonoda, Eder Martinez, Lior Shamir

Differential Geometry

Singularities of the Wigner Caustic and the Centre Symmetry Set of Frontal Curves in the Euclidean Plane

Motivated by the role of the Wigner caustic in semiclassical phase-space analysis, we extend it, together with the centre symmetry set, from regular planar curves to cooriented frontals. For an angularly regular parallel pair, the signed speed of the Wigner caustic is one half of the difference of the extended signed radii of curvature, while singular points of the finite centre symmetry set are the critical points of their projective ratio. These formulas yield criteria and explicit invariants for ordinary and higher cusps, an order-lowering relation between the two constructions, and a multiplicity-weighted extension of the classical cusp-count inequality for strictly convex ovals. We also analyse parallel pairs containing a singular frontal which is not a front. A $5/2$-cusp produces either a $5/2$- or, at a signed-radius resonance, a $5/3$-cusp on the Wigner caustic. The opposite resonance sends the centre symmetry set to infinity. We obtain the corresponding transfer results for a $5/3$-cusp and give a projective completion which resolves vanishing denominators and simultaneous finite-order zeros of the two signed radii.
Read the full abstract

Databases

Natural Language Knowledge Graph Query Execution: Leveraging Controlled Semantics in the LLM Context Window

Large Language Model (LLM) applications often transfer domain concepts into the model's context informally, through prompt prose, schema dumps, and examples. We show that for database queries, data model concepts pass to LLMs more effectively through representations whose vocabulary terms carry declared, machine-readable semantics (controlled semantics). NLKGQ is a working system and reusable framework that does this for data modeled in a knowledge graph. A formal OWL ontology serves as the transfer mechanism, concentrating the meaning of the data into semantically precise tokens the model can use directly. In a single LLM call, NLKGQ places in the context a system prompt instructing on SPARQL, the complete domain OWL ontology, and a domain-specific prompt addition, together with the user's natural language query. The model then generates the SPARQL query directly, zero-shot. Where the native vocabulary of an existing database or federation of databases is opaque, a wrapper ontology substitutes clean terms and a runtime rewriter restores the native forms. Evaluating on DBLP-QuAD 2.0 showed that its scores depend on the graph snapshot, the endpoint used, and the wording of its machine-generated questions, so we propose DBLP-QuAD 3.1, which maintains the intent of 2.0 while making reference results deterministic, revising reference SPARQL where needed, and rewriting the natural language questions, with a frontier model, to state each reference query's intent clearly and completely. We evaluate on the DBLP-QuAD 2.0 benchmark (57.6% Match under deterministic re-scoring), DBLP-QuAD 3.1 (89.9% Match on 1,000 questions), SemOpenAlex (98% Match against a published baseline's 86% on the identical test set), and neuroimaging metadata (100%).
Read the full abstract

Optimization and Control

Convergence of the Sinkhorn Riemannian metric for finitely supported measures

The Sinkhorn Riemannian metric characterizes the local behavior of the Sinkhorn divergence, a debiased version of entropy-regularized optimal transport. Although the Sinkhorn divergence is known to approximate the squared Wasserstein-2 distance as the regularization parameter $\varepsilon$ goes to $0$, the behavior of the Sinkhorn Riemannian metric as $\varepsilon\to 0$ remains largely open. To the best of our knowledge, the only existing argument is for measures with a density and is still a formal computation. In this work, we prove the convergence of the Sinkhorn Riemannian metric as $\varepsilon\to 0$ in the case of finitely supported measures undergoing horizontal perturbations. Under our assumptions, our techniques additionally allow us to prove the convergence of higher order derivatives of the Sinkhorn Riemannian metric, thereby ensuring the convergence of the associated Riemann curvature tensor and Christoffel symbols as $\varepsilon\to 0$. We also compare with the absolutely continuous setting and show where our approach fails for measures with a density.
Read the full abstract

General Relativity and Quantum Cosmology

Self-Gravitation of Mode Quanta in a Causal Resonator: One-Loop Finiteness in Linearized Quantum Gravity and the Emergence of the Dark-Energy Scale

Perturbative quantum gravity is ultraviolet divergent and, as shown by 't Hooft and Veltman, non-renormalizable. A recent object-relative, Lorentz-invariant weighting of internal electromagnetic modes renders selected one-loop contributions of quantum electrodynamics finite without counterterms. Here that weighting is derived rather than postulated: causality defines a mode resonator bounded by the Hubble radius and re-defined in every inertial frame, and the self-gravitation of each mode quantum deforms its spectrum. Applied to linearized gravity, the graviton self-coupling reaches order unity at the Planck wave number, so that the mode content available to a loop is suppressed beyond it. Any suppression exceeding a threshold fixed by the diagram renders the one-loop integrals finite; no cutoff is imposed and no mode structure above the Planck scale need be derived. For pure gravity the same power counting extends to every loop order, and the leading suppression already meets the resulting condition for every topology with more than one vertex; a single class of single-vertex insertions is excluded. The Bianchi identity takes the role of the Ward identity at one loop, and the transformation between the resonators of two inertial frames is unitary. The same finite mode content yields a vacuum energy density of order H^2/G, of the observed sign and magnitude, from the Planck and causal infrared wave numbers alone. A contribution tracking the instantaneous expansion rate is an exact rescaling of the gravitational constant and is excluded by primordial nucleosynthesis, so the mode basis must be frozen in any inertial system; referring it to the stationary de Sitter horizon fixes the remaining coefficient to the geometric value 6/pi and makes a value for the energy density of gravitational vacuum fluctuations possible which can be interpreted as origin of the dark-energy effect.
Read the full abstract

Cryptography and Security

CIG-MIA: Context-Induced Information Gain Membership Inference Attacks against Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) systems ground large language models on external knowledge bases, enabling access to private, domain-specific, and up-to-date knowledge without retraining. However, the same retrieval interface can expose whether a candidate document is contained in the knowledge base. This paper studies knowledge base membership inference against RAG systems under both gray-box and text-only black-box access. Existing RAG membership inference attacks rely on signals such as direct membership prompts, response similarity, mask recovery, or query perturbation, which can be sensitive to prompt defenses, semantically related retrieved documents, and the generator's parametric knowledge. We introduce CIG-MIA, a membership inference attack based on context-induced information gain. The key insight is that explicit candidate-document injection affects members and non-members differently: if a document is already available through retrieval, injection provides little additional support for document-derived answers; if it is absent, injection introduces new evidence and yields a larger likelihood gain. In the gray-box setting, CIG-MIA computes this gain directly from token-level likelihoods. In the black-box setting, it estimates the same gain from generated text by scoring selected answer tokens with a lightweight surrogate-based estimator using semantic similarity and exact-match features. We evaluate CIG-MIA on Natural Questions, MS-MARCO, and HealthCareMagic against recent RAG membership inference baselines. On Natural Questions, CIG-MIA achieves an AUC of 0.99 in the gray-box setting and 0.93 in the black-box setting. We further analyze the information-gain signal, RAG configuration effects, ablations, and robustness to paraphrasing and generation randomness.
Read the full abstract

All authors Tan Xue, Huo Chang, Wang Changhui, Yang Sanrui, Li Heng, Chen Ping, Luo Xiapu

Computation and Language

Optimizing Sparse Outcomes Through Dense Behavioral Signals via Value-Guided Preference Distillation

Aligning multi-turn dialogue agents is usually framed as matching turn-level human preferences, yet direct optimization of long-term outcomes is often ineffective and prone to reward hacking. We formulate long-horizon dialogue optimization as a multi-objective reinforcement learning problem and train a multi-head value model that predicts a vector of observed user behaviors across multiple look-ahead horizons. Our findings demonstrate that a scalarized composite of dense auxiliary behavioral signals enables effective credit assignment and optimization of sparse outcomes. However, optimizing unconstrained single-objective proxies might induce policy degradations that are harmful when the agent is exposed to real users. To identify these failure modes prior to deployment, we establish a safety framework combining counterfactual user simulation with a validated dialogue-level outcome model to evaluate preference weightings and policy optimization methods. Finally, we demonstrate that distilling multi-objective value preferences into the policy via reference-anchored preference optimization matches on-policy online RL at a small fraction of its compute budget. Live A/B testing confirms that our distilled policy significantly improves long-term user retention, while simultaneously enhancing the positive behaviors and therapeutic-process markers.
Read the full abstract

All authors Ziyi Zhu, Daniel R. Cahn, Thomas D. Hull, Caitlin A. Stamatis, Olivier Tieleman, Guilherme B. Freire, Jinghong Chen

Robotics

Skill Composition for Legged Robot Reinforcement Learning

Robots, and humanoid robots in particular, are increasingly competent at individual behaviors, each obtained by training a specialized controller. A specialized skill is quick to train, converges reliably because the problem it faces is narrow, and can be validated on its own, none of which is true of a single end-to-end policy asked to cover everything. What remains fragile is the transition between them. We argue that the composition of independent sub-policies deserves to be treated as a research problem in its own right, rather than as an implementation detail left to whatever mechanism happens to be at hand. Reliable composition is what turns a collection of separate skills into a repertoire that can be used, extended and shared. More fundamentally, if control can be passed between specialized policies safely, and at any moment, the choice of what the robot should do next can be delegated to a component of an entirely different nature, such as a planner, an automaton or a symbolic controller, whose behavior can be inspected in advance. The policies would then only ever have to act, and what the robot can be trusted to do would become verifiable.
Read the full abstract

All authors Daniel Gigliotti, Flavio Maiorana, Fabio Patrizi, Luca Iocchi

Computation and Language

CompCQR: Compositional Query Generation for Training-Free Conversational Search

Multi-turn interactions with LLMs are becoming increasingly common in information-seeking scenarios. However, user queries are often ambiguous and context-dependent, making them ill-suited for direct use as retriever queries. Conversational query reformulation (CQR) addresses this issue by rewriting the current utterance into a stand-alone query grounded in the dialogue history. Recent LLM-based CQR approaches achieve strong performance; however, their repeated LLM invocations and misalignment with downstream retrievers remain challenges. In this work, we begin from the observation that retrievers are highly sensitive to content ordering: simply reordering the same content can lead to changes in retrieval coverage and performance. Based on this, we propose a novel training-free method that generates a very large number of queries with minimal LLM usage by compositionally combining a small set of atomic components. We further apply LLM reasoning to construct a high-quality document set that balances precision and recall while capturing the user's core intent. Our framework generalizes across both open- and closed-source LLMs as well as dense and sparse retrievers. It achieves strong performance on four widely used conversational benchmarks, with up to 22.5% relative MRR improvement over the previous state-of-the-art baseline with far fewer LLM calls.
Read the full abstract

All authors Yunah Jang, Kang-il Lee, Joongbo Shin, Kyomin Jung

Metric Geometry

A short polyhedral proof of the infinitesimal Stoker conjecture

The infinitesimal Stoker conjecture states that an infinitesimal deformation of a convex polytope that preserves all dihedral angles also preserves all normal directions up to isometry. We give a short polyhedral proof, self-contained up to assuming spectral and geometric properties of the Izmestiev matrix.
Read the full abstract

Quantum Physics

Collective-dissipation-controlled thermal rectification and entropy production in a three-terminal three-qubit XXZ spin chain

We investigate thermal rectification and its thermodynamic cost in a three-qubit XXZ spin chain coupled to three thermal reservoirs: two local baths attached to the boundary qubits and a collective bath jointly coupled to the middle and right qubits. This asymmetric dissipative architecture allows the collective environment to actively reshape energy transport. We show that the heat current carried by the collective bath changes sign as its temperature is varied, and that the spin-chain anisotropy provides an effective internal control parameter for suppressing the boundary heat currents. These features yield thermal rectification coefficients of about $3.36\%$ at maximum bias and up to $20.1\%$ in the strongly anisotropic regime, although the latter comes at the cost of severely reduced heat currents. Beyond rectification, we evaluate the steady-state entropy production rate and identify a low-irreversibility operating window near the collective-current reversal, whereas increasing the local boundary couplings monotonically raises the thermodynamic cost. Our results establish collective dissipation as a tunable control mechanism for quantum heat transport and uncover a nontrivial trade-off among rectification, heat-current magnitude, and irreversible losses.
Read the full abstract

Hardware Architecture

BigMoMo: Efficient Inference of Large-Scale MoE with Speculative Decoding on Mobile Devices

Mixture-of-Experts (MoE) models expand language model capacity on smartphones, but expert offloading remains constrained by limited DRAM capacity and costly data movement. Sequential token routing couples expert execution to fragmented flash reads and multistage NPU preparation, leaving sparse computation stalled on weight transfers. Each transfer serves few tokens before execution moves on. We exploit the multi-token verification window of speculative decoding to decouple expert movement from single-token execution, enabling weight reuse, contiguous flash reads, and load-compute overlap. We present BigMoMo, a mobile MoE runtime that exploits this window across the memory hierarchy. It prunes speculative branches and expert activations using acceptance rates, routing impact, and movement cost; reorganizes on-flash experts according to runtime co-loading patterns; and batches ready experts to overlap NPU computation with pending transfers. Across four MoE models and five benchmarks on two mobile platforms, BigMoMo achieves mean decoding speedups of $4.83\times$ over on-demand autoregressive offloading and $1.82\times$ over the best speculative MoE baseline, supporting MoE models up to 80B parameter.
Read the full abstract

All authors Maoliang Li, Hailong Zou, Taohong Han, Haoze Chi, Jiayu Chen, Zihao Zheng, Zimo Guo, Jie Zhang, Guojie Luo, Xiang Chen

Artificial Intelligence

Diffusion-Based Generation of Gait Trajectories

Generation of musculoskeletal gait trajectories conditioned on patient-specific parameters remains a key challenge for wearable robotics and rehabilitation. Assistive systems such as lower-limb exoskeletons require reference trajectories that adapt to individual morphology and therapeutic goals while preserving biomechanical realism. Traditional approaches rely on hand-crafted gait templates or optimization procedures that scale poorly across subjects and walking conditions. In this work, we explore conditional diffusion models for generating lower-limb joint-angle trajectories conditioned on gait parameters such as step length. We compare a baseline transformer diffusion model with a controllable diffusion transformer variant incorporating adaptive normalization and classifier-free guidance. Experiments on a dataset of 4,590 gait cycles show that diffusion models can generate realistic periodic gait trajectories while enabling some controllability variation in gait characteristics, highlighting their potential for personalized gait synthesis in assistive robotics.
Read the full abstract

All authors Damian Benasco, Juan Carballeira-Lopez, Jaime Ramos-Rojas, Julio S. Lora-Millan, Antonio J. Del-Ama, David Rodriguez-Cianca, Pablo Lanillos

Functional Analysis

Sphere version of Banach–Kadec–Paley theorem

We study almost Lipschitz embeddings in the quantitative geometry of unit spheres of classical \(L_p\)-spaces. Our main result is a sphere version of the classical Banach–Kadec–Paley theorem: for \(1\le p,q<\infty\), the unit sphere \(S_{L_q}\) admits an almost Lipschitz embedding into \(L_p\) if and only if \(L_q\) admits a linear embedding into \(L_p\). Consequently, the almost Lipschitz geometry of \(L_p\)-spheres completely determines the underlying linear structure of \(L_p\)-spaces. The proof relies on quantitative estimates for moduli of continuity of sphere mappings. These estimates are derived from Kalton–Randrianarivony's concentration inequalities, Mendel–Naor's metric cotype inequality, and Naor's sharp metric \(X_p\) inequality. They also yield several sharp results on the Hölder geometry of \(L_p\)-spheres.
Read the full abstract

Distributed Computing

PaxosLease Revisited: A Checked Model of Diskless Distributed Leases

PaxosLease is a protocol by which a quorum of acceptors grants time-bounded exclusive ownership with no durable acceptor lease state and no disk write on the lease acquisition path. This paper gives a precise, machine-checked statement of the protocol and of its standard use, electing a Multi-Paxos leader. Formalizing and model checking the original protocol changes three rules of its acceptor: two are required for safety, the third allows shorter restart quarantines. The protocol is formalized in TLA+ and checked by TLC, its timing arithmetic is proved in TLAPS, and an executable Python model demonstrates the distributed algorithm for human readers.
Read the full abstract

Human-Computer Interaction

Understanding the Design Taxonomy of AI-Mediated Interpersonal Communication Experiences in HCI: A Scoping Analysis

Interpersonal communication is a fundamental aspect of everyday life, shaping interactions across workplaces, education, entertainment, healthcare, and beyond. While computer-mediated communication has been extensively studied, a comprehensive understanding of AI-Mediated Interpersonal Communication (AIMIC) remains lacking. An in-depth scoping analysis is urgently needed to understand the research landscape of AIMIC in HCI, particularly following the recent growth of large foundation models, and AI agent research. We conducted a scoping analysis to understand AIMIC by performing an in-depth review of prior HCI literature published over the past decade (January, 2016 - May, 2026). Grounded in the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) approach, we curated 52 full-paper publications from the HCI literature spanning a range of interpersonal communication contexts. We analyzed this corpus by examining the types of AIMIC studied, AI integration approaches and human-AI interaction design, reported outcomes and benefits, and key challenges and future research opportunities.
Read the full abstract

All authors Chen Chen, Lingyao Li, Renkai Ma, Rawan Alghofaili, Shaoze Zhou, Bojun Zhang, Xian Su, Weidong Zhu, Christine Lisetti, Mo Sha

Human-Computer Interaction

Investigating the Impacts of Generative AI on Information Seeking

This paper is an encore submission of our 2026 journal article "Expertise and Information Seeking in the Age of Generative AI: New Procedures, New Problematics" with an extended discussion for the CSCW 2026 "Broader Impacts of GenAI in Communication" Workshop on October 10, 2026. In the original article, we employ procedural rhetoric to analyze how generative AI chatbots leverage natural language signifiers of expertise and intelligence to influence users' perception of their trustworthiness. In this submission, we extend our conversation in the CSCW community with the goal of cultivating a cross-disciplinary vocabulary for describing, analyzing, and mitigating the risks posed by the integration of generative AI into human communication practices. It is important to develop an understanding of how the procedures surrounding information-seeking practices are informed by users' values, experiences, and expectations - and how these procedures might in future be altered by the emerging turn toward AI "experts" and authority.
Read the full abstract

Artificial Intelligence

DynSTEER: Dynamic Stage-wise Trajectory Evaluation and Execution-time Review for Agents

Large language model agents are increasingly deployed for long-horizon task execution, raising a central granularity question for trajectory evaluation: whole-trajectory verification is too coarse to capture concrete failures and their associated evidence in long trajectories, while atomic-step scoring is too fine-grained, noise-sensitive, and computationally expensive. This granularity gap makes a single-reference trajectory paradigm inadequate for assessing the rich space of valid agent execution paths and delays timely feedback and early stopping in long-horizon tasks. To address these issues, we propose DynSTEER, a dynamic stage-wise framework for agent trajectory evaluation. DynSTEER bridges the granularity gap through stage-wise dynamic evaluation that segments rollouts at key execution nodes and adapts its multi-level review strategy based on stage-level results; it compiles a path-tolerant milestone graph from available task inputs to preserve diverse legal paths without reference leakage; and it supports terminating unrecoverable agent executions to curb resource waste. Experimental results show that DynSTEER improves evaluation discriminability by over 85% compared with whole-trajectory evaluation and saves 17.74% of execution steps. The code is available at https://github.com/zhichao-stone/DynSTEER
Read the full abstract

All authors Zhichao Shi, Xuhui Jiang, Wenjie Zhang, Xiaojun Wu, Cehao Yang, Chengjin Xu, Jian Guo, Yuanzhuo Wang