Visual reinforcement learning is appealing for robotics but expensive -- off-policy methods are sample-efficient yet slow; on-policy methods parallelize well but waste samples. Recent work has shown that off-policy methods can train faster than on-policy methods in wall-clock time for state-based control. Extending this to vision remains challenging, where high-dimensional input images complicate training dynamics and introduce substantial storage and encoding overhead. To address these challenges, we introduce Squint, a visual Soft Actor Critic method that achieves faster wall-clock training than prior visual off-policy and on-policy methods. Squint achieves this via parallel simulation, a distributional critic, resolution squinting, layer normalization, a tuned update-to-data ratio, and an optimized implementation. We evaluate on the SO-101 Task Set, a new suite of eight manipulation tasks in ManiSkill3 with heavy domain randomization, and demonstrate sim-to-real transfer to a real SO-101 robot. We train policies for 15 minutes on a single RTX 3090 GPU, with most tasks converging in under 6 minutes.

We study efficient multi-vector retrieval for late interaction in any modality. Late interaction has emerged as a dominant paradigm for information retrieval in text, images, visual documents, and videos, but its computation and storage costs grow linearly with document length, making it costly for image-, video-, and audio-rich corpora. To address this limitation, we explore query-agnostic methods for compressing multi-vector document representations under a constant vector budget. We introduce four approaches for index compression: sequence resizing, memory tokens, hierarchical pooling, and a novel attention-guided clustering (AGC). AGC uses an attention-guided mechanism to identify the most semantically salient regions of a document as cluster centroids and to weight token aggregation. Evaluating these methods on retrieval tasks spanning text (BEIR), visual-document (ViDoRe), and video (MSR-VTT, MultiVENT 2.0), we show that attention-guided clustering consistently outperforms other parameterized compression methods (sequence resizing and memory tokens), provides greater flexibility in index size than non-parametric hierarchical clustering, and achieves competitive or improved performance compared to a full, uncompressed index. The source code is available at: github.com/hanxiangqin/omni-col-press.

A numerical study of tetrahedral Raviart-Thomas mixed finite element methods is presented in the solution of model second order boundary value problems posed in a curved spatial domain. An emphasis is given to the case where normal fluxes are prescribed on a boundary portion. In this case the question on the best way to enforce known boundary degrees of freedom is raised. It seems intuitive that the normal component of the flux variable should preferably not take up corresponding prescribed values at nodes shifted to the boundary of the approximating polyhedron in the underlying normal direction. This is because an accuracy downgrade is to be expected, as shown in https://doi.org/10.1137/15M1045442 and https://doi.org/10.1051/m2an/2025028. In the former work accuracy improvement is achieved by means of a standard Galerkin formulation with parametric elements. The latter one in turn advocates the use of straight-edged triangles combined with a Petrov-Galerkin formulation, in which the aforementioned shift applies only to the test-flux space, while the shape-flux space consists of fields whose fluxes satisfy the prescribed conditions on the true boundary. The first purpose of this article is to show that the method studied in https://doi.org/10.1051/m2an/2025028 for two-dimensional problems can be extended quite naturally to the three-dimensional case. More particularly we illustrate this by carrying out numerical experimentation with such a version for the two lowest order methods of this family, as compared to the corresponding do-nothing strategy. In the case of the lowest order method this comparative study is enriched by assessing as well the performance of its Hermite analog introduced in https://doi.org/10.1016/j.cam.2012.08.027.

Efficiently processing long sequences with Transformer models usually requires splitting the computations across accelerators via context parallelism. The dominant approaches in this family of methods, such as Ring Attention or DeepSpeed Ulysses, enable scaling over the context dimension but do not focus on memory efficiency, which limits the sequence lengths they can support. More advanced techniques, such as Fully Pipelined Distributed Transformer or activation offloading, can further extend the possible context length at the cost of training throughput. In this paper, we present UPipe, a simple yet effective context parallelism technique that performs fine-grained chunking at the attention head level. This technique significantly reduces the activation memory usage of self-attention, breaking the activation memory barrier and unlocking much longer context lengths. Our approach reduces intermediate tensor memory usage in the attention layer by as much as 87.5$\%$ for 32B Transformers, while matching previous context parallelism techniques in terms of training speed. UPipe can support the context length of 5M tokens when training Llama3-8B on a single 8$\times$H100 node, improving upon prior methods by over 25$\%$.

Cryo-electron tomography (cryo-ET) enables high resolution, three-dimensional reconstruction of biological structures, including membranes and membrane proteins. Identification of regions of interest (ROIs) is central to scientific imaging, as it enables isolation and quantitative analysis of specific structural features within complex datasets. In practice, however, ROIs are typically derived indirectly through full structure segmentation followed by post hoc analysis. This limitation is especially apparent for continuous and geometrically complex structures such as membranes, which are segmented as single entities. Here, we developed TomoROIS-SurfORA, a two step framework for direct, shape-agnostic ROI segmentation and morphological surface analysis. TomoROIS performs deep learning-based ROI segmentation and can be trained from scratch using small annotated datasets, enabling practical application across diverse imaging data. SurfORA processes segmented structures as point clouds and surface meshes to extract quantitative morphological features, including inter-membrane distances, curvature, and surface roughness. It supports both closed and open surfaces, with specific considerations for open surfaces, which are common in cryo-ET due to the missing wedge effect. We demonstrate both tools using in vitro reconstituted membrane systems containing deformable vesicles with complex geometries, enabling automatic quantitative analysis of membrane contact sites and remodeling events such as invagination. While demonstrated here on cryo-ET membrane data, the combined approach is applicable to ROI detection and surface analysis in broader scientific imaging contexts.

Despite rapid recent progress in the terminal capabilities of large language models, the training data strategies behind state-of-the-art terminal agents remain largely undisclosed. We address this gap through a systematic study of data engineering practices for terminal agents, making two key contributions: (1) Terminal-Task-Gen, a lightweight synthetic task generation pipeline that supports seed-based and skill-based task construction, and (2) a comprehensive analysis of data and training strategies, including filtering, curriculum learning, long context training, and scaling behavior. Our pipeline yields Terminal-Corpus, a large-scale open-source dataset for terminal tasks. Using this dataset, we train Nemotron-Terminal, a family of models initialized from Qwen3(8B, 14B, 32B) that achieve substantial gains on Terminal-Bench 2.0: Nemotron-Terminal-8B improves from 2.5% to 13.0% Nemotron-Terminal-14B improves from 4.0% to 20.2%, and Nemotron-Terminal-32B improves from 3.4% to 27.4%, matching the performance of significantly larger models. To accelerate research in this domain, we open-source our model checkpoints and most of our synthetic datasets at https://huggingface.co/collections/nvidia/nemotron-terminal.

We study the complexity of smoothed agnostic learning, recently introduced by~\cite{CKKMS24}, in which the learner competes with the best classifier in a target class under slight Gaussian perturbations of the inputs. Specifically, we focus on the prototypical task of agnostically learning halfspaces under subgaussian distributions in the smoothed model. The best known upper bound for this problem relies on $L_1$-polynomial regression and has complexity $d^{\tilde{O}(1/σ^2) \log(1/ε)}$, where $σ$ is the smoothing parameter and $ε$ is the excess error. Our main result is a Statistical Query (SQ) lower bound providing formal evidence that this upper bound is close to best possible. In more detail, we show that (even for Gaussian marginals) any SQ algorithm for smoothed agnostic learning of halfspaces requires complexity $d^{Ω(1/σ^{2}+\log(1/ε))}$. This is the first non-trivial lower bound on the complexity of this task and nearly matches the known upper bound. Roughly speaking, we show that applying $L_1$-polynomial regression to a smoothed version of the function is essentially best possible. Our techniques involve finding a moment-matching hard distribution by way of linear programming duality. This dual program corresponds exactly to finding a low-degree approximating polynomial to the smoothed version of the target function (which turns out to be the same condition required for the $L_1$-polynomial regression to work). Our explicit SQ lower bound then comes from proving lower bounds on this approximation degree for the class of halfspaces.

Recent diffusion methods have made significant progress in generating videos from single images due to their powerful visual generation capabilities. However, challenges persist in image-to-video synthesis, particularly in human video generation, where inferring view-consistent, motion-dependent clothing wrinkles from a single image remains a formidable problem. In this paper, we present Human Video Generation in 4D (HVG), a latent video diffusion model capable of generating high-quality, multi-view, spatiotemporally coherent human videos from a single image with 3D pose and view control. HVG achieves this through three key designs: (i) Articulated Pose Modulation, which captures the anatomical relationships of 3D joints via a novel dual-dimensional bone map and resolves self-occlusions across views by introducing 3D information; (ii) View and Temporal Alignment, which ensures multi-view consistency and alignment between a reference image and pose sequences for frame-to-frame stability; and (iii) Progressive Spatio-Temporal Sampling with temporal alignment to maintain smooth transitions in long multi-view animations. Extensive experiments on image-to-video tasks demonstrate that HVG outperforms existing methods in generating high-quality 4D human videos from diverse human images and pose inputs.

While Vision-Language Models (VLMs) exhibit exceptional 2D visual understanding, their ability to comprehend and reason about 3D space--a cornerstone of spatial intelligence--remains superficial. Current methodologies attempt to bridge this domain gap either by relying on explicit 3D modalities or by augmenting VLMs with partial, view-conditioned geometric priors. However, such approaches hinder scalability and ultimately burden the language model with the ill-posed task of implicitly reconstructing holistic 3D geometry from sparse cues. In this paper, we argue that spatial intelligence can emerge inherently from 2D vision alone, rather than being imposed via explicit spatial instruction tuning. To this end, we introduce Spa3R, a self-supervised framework that learns a unified, view-invariant spatial representation directly from unposed multi-view images. Spa3R is built upon the proposed Predictive Spatial Field Modeling (PSFM) paradigm, where Spa3R learns to synthesize feature fields for arbitrary unseen views conditioned on a compact latent representation, thereby internalizing a holistic and coherent understanding of the underlying 3D scene. We further integrate the pre-trained Spa3R Encoder into existing VLMs via a lightweight adapter to form Spa3-VLM, effectively grounding language reasoning in a global spatial context. Experiments on the challenging VSI-Bench demonstrate that Spa3-VLM achieves state-of-the-art accuracy of 58.6% on 3D VQA, significantly outperforming prior methods. These results highlight PSFM as a scalable path toward advancing spatial intelligence. Code is available at https://github.com/hustvl/Spa3R.

Bengali, despite being one of the most widely spoken languages globally, remains underrepresented in long form speech technology, particularly in systems addressing transcription and speaker attribution. We present frameworks for long form Bengali speech intelligence that address automatic speech recognition using a Whisper Medium based model and speaker diarization using a finetuned segmentation model. The ASR pipeline incorporates vocal separation, voice activity detection, and a gap aware windowing strategy to construct context preserving segments for stable decoding. For diarization, a pretrained speaker segmentation model is finetuned on the official competition dataset (provided as part of the DL Sprint 4.0 competition organized under BUET CSE Fest), to better capture Bengali conversational patterns. The resulting systems deliver both efficient transcription of long form audio and speaker aware transcription to provide scalable speech technology solutions for low resource languages.

The CAP theorem is routinely treated as a systems law: under network partition, a replicated service must sacrifice either consistency or availability. The theorem is correct within its standard asynchronous network model, but operational practice depends on where partition-like phenomena become observable and on how lower layers discard or preserve semantic information about message fate. This paper argues that Open Atomic Ethernet (OAE) shifts the engineering regime in which CAP tradeoffs become application-visible by (i) replacing fire-and-forget link semantics with bounded-time bilateral reconciliation of endpoint state -- the property we call bisynchrony -- and (ii) avoiding Clos funnel points via an octavalent mesh in which each node can act as the root of a locally repaired spanning tree. The result is not the elimination of hard graph cuts, but a drastic reduction in the frequency and duration of application-visible "soft partitions" by detecting and healing dominant fabric faults within hundreds of nanoseconds. We connect this view to Brewer's original CAP framing, the formalization by Gilbert and Lynch, the CAL theorem of Lee et al., which replaces binary partition tolerance with a quantitative measure of apparent latency, and Abadi's PACELC extension.

Machine learning and data systems increasingly function as infrastructures of memory: they ingest, store, and operationalize traces of personal, political, and cultural life. Yet contemporary governance demands credible forms of forgetting, from GDPR-backed deletion to harm-mitigation and the removal of manipulative content, while technical infrastructures are optimized to retain, replicate, and reuse. This work argues that "forgetting" in computational systems cannot be reduced to a single operation (e.g., record deletion) and should instead be treated as a sociotechnical practice with distinct mechanisms and consequences. We clarify a vocabulary that separates erasure (removing or disabling access to data artifacts), unlearning (interventions that bound or remove a data point influence on learned parameters and outputs), exclusion (upstream non-collection and omission), and forgetting as an umbrella term spanning agency, temporality, reversibility, and scale. Building on examples from machine unlearning, semantic dependencies in data management, participatory data modeling, and manipulation at scale, we show how forgetting can simultaneously protect rights and enable silencing. We propose reframing unlearning as a first-class capability in knowledge infrastructures, evaluated not only by compliance or utility retention, but by its governance properties: transparency, accountability, and epistemic justice.

Graph-based medical image segmentation represents anatomical structures using boundary graphs, providing fixed-topology landmarks and inherent population-level correspondences. However, their clinical adoption has been hindered by a major requirement: training datasets with manually annotated landmarks that maintain point-to-point correspondences across patients rarely exist in practice. We introduce Mask-HybridGNet, a framework that trains graph-based models directly using standard pixel-wise masks, eliminating the need for manual landmark annotations. Our approach aligns variable-length ground truth boundaries with fixed-length landmark predictions by combining Chamfer distance supervision and edge-based regularization to ensure local smoothness and regular landmark distribution, further refined via differentiable rasterization. A significant emergent property of this framework is that predicted landmark positions become consistently associated with specific anatomical locations across patients without explicit correspondence supervision. This implicit atlas learning enables temporal tracking, cross-slice reconstruction, and morphological population analyses. Beyond direct segmentation, Mask-HybridGNet can extract correspondences from existing segmentation masks, allowing it to generate stable anatomical atlases from any high-quality pixel-based model. Experiments across chest radiography, cardiac ultrasound, cardiac MRI, and fetal imaging demonstrate that our model achieves competitive results against state-of-the-art pixel-based methods, while ensuring anatomical plausibility by enforcing boundary connectivity through a fixed graph adjacency matrix. This framework leverages the vast availability of standard segmentation masks to build structured models that maintain topological integrity and provide implicit correspondences.

Deep learning has significantly advanced automated brain tumor diagnosis, yet clinical adoption remains limited by interpretability and computational constraints. Conventional models often act as opaque ''black boxes'' and fail to quantify the complex, irregular tumor boundaries that characterize malignant growth. To address these challenges, we present XMorph, an explainable and computationally efficient framework for fine-grained classification of three prominent brain tumor types: glioma, meningioma, and pituitary tumors. We propose an Information-Weighted Boundary Normalization (IWBN) mechanism that emphasizes diagnostically relevant boundary regions alongside nonlinear chaotic and clinically validated features, enabling a richer morphological representation of tumor growth. A dual-channel explainable AI module combines GradCAM++ visual cues with LLM-generated textual rationales, translating model reasoning into clinically interpretable insights. The proposed framework achieves a classification accuracy of 96.0%, demonstrating that explainability and high performance can co-exist in AI-based medical imaging systems. The source code and materials for XMorph are all publicly available at: https://github.com/ALSER-Lab/XMorph.

Text-to-image retrieval is a fundamental task in vision-language learning, yet in real-world scenarios it is often challenged by short and underspecified user queries. Such queries are typically only one or two words long, rendering them semantically ambiguous, prone to collisions across diverse visual interpretations, and lacking explicit control over the quality of retrieved images. To address these issues, we propose a new paradigm of quality-controllable retrieval, which enriches short queries with contextual details while incorporating explicit notions of image quality. Our key idea is to leverage a generative language model as a query completion function, extending underspecified queries into descriptive forms that capture fine-grained visual attributes such as pose, scene, and aesthetics. We introduce a general framework that conditions query completion on discretized quality levels, derived from relevance and aesthetic scoring models, so that query enrichment is not only semantically meaningful but also quality-aware. The resulting system provides three key advantages: 1) flexibility, it is compatible with any pretrained vision-language model (VLMs) without modification; 2) transparency, enriched queries are explicitly interpretable by users; and 3) controllability, enabling retrieval results to be steered toward user-preferred quality levels. Extensive experiments demonstrate that our proposed approach significantly improves retrieval results and provides effective quality control, bridging the gap between the expressive capacity of modern VLMs and the underspecified nature of short user queries. Our code is available at https://github.com/Jianglin954/QCQC.

Hierarchical, multi-resolution volumetric mapping approaches are widely used to represent large and complex environments as they can efficiently capture their occupancy and connectivity information. Yet widely used path planning methods such as sampling and trajectory optimization do not exploit this explicit connectivity information, and search-based methods such as A* suffer from scalability issues in large-scale high-resolution maps. In many applications, Euclidean shortest paths form the underpinning of the navigation system. For such applications, any-angle planning methods, which find optimal paths by connecting corners of obstacles with straight-line segments, provide a simple and efficient solution. In this paper, we present a method that has the optimality and completeness properties of any-angle planners while overcoming computational tractability issues common to search-based methods by exploiting multi-resolution representations. Extensive experiments on real and synthetic environments demonstrate the proposed approach's solution quality and speed, outperforming even sampling-based methods. The framework is open-sourced to allow the robotics and planning community to build on our research.

Counterfactual inference enables clinicians to ask "what if" questions about patient outcomes, but standard methods assume feature independence and simultaneous modifiability -- assumptions violated by longitudinal clinical data. We introduce the Sequential Counterfactual Framework, which respects temporal dependencies in electronic health records by distinguishing immutable features (chronic diagnoses) from controllable features (lab values) and modeling how interventions propagate through time. Applied to 2,723 COVID-19 patients (383 Long COVID heart failure cases, 2,340 matched controls), we demonstrate that 38-67% of patients with chronic conditions would require biologically impossible counterfactuals under naive methods. We identify a cardiorenal cascade (CKD -> AKI -> HF) with relative risks of 2.27 and 1.19 at each step, illustrating temporal propagation that sequential -- but not naive -- counterfactuals can capture. Our framework transforms counterfactual explanation from "what if this feature were different?" to "what if we had intervened earlier, and how would that propagate forward?" -- yielding clinically actionable insights grounded in biological plausibility.

Pinching-antenna systems have recently emerged as a promising solution for enhancing coverage in high-frequency wireless communications by guiding signals through dielectric waveguides and radiating them via position-adjustable antennas. However, their practical deployment is fundamentally constrained by waveguide attenuation and line-installation requirements, which limit the achievable coverage range. To address this challenge, this paper investigates a wireless-fed pinching-antenna architecture that employs highly directional horn antennas to enable efficient coverage extension. Specifically, a full-duplex amplify-and-forward relay equipped with horn antennas is introduced between the base station and the waveguide input, which significantly improves the link budget in high-frequency bands while effectively eliminating self-interference. On this basis, we formulate a total power minimization problem subject to a quality-of-service constraint at the user equipment, involving the joint optimization of the pinching-antenna position, the relay amplification gain, and the base station transmit power. By exploiting the structure of the end-to-end signal-to-noise ratio, the optimal pinching-antenna position is first obtained in closed form by balancing waveguide attenuation and free-space path loss. Subsequently, closed-form expressions for the optimal relay gain and transmit power are derived. Numerical results demonstrate that the proposed scheme substantially outperforms conventional systems without relaying and relay-assisted transmission with fixed antennas in terms of total power consumption.

Patient-generated text such as secure messages, surveys, and interviews contains rich expressions of the patient voice (PV), reflecting communicative behaviors and social determinants of health (SDoH). Traditional qualitative coding frameworks are labor intensive and do not scale to large volumes of patient-authored messages across health systems. Existing machine learning (ML) and natural language processing (NLP) approaches provide partial solutions but often treat patient-centered communication (PCC) and SDoH as separate tasks or rely on models not well suited to patient-facing language. We introduce PVminer, a domain-adapted NLP framework for structuring patient voice in secure patient-provider communication. PVminer formulates PV detection as a multi-label, multi-class prediction task integrating patient-specific BERT encoders (PV-BERT-base and PV-BERT-large), unsupervised topic modeling for thematic augmentation (PV-Topic-BERT), and fine-tuned classifiers for Code, Subcode, and Combo-level labels. Topic representations are incorporated during fine-tuning and inference to enrich semantic inputs. PVminer achieves strong performance across hierarchical tasks and outperforms biomedical and clinical pre-trained baselines, achieving F1 scores of 82.25% (Code), 80.14% (Subcode), and up to 77.87% (Combo). An ablation study further shows that author identity and topic-based augmentation each contribute meaningful gains. Pre-trained models, source code, and documentation will be publicly released, with annotated datasets available upon request for research use.

Pinching antenna systems (PASS) have recently emerged as a promising architecture for high-frequency wireless communications. In this letter, we investigate localization in PASS by jointly exploiting the received signal amplitude and phase information, unlike recent works that consider only the amplitude information. A complex baseband signal model is formulated to capture free-space path loss, waveguide attenuation, and distance-dependent phase rotation between the user and each pinching antenna. Using this model, we derive the Fisher information matrix (FIM) with respect to the user location and obtain closed-form expressions for the Cramer-Rao lower bound (CRLB) and the position error bound (PEB). A maximum likelihood (ML) estimator that jointly considers the received signal amplitude and phase is developed to estimate the unknown user location. Given the non-convexity of the estimation problem, a two-stage solution combining coarse grid search and Levenberg-Marquardt refinement is proposed. Numerical results demonstrate that the proposed phase-aware estimator outperforms existing amplitude-only method in terms of positioning accuracy.

Classical robotic systems typically rely on custom planners designed for constrained environments. While effective in restricted settings, these systems lack generalization capabilities, limiting the scalability of embodied AI and general-purpose robots. Recent data-driven Vision-Language-Action (VLA) approaches aim to learn policies from large-scale simulation and real-world data. However, the continuous action space of the physical world significantly exceeds the representational capacity of linguistic tokens, making it unclear if scaling data alone can yield general robotic intelligence. To address this gap, we propose ActionReasoning, an LLM-driven framework that performs explicit action reasoning to produce physics-consistent, prior-guided decisions for robotic manipulation. ActionReasoning leverages the physical priors and real-world knowledge already encoded in Large Language Models (LLMs) and structures them within a multi-agent architecture. We instantiate this framework on a tractable case study of brick stacking, where the environment states are assumed to be already accurately measured. The environmental states are then serialized and passed to a multi-agent LLM framework that generates physics-aware action plans. The experiments demonstrate that the proposed multi-agent LLM framework enables stable brick placement while shifting effort from low-level domain-specific coding to high-level tool invocation and prompting, highlighting its potential for broader generalization. This work introduces a promising approach to bridging perception and execution in robotic manipulation by integrating physical reasoning with LLMs.

Accurate interpretation of electrocardiogram (ECG) signals is crucial for diagnosing cardiovascular diseases. Recent multimodal approaches that integrate ECGs with accompanying clinical reports show strong potential, but they still face two main concerns from a modality perspective: (1) intra-modality: existing models process ECGs in a lead-agnostic manner, overlooking spatial-temporal dependencies across leads, which restricts their effectiveness in modeling fine-grained diagnostic patterns; (2) inter-modality: existing methods directly align ECG signals with clinical reports, introducing modality-specific biases due to the free-text nature of the reports. In light of these two issues, we propose CG-DMER, a contrastive-generative framework for disentangled multimodal ECG representation learning, powered by two key designs: (1) Spatial-temporal masked modeling is designed to better capture fine-grained temporal dynamics and inter-lead spatial dependencies by applying masking across both spatial and temporal dimensions and reconstructing the missing information. (2) A representation disentanglement and alignment strategy is designed to mitigate unnecessary noise and modality-specific biases by introducing modality-specific and modality-shared encoders, ensuring a clearer separation between modality-invariant and modality-specific representations. Experiments on three public datasets demonstrate that CG-DMER achieves state-of-the-art performance across diverse downstream tasks.

We present SPRITETOMESH, a fully automatic pipeline for converting 2D game sprite images into triangle meshes compatible with skeletal animation frameworks such as Spine2D. Creating animation-ready meshes is traditionally a tedious manual process requiring artists to carefully place vertices along visual boundaries, a task that typically takes 15-60 minutes per sprite. Our method addresses this through a hybrid learned-algorithmic approach. A segmentation network (EfficientNet-B0 encoder with U-Net decoder) trained on over 100,000 sprite-mask pairs from 172 games achieves an IoU of 0.87, providing accurate binary masks from arbitrary input images. From these masks, we extract exterior contour vertices using Douglas-Peucker simplification with adaptive arc subdivision, and interior vertices along visual boundaries detected via bilateral-filtered multi-channel Canny edge detection with contour-following placement. Delaunay triangulation with mask-based centroid filtering produces the final mesh. Through controlled experiments, we demonstrate that direct vertex position prediction via neural network heatmap regression is fundamentally not viable for this task: the heatmap decoder consistently fails to converge (loss plateau at 0.061) while the segmentation decoder trains normally under identical conditions. We attribute this to the inherently artistic nature of vertex placement - the same sprite can be meshed validly in many different ways. This negative result validates our hybrid design: learned segmentation where ground truth is unambiguous, algorithmic placement where domain heuristics are appropriate. The complete pipeline processes a sprite in under 3 seconds, representing a speedup of 300x-1200x over manual creation. We release our trained model to the game development community.

In this paper, we propose the problem of Encounter-Driven Information Diffusion (EDID). In EDID, robots are allowed to exchange information only upon meeting. Crucially, EDID assumes that the robots are not allowed to schedule their meetings. As such, the robots have no means to anticipate when, where, and who they will meet. As a step towards the design of storage and routing algorithms for EDID, in this paper we propose a model of information diffusion that captures the essential dynamics of EDID. The model is derived from first principles and is composed of two levels: a micro model, based on a generalization of the concept of `mean free path'; and a macro model, which captures the global dynamics of information diffusion. We validate the model through extensive robot simulations, in which we consider swarm size, communication range, environment size, and different random motion regimes. We conclude the paper with a discussion of the implications of this model on the algorithms that best support information diffusion according to the parameters of interest.

While tensor-based methods excel at Direction-of-Arrival (DOA) estimation, their performance degrades severely with faulty or sparse arrays that violate the required manifold structure. To address this challenge, we propose Tensor Completion for Defective Arrays (TCDA), a robust algorithm that reformulates the physical imperfection problem as a data recovery task within a virtual tensor space. We present a detailed derivation for constructing an incomplete third-order Parallel Factor Analysis (PARAFAC) tensor from the faulty array signals via subarray partitioning, cross-correlation, and dimensional reshaping. Leveraging the tensor's inherent low-rank structure, an Alternating Least Squares (ALS)-based algorithm directly recovers the factor matrices embedding the DOA parameters from the incomplete observations. This approach provides a software-defined 'self-healing' capability, demonstrating exceptional robustness against random element failures without requiring additional processing steps for DOA estimation.

Selective state space models (SSMs) have rapidly become a compelling backbone for large language models, especially for long-context workloads. Yet in deployment, their inference performance is often bounded by the memory capacity, bandwidth, and latency limits of a single GPU, making multi-GPU execution increasingly necessary. Although tensor parallelism (TP) is widely used to scale Transformer inference, applying it to selective SSM blocks is non-trivial because the SSM mixer couples large projections with a sequence-wise recurrent state update and local mixing whose efficiency depends on preserving locality and avoiding synchronization in the critical path. This paper presents a communication-efficient TP design for selective SSM inference that addresses three practical engineering challenges: enabling TTFT improvements via an SSM state cache across prefill and decode, partitioning the mixer's packed parameter tensor so that recurrent updates remain local while minimizing communication, and reducing TP aggregation overhead with quantized AllReduce. We evaluate on three representative SSM-based LLMs spanning pure-SSM and hybrid architectures - Mamba, Falcon-Mamba, and Zamba - on NVIDIA A6000 and A100 clusters. Our experiments show substantial throughput gains from tensor-parallel SSM inference, improving batch-request throughput by ~1.6-2.1x on 2 GPUs and ~2.6-4.0x on 4 GPUs for Mamba, with the largest benefits at long context lengths, and achieving a further ~10-18% throughput improvement from quantized all-reduce by lowering synchronization bandwidth overhead.

Large language model (LLM)-based agents are increasingly used to solve complex tasks involving tool use, such as web browsing, code execution, and data analysis. However, current evaluation benchmarks do not adequately assess their ability to solve real-world tasks that require synthesizing information from multiple sources and inferring insights beyond simple fact retrieval. To address this, we introduce DEEPSYNTH, a novel benchmark designed to evaluate agents on realistic, time-consuming problems that combine information gathering, synthesis, and structured reasoning to produce insights. DEEPSYNTH contains 120 tasks collected across 7 domains and data sources covering 67 countries. DEEPSYNTH is constructed using a multi-stage data collection pipeline that requires annotators to collect official data sources, create hypotheses, perform manual analysis, and design tasks with verifiable answers. When evaluated on DEEPSYNTH, 11 state-of-the-art LLMs and deep research agents achieve a maximum F1 score of 8.97 and 17.5 on the LLM-judge metric, underscoring the difficulty of the benchmark. Our analysis reveals that current agents struggle with hallucinations and reasoning over large information spaces, highlighting DEEPSYNTH as a crucial benchmark for guiding future research.

Large vision-language models (VLMs) have evolved from general-purpose applications to specialized use cases such as in the clinical domain, demonstrating potential for decision support in radiology. One promising application is assisting radiologists in decision-making by the analysis of radiology imaging data such as chest X-rays (CXR) via a visual and natural language question-answering (VQA) interface. When longitudinal imaging is available, radiologists analyze temporal changes, which are essential for accurate diagnosis and prognosis. The manual longitudinal analysis is a time-consuming process, motivating the development of a training framework that can provide prognostic capabilities. We introduce a novel training framework LUMEN, that is optimized for longitudinal CXR interpretation, leveraging multi-image and multi-task instruction fine-tuning to enhance prognostic and diagnostic performance. We conduct experiments on the publicly available MIMIC-CXR and its associated Medical-Diff-VQA datasets. We further formulate and construct a novel instruction-following dataset incorporating longitudinal studies, enabling the development of a prognostic VQA task. Our method demonstrates significant improvements over baseline models in diagnostic VQA tasks, and more importantly, shows promising potential for prognostic capabilities. These results underscore the value of well-designed, instruction-tuned VLMs in enabling more accurate and clinically meaningful radiological interpretation of longitudinal radiological imaging data.

As LLM deployments scale over more hardware, the probability of a single failure in a system increases significantly, and cloud operators must consider robust countermeasures to handle these inevitable failures. A common recovery approach is to simply restart the LLM serving instance; however, this is costly in model-as-a-service (MaaS) inference settings, where reloading model weights and recompiling computation graphs can introduce significant delays to incoming requests. We propose ReviveMoE, a method for rapid failure recovery in large-scale LLM deployments without restarting the serving instance. ReviveMoE is designed to support both the traditional LLM architecture, which collocates MoE and attention on the same hardware, and the disaggregated architectures, which separate MoE from attention. Integrated into Huawei Cloud's MaaS, ReviveMoE is built on top of Huawei's xDeepServe serving platform and the XCCL communications library.

Understanding the complex, multi-agent dynamics of urban traffic remains a fundamental challenge for video language models. This paper introduces Urban Dynamics VideoQA, a benchmark dataset that captures the unscripted real-world behavior of dynamic urban scenes. UDVideoQA is curated from 16 hours of traffic footage recorded at multiple city intersections under diverse traffic, weather, and lighting conditions. It employs an event-driven dynamic blur technique to ensure privacy preservation without compromising scene fidelity. Using a unified annotation pipeline, the dataset contains 28K question-answer pairs generated across 8 hours of densely annotated video, averaging one question per second. Its taxonomy follows a hierarchical reasoning level, spanning basic understanding and attribution to event reasoning, reverse reasoning, and counterfactual inference, enabling systematic evaluation of both visual grounding and causal reasoning. Comprehensive experiments benchmark 10 SOTA VideoLMs on UDVideoQA and 8 models on a complementary video question generation benchmark. Results reveal a persistent perception-reasoning gap, showing models that excel in abstract inference often fail with fundamental visual grounding. While models like Gemini Pro achieve the highest zero-shot accuracy, fine-tuning the smaller Qwen2.5-VL 7B model on UDVideoQA bridges this gap, achieving performance comparable to proprietary systems. In VideoQGen, Gemini 2.5 Pro, and Qwen3 Max generate the most relevant and complex questions, though all models exhibit limited linguistic diversity, underscoring the need for human-centric evaluation. The UDVideoQA suite, including the dataset, annotation tools, and benchmarks for both VideoQA and VideoQGen, provides a foundation for advancing robust, privacy-aware, and real-world multimodal reasoning. UDVideoQA is available at https://ud-videoqa.github.io/UD-VideoQA/UD-VideoQA/.