Papers

In general relativity, the geometric optics approximation describes the propagation of high-frequency electromagnetic waves along null geodesics. Beyond this approximation, localized wave packets acquire frequency- and polarization-dependent corrections to null geodesic motion, described by the gravitational spin Hall equations. In this paper, these equations are integrated analytically, to first order in wavelength, in a large class of type D black hole spacetimes, whenever the corresponding null geodesics admit analytic solutions. The same framework yields analytic expressions for the shape of black hole shadows with polarization-dependent corrections.

Nebular emission lines and stellar continua probe the conditions of $\mathrm{H\ II}$ regions. The optical emission line $\mathrm{[O\ III]}\lambda 5007$ is important because it is observable across a wide redshift range, enabling studies of interstellar medium (ISM) conditions across cosmic time. We investigate the equivalent width of $\mathrm{[O\ III]}\lambda 5007$ (EWO3) as an ionization-parameter diagnostic, providing an alternative to the $\mathrm{[O\ III]} \lambda 5007 / \mathrm{[O\ II]} \lambda 3727$ (O32) ratio, which requires broad wavelength coverage and dust corrections that may not be available at high redshift. We employ the MAPPINGS photoionization code with input spectra from BPASS, comparing single-star and binary-inclusive stellar populations. We find that binary-inclusive populations sustain harder ionizing spectra and strong $\mathrm{[O\ III]}\lambda 5007$ emission to later ages than single-star populations. EWO3 is highly sensitive to the ionization parameter and exhibits systematic dependencies on metallicity, gas pressure, and stellar population age. While these dependencies broadly follow those of O32, EWO3 is less sensitive to gas pressure in the low-metallicity, high-ionization regime, making EWO3 well suited for studies of metal-poor and high-redshift galaxies. Comparing the models with observations, we find that local galaxies are reproduced by lower-pressure models, whereas intermediate- and high-redshift galaxies require higher gas pressures, indicating evolving ISM conditions across cosmic time. We present metallicity- and pressure-dependent calibrations of EWO3 as an ionization-parameter diagnostic. When validated against ionization parameters derived from O32, the $\log(P/k)=6$ calibration provides the best agreement for local galaxies, while the $\log(P/k)=7$ calibration best reproduces intermediate- and high-redshift galaxies.

Although generalized hyperbolicity has proved to be a fundamental notion in linear dynamics, recent results show that it is too restrictive to characterize the shadowing property of invertible bounded linear operators on Banach spaces. This motivates the introduction of generalized pseudo-hyperbolicity, a weaker notion formulated through continuous homogeneous maps generating exponentially decaying Green families. We prove that, for arbitrary sequences of invertible bounded linear operators on Banach spaces, generalized pseudo-hyperbolicity is equivalent to both shadowing and topological stability. In the autonomous setting, this yields the equivalence between shadowing and topological stability, removing additional assumptions from previous results. We also prove robustness of shadowing and topological stability for sequences of invertible bounded linear operators.

Vision-language-action (VLA) policies infer grasp actions primarily from visual observations and robot state, but do not explicitly represent the physical response observed after contact. We present MIM-VLA, a motor-feedback-based architecture that encodes recent gripper current, position, velocity, and signal validity as a 128-dimensional interaction token. A motor-only Motor Interaction Module (MIM) is pretrained with human-reviewed contact and interaction-phase labels and then conditions only the gripper-action pathway of SmolVLA; arm actions and the position-control interface remain unchanged. The same token supports the MEM selector VLM that compares candidate interactions and produces evidence-conditioned selections and explanations. We evaluate MIM-VLA in three real-world settings: comparing the interaction resistance of visually different objects, disambiguating visually similar real and replica objects through active probing, and gently grasping fragile objects, including held-out instances. Across 13 object pairs, MIM-VLA selects the higher-resistance object in 75.0% of trials, compared with 48.8% for the SmolVLA baseline. For the evaluated tasks, the approach uses motor feedback already available from the gripper and does not require an additional tactile array, force-torque sensor, calibrated force estimate, or direct current control.

Let $f: \mathbb{N} \rightarrow (0,\infty)$ be a multiplicative function, and let $\omega > 0$. We consider the frequency with which patterns of the form $$ f(n) > \omega \max\{f(n-1),f(n+1)\}, \quad f(n) < \frac{1}{\omega}\min\{f(n-1),f(n+1)\} $$ occur in the sequence $(f(n))_n$, and give necessary conditions on $f$ so that the set of occurrences of at least one of these patterns has logarithmic density $0$. This includes a complete classification of all such completely multiplicative functions. As a particular application, we characterise (up to prime powers $p^k$, $k \geq 2$) all multiplicative functions $f: \mathbb{N} \rightarrow \mathbb{N}$ that have either finitely many local maxima, or finitely many local minima. This answers, in a strong form, two conjectures of A. Sárközy from 2001. \\ The proof is based on a natural “pretentiousness” dichotomy for additive functions. Key tools in the analysis include an analogue of the Matomäki-Radziwiłłtheorem for additive functions, due to the author, as well as a new bound for the concentration of the gaps $g(n+1)-g(n)$ of an additive function $g$, which may be of independent interest.

We investigate the statistical mechanics of a system coupled to multiple heat reservoirs at different temperatures, a setting which typically sustains nontrivial circulation. By applying a canonical transformation, we map the system onto that of a particle coupled to a single reservoir yet driven by a deterministic curl-force—a reversible, velocity-independent non-conservative force with non-vanishing curl—that can be described by a Hamiltonian with an anisotropic kinetic energy term. Despite the Hamiltonian structure, the system does not relax to the corresponding Boltzmann distribution, and is genuinely out of equilibrium. The mapping provides a mechanical interpretation of the currents in multi-temperature systems, as we demonstrate by means of two examples. In particular, the direction of circulation is qualitatively predictable from the deterministic force field. For the exactly solvable case of an $N$-dimensional underdamped Ornstein-Uhlenbeck process, we derive explicit expressions for the steady-state distribution and the entropy production rate, highlighting the role of the (generalized) curl of the force field in driving the system out of thermal equilibrium. Finally, we show how a non-conservative force can be used to exactly equilibrate any system coupled to multiple reservoirs, and obtain the corresponding equilibrium distribution.

In this paper, we construct the Mehler-Fock transform of arbitrary order MF(mu,k) by removing the restriction Re(mu) > -1/2 with the help of a positive integer k such that Re(mu+k) > -1/2 and obtain some transformation formulae based on the differential operator M(mu)N(mu). Furthermore, we define the Mehler-Fock translation and convolution operators of arbitrary order and demonstrate that the factorization property for the convolution holds in the test function space H(mu).

A stable grasp does not guarantee kinematically feasible robotic insertion because the object-in-gripper transform may force the robot towards singularities or joint limits along the prescribed insertion path. We study post-grasp kinematic feasibility repair through object-in-gripper reorientation. Given an achieved grasp and a fixed insertion path, we seek a small reorientation that restores kinematic feasibility. Sequential IK can miss such candidates by following an unfavorable joint-space path, while the nonsmooth feasibility landscape makes the search computationally expensive. We evaluate candidates using a branch-aware IK graph that maximizes the minimum feasibility margin over the discretized insertion path and use a learned task-conditioned prior to improve query ordering. The selected reorientation is executed through tactile-based extrinsic manipulation. In UR5e simulations, the planner without learned ranking reduces mean reorientation over successful trials by 51.5% compared with grid-based sequential IK. Adding learned ranking reduces this planner's mean planning time by an additional 51.1%. Real-robot experiments validate the complete pipeline.

We establish a polynomial sample complexity separation between symmetry-aware and symmetry-agnostic feature learning. We study growing-rank multi-index models with high-dimensional Gaussian covariates in $\mathbb{R}^d$ and $r=\Theta(d^\delta)$ teacher directions forming a cyclic symmetry orbit, where $0<\delta<1/2$. We compare three ways of exploiting this structure: architectural weight sharing, data augmentation over the full symmetry group, and learning without access to the symmetry. In particular, we analyze a symmetry-tied convolutional network, an untied network, and the same untied network trained with full-group data augmentation, using spherical online SGD with correlation loss. For a class of polynomial links with information exponent $p\ge3$, we prove matching sample complexity bounds up to logarithmic factors: the tied and augmented learners achieve weak directional recovery in $\widetilde{\Theta}(d^{p-1})$ samples, whereas the symmetry-agnostic learner requires $\widetilde{\Theta}(rd^{p-1})$. For the pure quadratic Hermite link, the same separation holds for weak recovery of the teacher subspace, with sample complexities $\widetilde{\Theta}(d)$ and $\widetilde{\Theta}(rd)$, respectively. Thus, full-group data augmentation matches the sample efficiency of architectural weight sharing, and both provide a polynomial advantage over training without symmetry. For $p\ge3$, the proof reveals a two-stage mechanism: fluctuations at initialization select one direction in the teacher orbit, after which localized growth amplifies its overlap to the weak recovery scale while competing overlaps remain near their initialization scale.

Slice-to-volume registration between ultrasound (US) and preoperative computed tomography (CT) imaging would enhance many minimally invasive interventions, for example by locating soft tissue structures intra-operatively that are discernible in CT. While optical tracking enables initial rigid registration, contact from the probe induces soft tissue deformations that inhibit accurate alignment. In this work, we introduce a deformable CT-ultrasound registration framework that incorporates anatomical priors derived from CT to improve registration under deformation. Rigid registration is first established using a robot-assisted optical tracking system, after which a deformable transformation is estimated using a sinusoidal implicit neural representation (SIREN) optimized per frame. Tissue stiffness is approximated from CT-based HU values and used as spatially varying regularization, suppressing deformation in rigid structures such as bone while allowing more flexibility in soft tissue. Two additional constraints capture the physics of probe contact: a contact-zone displacement prior that drives the displacement field to compress tissue below the probe face, and a fan-geometry regularization term based on beam direction and convex transducer field of view. Model parameters are optimized with a normalized gradient field (NGF). The proposed approach improves alignment over rigid initialisation by 17% and outperforms classical deformable baselines while maintaining near-zero topological folding.

Biological connectomes encode structured solutions to visual computation that may provide reusable inductive biases for artificial vision. We develop ConnectomeX around FlyVision, a trainable architecture that preserves parallel ON/OFF processing, recurrent computation and population-level graph interaction while scaling model capacity across tasks. FlyVision reached 99.34% accuracy on MNIST with 80,608 parameters and 78.03% on CIFAR-10 with 81,408 parameters. On ImageNet-1K, FlyVision Base and Large reached 60.79% and 66.25% top-1 accuracy with 1.8 and 3.7 million parameters, while a Large local-k7 model with a learned low-frequency branch reached 66.53%, compared with 69.25% for ResNet18 with 11.7 million parameters. On a 22-class skin-disease benchmark, FlyVision Large achieved 63.78% accuracy and 95.28% macro-AUROC with 2.99 million parameters. In four-class chest radiography, ImageNet-pretrained FlyVision Base and Large reached 92.60% and 92.76% accuracy with 1.33 and 2.97 million parameters, compared with 91.56% for ImageNet-pretrained ResNet18 with 11.18 million. BrainAGE extends FlyVision to volumetric T1-weighted MRI by applying a shared ImageNet-pretrained FlyVision Large encoder to 24 sagittal, coronal and axial slices per scan and combining slice-level age estimates by confidence-modulated Gaussian voting. On 433 held-out scans, three-axis fusion achieved a mean absolute error of 5.98 years and R^2 = 0.868. Across the 224x224 classification tasks, the best FlyVision configuration remained within three percentage points of ResNet18 on ImageNet-1K and skin-disease classification and exceeded it on chest radiography with substantially fewer parameters. These results show that a conserved connectome-informed computation can scale from compact recognition to large-scale natural and biomedical vision.

Recent advances in LipSync generation technology have led to the creation of highly realistic videos, posing severe societal risks. However, existing defense strategies struggle against LipSync forgeries, as advanced LipSync generation methods not only achieve better lip synchronization but also eliminate visual artifacts. An important reason is that they overlook an inherent biological coupling between lip movements and head poses in natural speech videos. In this paper, we propose LipDA, a novel framework for joint LipSync Detection and Attribution, which takes advantage of the inconsistency between head and lip. For detection, the framework learns to quantify this discrepancy by contrasting lip and pose features from authentic versus forged videos. For attribution, our method is designed to capture the unique temporal dynamics and audio-visual synchronization patterns that act as the fingerprint of models, enabling source tracing. We conduct extensive experiments on two challenging LipSync datasets as well as our own proposed large-scale and multi-generator dataset. LipDA achieves over 97% AUC in detection and 97.5% accuracy in model attribution, significantly outperforming existing methods. Code and the proposed LipSync-A dataset are available at https://github.com/AnsonShe/LipDA.

Plane Cremona maps of degree four are divided into two different classes according to their homaloidal type: either they are de Jonquières, that is, they have a base point of multiplicity 3, or they are not de Jonquières. The maps of the former class have length one, while the maps of the latter class have length two, according to the definition of length of a plane Cremona map given by Blanc and Furter. In this paper, we classify plane Cremona maps of degree four and length two, up to the action of automorphisms of the projective plane on the left-hand side and on the right-hand side of the map. This classification also allows us to compute the ordinary quadratic length of all such maps. In particular, we show that this ordinary quadratic length is at most 7 and we classify those maps which attain this upper bound.

Motivated by the role of finite-time restriction in one-dimensional (1D) systems, we investigate the restriction of two-dimensional (2D) systems defined over $\mathbb{N}^2$ to a vertical strip. We give a constructive proof that such a restriction induces a 1D system. A Gröbner basis-based algorithm is then developed to compute this restricted 1D system from a given 2D system. The framework is then applied to develop an algorithm for testing the time-controllability and zero-time-controllability of scalar autonomous 2D systems; these properties are known to be equivalent to the existence of a deadbeat controller. Finally, we show that, for time-controllable or zero-time-controllable scalar autonomous 2D systems, the induced 1D system generates all the admissible trajectories of the original 2D system, thereby reducing the complexity of trajectory generation.

Image Bitstream Fine-grained Understanding (IBFU) aims to directly perform fine-grained classification and semantic description generation from encoded image byte sequences. In contrast to conventional pixel-domain visual understanding, IBFU conducts semantic analysis without fully decoding images into the pixel domain. Since pixel-level visual content is not explicitly reconstructed during inference, this paradigm reduces visual exposure within the processing pipeline and suits privacy-friendly Artificial Intelligence of Things (AIoT) applications. In this paper, we propose Bitstream Fine-grained Generator (BFG), a novel foundation model tailored for IBFU. BFG consists of two main components: a Bitstream Semantic Encoder (BSeE) and a Fine-grained Semantic Generator (FSeG). BSeE directly models semantic representations from encoded image bitstreams without explicit pixel reconstruction, while FSeG transforms the extracted bitstream semantics into detailed natural-language descriptions through autoregressive generation. To train BFG and comprehensively evaluate IBFU in practical AIoT scenarios, where image bitstreams may suffer corruption during transmission and storage, we construct a large-scale Corrupted-bitstream Fine-grained Understanding dataset (CFU-D), containing both intact bitstreams and corrupted variants across multiple corruption types and severity levels. Experiments show that BFG maintains stable fine-grained caption generation under bitstream corruption. For example, the performance only has slight change from 0.6339 to 0.6077 in terms of average CIDEr score on Stanford Dogs Caption dataset, while vision-language models, such as Qwen-VL-Chat, BLIP-2, GLM, Gemini, and GPT suffer severe performance decrease. This paper provides a practical paradigm for privacy-friendly fine-grained understanding in AIoT.

Large language models routinely answer questions that cannot be answered from the information given, and in dialogue they answer before enough has been said. Unanswerability is linearly decodable from hidden states, but it is unclear which of its forms share a representation and whether the signal is useful in dialogue. We contribute a turn-labeled multi-turn benchmark (423 conversations, 1,661 labeled turn-states) and an evaluation harness with a simulated user who answers clarifying questions, and use them with six datasets and six open-weight LLMs to test how far probes for unanswerability carry. Probes transfer robustly between datasets that share a ground of unanswerability: missing information in math (AUROC 0.77-0.97) and in a passage (SQuAD 2.0<->MuSiQue, 0.77-0.90). Probes for epistemic "known-unknowns" transfer poorly to math, but this separation weakens under lexical controls and changes with layer and coordinate system, so it remains unresolved. Single-turn probes fail zero-shot to detect when a conversation becomes answerable; in-structure probes recover it, but no better than a bag-of-words classifier. A gate on the calibrated probe, with no model fine-tuning, fires on underspecified turns far more precisely than chance, and its end-task success comes within 0.08 of a gate given the true labels. Yet across four models it does not reliably beat vanilla generation or prompted consolidation. The remaining gap lies mostly in how models use a clarification, not in detection.

Collaborative inference between the edge cloud and user equipment (UE) is a promising paradigm for deploying large deep neural networks (DNNs) in 6G networks. However, existing cloud-edge inference and distillation schemes often require the real-time transmission of high-dimensional intermediate features or soft probability vectors, which imposes a substantial communication burden on bandwidth-limited wireless links. To address this challenge, we propose HintKD, a bandwidth-constrained distillation framework for cloud-edge inference. The core idea is to compress the teacher's guidance into compact discrete hints rather than transmit raw features directly. Specifically, a cloud-side teacher is first trained with a maximum conditional mutual information (MCMI) objective to preserve informative intra-class variations. Its latent representation is then mapped to a compact codebook through differentiable vector quantization. During deployment, the cloud transmits only a hint index, and the UE uses a lightweight FiLM-based adapter to modulate student features according to the received codeword. Experiments on the DeepSense 6G beam prediction task show that HintKD preserves competitive accuracy while reducing the inference-side communication payload from O(D_t) for feature transmission or O(C) for logit transmission to O(1) for discrete hint transmission. In addition, experiments on the MNIST dataset further verify the generality of the proposed distillation method across different datasets and tasks. These results show that HintKD achieves a favorable accuracy-bandwidth trade-off by replacing high-dimensional feature or logit transmission with a compact discrete hint index.

We report a 3D-printed microwave-resonator Paul trap that operates at the GHz frequencies required for electron confinement while preserving an open geometry suitable for laser access and imaging ion Coulomb crystals. The resonator exhibits moderate quality factors of approximately 1000, enabling large confining electric-field amplitudes at low input power, and is in good agreement with finite-element-method (FEM) predictions despite manufacturing imperfections, which we experimentally characterize. The validated FEM model is then used to calculate the electromagnetic fields in the trapping region. These field distributions, together with surface-roughness measurements, are incorporated into a model of electron spin-qubit decoherence to estimate the coherence time. Our analysis predicts that electron spin-qubit coherence times on the order of 10 ms should be achievable under the assumption that motional heating is dominated by Johnson noise, making such systems relevant for quantum information processing. The method of obtaining the coherence time is applicable to any electron Paul trap.

A recognizer that compares rotation-invariant descriptors sees a surface only up to the fiber of the descriptor. We measure this fiber by its radius in the orbit distance from the enrolled surface. A large radius admits decoys, that is, distant shapes that pass the matcher. A small radius discloses the enrolled shape to anyone who captures the stored value. For star-shaped surfaces truncated to spherical harmonics of degree at most $L$, with $n$ coefficients, a descriptor of generic rank $r$ has generic fibers of dimension $n-3-r$ modulo rotations. The standard pool of band powers, even bispectra, and three invariants of the degree-three band therefore admits decoy families of dimension $5$, $13$, $20$ at $L=4,6,8$. Its rank first reaches $n-3$ at $L=16$, and a mirror decoy remains at every $L$. The odd bispectra remove the mirror decoy generically for $L \geq 4$. Yet at fixed mean radius the same pool determines the enclosed volume exactly, and it does not determine whether a surface meets a clearance requirement. We certify two cases by exact and interval arithmetic. At $L=6$ a decoy matches all $32$ invariants to relative precision $2 \cdot 10^{-18}$ at orbit distance at least $0.87$ times the norm of the enrolled tuple. For the radar shape model of asteroid (101955) Bennu, the pool recovers the modeled volume, misses the handedness, and leaves the keep-out radius uncertain by more than $7 \, \mathrm{m}$.

This work studies how Bayesian machine learning methods can be used for large-scale demand estimation with many product categories. I compare two model classes, a latent factorization model and a mixed logit model and two Bayesian estimation approaches, Markov Chain Monte Carlo (MCMC) and Variational Inference (VI). The analysis combines a simulation study with an application to supermarket scanner data. The results show that the latent factorization model benefits from information across categories and improves its predictive performance as the dimensionality of the choice environment increases, whereas the mixed logit model does not exhibit the same pattern. MCMC delivers the highest predictive accuracy but is computationally intensive. VI achieves slightly lower predictive performance while substantially reducing runtime. In the empirical application, VI also outperforms the mixed logit benchmark. These findings highlight a trade-off between accuracy and computational feasibility in multi-category demand estimation.

We analyse the boundary behaviour of the gradient of the torsion function $w_\Om$ in convex domains $\Om\subset\mathbb{R}^d$. We prove that its $L^\infty$-norm is attained at a boundary point and establish the stability of this norm under general (convex) perturbations. For problems of the form $$\sup\{\|\nabla w_{\Om}\|_\infty: \Om \subset \R^d,\textup{open, bounded and convex}, G(\Om)=m\},$$ where $G$ is either volume or perimeter, we show the existence of maximisers which are $C^1$-smooth, rotationally symmetric about an axis and contain a $(d-1)$- dimensional ball (of estimated size) in their boundary. Other constraints, whether geometric (such as diameter or circumradius) or variational (such as torsion or first Dirichlet eigenvalue), are also briefly discussed, and some of their characteristic behaviours are highlighted.

Contextual information, capturing the circumstances of a user-item interaction, is central to recommender systems. Prior work draws context from location, time, or reviews, but not images; multimodal recommender systems mainly use images to enrich item or user representations, not identify situational context. We propose a new representation of context derived from images, spanning physical, social, and modal categories learned via a vision-language model. We introduce ICE-Fuse, a pipeline for evaluating this representation that fuses these categories and integrates them into a context-aware recommender system, using TripAdvisor data and Review-aware Graph Contrastive Learning as the recommendation algorithm. Image context does not outperform established signals standalone, but improves them combined, indicating complementary information. Semantic analysis shows image- and review-derived context capture distinct aspects of the interaction, positioning images as complementary context.

Dynamic Application Security Testing (DAST) scanners achieve high recall but also produce a large number of false positives, resulting in substantial manual triage costs. Large Language Models (LLMs), when used for independent detection, achieve extremely high recall (95.4%-100%) but also exhibit prohibitively high false positive rates (49.6%-85.0%), precluding their use as standalone replacements for scanners. A natural solution is a two-stage cascade consisting of scanner detection followed by LLM verification. However, a verification-granularity issue that has long been overlooked in practice creates a structural bottleneck: alert aggregation binds multiple true and false cases into a shared decision unit, such that removing a false positive inevitably eliminates true positives aggregated within the same alert group. This creates a trade-off bottleneck between the False-positive Reduction Rate (FRR) and the True-positive Rate (TPR). We formalize this bottleneck by showing that the alert-level false-positive set is a subset of the case-level false-positive set, and introduce a Case-Level, per-case verification strategy that shifts the decision granularity from the alert level to the instance level, independently replaying HTTP requests and making an independent determination for each detected case. Evaluation on the dual testbeds of Damn Vulnerable Web Application (DVWA) and WebGoat shows that the empirically best Alert-Level operating point achieves FRR=42.86% (TPR=51.7%). The Case-Level Baseline achieves FRR=47.6%, an improvement of 4.7 percentage points (+4.7 pp), while the Case-Focused Evidence Verification Prompt (CEV-Prompt) increases TPR from 55.2% to 62.1% at the same FRR.

Large Language Models have emerged as promising tools for software vulnerability analysis, but their effectiveness depends heavily on prompt design. Existing research primarily compares prompting strategies using aggregate performance metrics, providing limited insight into why models fail or how prompts can be improved systematically. We propose Failure-Driven Prompt Refinement (FDPR), a methodology that analyzes recurring model failures to guide evidence-based prompt refinement. Using the Damn Vulnerable Java Application (DVJA), we identify recurring failure modes, including false positives, false negatives, unsupported reasoning, and CWE misclassification, and translate them into targeted prompt refinements. We then evaluate the resulting prompt on the Juliet Test Suite and perform cross-model validation to assess generalizability. The results show that failure-driven refinement improves the reliability of LLM-based vulnerability analysis while yielding reusable prompt design principles. More broadly, this work demonstrates that recurring model failures provide a principled foundation for prompt engineering, enabling the systematic development of more reliable LLM-based vulnerability analysis systems.

In this letter, we derive a necessary and sufficient condition for the existence of a deadbeat observer (DBO) for systems described by higher-order difference equations based on finite historical data. Furthermore, we develop an algorithm to construct a consistent DBO and to parametrise all DBOs directly from the historical data. Finally, we apply the proposed approach to the problem of state estimation in the presence of unknown inputs.

Large Language Models (LLMs) require substantial computational resources, limiting their deployment on resource-constrained hardware. Ternary LLMs mitigate these demands through weight quantization via ternary values, achieving significant compression often with 50-90% sparsity. However, existing approaches have limitations: methods optimized for ternary weights, such as BitNet, redundant segment reduction (RSR), and its improved version RSR++, do not exploit sparsity structures, while conventional sparse formats neglect ternary characteristics, foregoing dual optimization opportunities. In this paper, we introduce Sparse Segment Reduction (SSR), a ternary matrix multiplication method designed to accelerate the inference of ternary LLMs and general Ternary Weight Networks (TWNs). SSR has a dedicated optimized ternary data format and an algorithm that systematically exploits sparsity patterns through computation trees that scale with the sparsity. SSR provides theoretical gains with asymptotically faster inference than RSR++ for sparsity above 50%, while practical evaluations reveal performance improvements across all sparsity levels. Evaluation results show that SSR achieves 2.1-11.3x speedup over RSR++ on ternary GEMM with 45-95% sparsity. Furthermore, SSR achieves 3.5-6.3x end-to-end speedup and 4.9% of memory saving over RSR++ on the Llama-3 1B model inference.

Multi-turn LLM agents often receive sparse task feedback across several interactions, while generating each response token by token. This creates two related credit-assignment questions: which responses helped achieve the outcome, and which generation decisions mattered within each response? Existing methods typically focus on only one level: turn-level methods evaluate complete responses but do not distinguish the decisions within them; token-level methods can propagate feedback across turns but do not explicitly model credit for each response. These complementary limitations motivate learning credit at both levels and coordinating it in a single policy update. We introduce VETTA, a credit assignment method that jointly learns turn- and token-level values through separate heads on a shared lightweight critic. VETTA computes advantages along both temporal sequences and combines each turn advantage with a within-response-centered token residual for PPO updates. Furthermore, to reduce value-learning cost, the critic retains only early Transformer blocks from the pretrained checkpoint used to initialize the actor. Experiments on four agent benchmarks demonstrate VETTA's strong performance. With Qwen2.5-7B-Instruct, VETTA achieves success rates of 95.5% on ALFWorld, 76.0% on WebShop, and 52.2% on ScienceWorld. With Qwen2.5-VL-3B-Instruct, it reaches 84.9% and 80.7% success on Sokoban under sparse and dense rewards, respectively. Critic-depth comparisons further show strong task performance with substantially lower critic-side computation. These results demonstrate that coordinating turn- and token-level credit with a lightweight shared critic improves agent performance while keeping value estimation efficient. Code is available at https://github.com/Jiaju-Chen/VETTA-official.

While recent vision-language models (VLMs) have shown outstanding performance across diverse applications, they tend to under-use visual information and over-rely on textual context. In this work, we propose GeoPID, a training-free framework that analyzes multimodal information within VLMs from a geometric perspective. GeoPID decomposes information into Redundant, Modality-Unique, and Synergistic components through the geometric relationships between visual and textual representation subspaces. Through an extensive analysis across 22 VLMs and 14 benchmarks, we confirm that correct predictions exhibit stronger vision-unique components when questions strongly require visual grounding. Building on this geometric analysis, we introduce a targeted intervention technique that selectively amplifies visual representations along the vision-unique subspace during inference. As a result, visual grounding capabilities were enhanced without any additional model parameter updates, achieving an average relative accuracy gain of 7.63%.

Large-scale self-supervised pretraining has reshaped modern machine learning, substantially advancing the ability of language and vision models to generalize across downstream tasks. While deep learning has driven considerable progress in modeling atomistic systems in recent years, self-supervised pretraining in this domain has not yet achieved comparable downstream generalization. To address this, we introduce Atom-JEPA, a self-supervised pretraining framework that learns latent representations from unlabeled 3D structures through complementary atom-level and substructure-level objectives inspired by joint-embedding predictive architectures. We pretrain Atom-JEPA on large-scale molecular and crystalline datasets and evaluate its transfer performance by fine-tuning on a diverse set of downstream property prediction tasks. Atom-JEPA achieves state-of-the-art performance on molecular ADMET and quantum-chemical property prediction tasks, and is highly competitive in predicting the physical properties of crystalline materials. These results demonstrate the potential of latent-space predictive pretraining to support broad downstream generalization from structural data alone. Code and pretrained model checkpoints are publicly available at https://github.com/khelverskovp/atom-jepa

We investigate strong gravitational lensing of gravitational waves from a compact source orbiting a Kerr black hole, motivated by self-lensing in binary extreme-mass-ratio inspirals. In the geometric-optics limit, we construct a point-to-point lens map from null-geodesic integrals and use the Jacobi map to compute the strain magnification of individual images. While lensed signals are generally demagnified, we show that they can be strongly amplified near the extended caustics of the Kerr lens map. For suitable source–observer geometries, the orbit crosses a caustic twice per revolution, producing two amplification peaks. Their arrival-time separation depends on the black-hole spin, observer inclination, and orbital radius through both the source motion and propagation delays. These signatures might offer a means of constraining the source–observer geometry and probing the spacetime of the central black hole.