arXiv:2607.01617v1 Announce Type: new Abstract: Large language model (LLM) serving increasingly combines prefill-decode (PD) disaggregation with tensor parallelism (TP) to support large models and long contexts. In conventional 2D/2.5D chiplet architectures, layer-wise prefill-to-decode KV-cache transfer decode-side TP collectives share the same lateral die-to-die (D2D) interconnect, creating mixed-traffic contention on the decode critical path. This contention increases communication latency, prolongs token generation intervals, and degrades end-to-end serving performance. We propose 3DLS, a logic-on-logic 3D-stacked chiplet architecture that separates traffic classes by routing KV-cache transfers through vertical interconnects while preserving decode-side TP collectives on the lateral D2D fabric. 3DLS achieves up to 1.49$\times$ throughput and 60.2\% lower end-to-end (E2E) latency over the shared-fabric planar baseline, and still achieves up to 1.17$\times$ throughput and 31.4\% lower E2E latency over a workload-aware priority-managed planar baseline. These results highlight that physical isolation is an important design principle for future chiplet-based PD-disaggregated LLM serving systems.
Science Journals
arXiv:2607.01602v1 Announce Type: new Abstract: SRAM-based FPGAs provide an attractive platform for energy- and latency-constrained CNN inference at the network edge, yet transient faults can lead to silent errors that compromise reliability. Always-on redundancy (e.g., full TMR) improves correctness but incurs substantial performance and energy overhead, while reactive recovery may introduce unacceptable latency on the critical path. We propose \textbf{ProWAFT}, a proactive workload-aware fault-tolerance framework for FPGA-based CNN accelerators that uses partial reconfiguration to selectively apply TMR across reconfigurable partitions. ProWAFT quantifies workload criticality, models fault propagation and reconfiguration overhead, and selects configurations that minimize a composite objective over latency, energy, and reliability risk. Implemented on a Xilinx Zynq UltraScale+ ZCU104 platform with six reconfigurable regions and evaluated on a 500-task trace derived from ResNet-18, MobileNetV2, and EfficientNet-Lite under time-varying SEU injection, ProWAFT achieves lower composite cost than static TMR and reactive reconfiguration while maintaining high task success rate and near-baseline throughput with low online decision overhead.
arXiv:2607.01605v1 Announce Type: new Abstract: CityGML, the OGC standard for modeling, storage, and exchange of semantic 3D city models, describes urban objects with detailed semantics, geometry, and topology. Yet this richness is difficult to query directly: CityGML's XML encoding is designed for exchange rather than analysis, and relational mappings expose it through schemas requiring expert knowledge. We present pykci (Python Knowledge Graph for Cities), an open-source system that transforms CityGML 2.0 datasets into a compact urban knowledge graph in Neo4j and makes it queryable in natural language. The graph schema covers all thematic feature modules of CityGML 2.0 across all levels of detail and is spatially indexed with an R-tree for efficient geometric retrieval. A complete end-to-end Python pipeline ingests CityGML datasets into the knowledge graph, exports them to OGC 3D Tiles for interactive visualization, and supports lossless round-trip export of all content back to CityGML. For querying, the graph is paired with a large language model through a model-agnostic text-to-Cypher mechanism: the graph schema is supplied as context, and the model translates natural-language questions into Cypher queries executed against the graph. We evaluate both a locally running open-weight model, which keeps sensitive city data on-premise, and a state-of-the-art commercial model for the most demanding spatial and semantic queries. Answers are grounded in exact city data rather than the model's parametric memory, reducing hallucination and providing auditable provenance for every response. We demonstrate the system on open-government CityGML LoD2 datasets from Hamburg, Germany, including complex semantic and spatial queries such as identifying roof surfaces suitable for greening. pykci enables urban planners, GIS practitioners, and citizens to interact with semantic 3D city models without expertise in query languages and database schemas.
arXiv:2511.20274v2 Announce Type: replace Abstract: In the paradigm of computer vision-based precise real-world scene understanding, joint reasoning in terms of contextual understanding about the objects present in a scene, their inter-object relations, and the action being performed is an essential prerequisite. However, prior works have not addressed all three jointly, and no large-scale dataset provides grounded annotations at all three levels across diverse visual scenarios. Hence, this work introduces the SCLARO (Scene-Contextual Localisation of Actions, Relations & Objects) dataset, consisting of 615,805 images spanning indoor, outdoor, and driving scenarios, annotated with global action captions, object bounding boxes, and relation triplets that supply structured scene context beyond a free-text caption. To benchmark the dataset, we propose ScenarioCLIP, a tri-level reference model that jointly encodes global scene context, objects, and inter-object relations using disentangled encoders and EMA-based knowledge distillation. We benchmark across a comprehensive suite of tasks on the SCLARO Dataset, namely zero-shot retrieval, linear probe, object detection, predicate classification, scene-graph classification, and out-of-domain generalisation. ScenarioCLIP's disentangled encoders improve over the previous works, such as PyramidCLIP's shared encoder, most notably at the object and relation levels and on out-of-domain generalisation. Code for the data generation pipeline and ScenarioCLIP is available at https://github.com/scenario-clip/SCLARO-ScenarioCLIP
arXiv:2511.22459v2 Announce Type: replace Abstract: We propose a method to reconstruct dynamic fire in 3D from a limited set of camera views with a Gaussian-based spatiotemporal representation. Capturing and reconstructing fire and its dynamics is highly challenging due to its volatile nature, transparent quality, and multitude of high-frequency features. Despite these challenges, we aim to reconstruct fire from only three views, which consequently requires solving for under-constrained geometry. We solve this by separating the static background from the dynamic fire region by combining dense multi-view stereo images with monocular depth priors. The fire is initialized as a 3D flow field, obtained by fusing per-view dense optical flow projections. To capture the high frequency features of fire, each 3D Gaussian encodes a lifetime and linear velocity to match the dense optical flow. To ensure sub-frame temporal alignment across cameras we employ a custom hardware synchronization pattern -- allowing us to reconstruct fire with affordable commodity hardware. Our quantitative and qualitative validations across numerous reconstruction experiments demonstrate robust performance for diverse and challenging real fire scenarios.
arXiv:2607.01642v1 Announce Type: new Abstract: Hardware-agnostic strategies for accelerating text-to-image diffusion, such as timestep distillation and feature caching, can reduce inference time without custom kernels or system-level optimization. Among them, multi-resolution generation strategies have recently received broad attention, attaining more than 5x speedup without any training. However, the design of performing upsampling in the latent space, together with the selective modification of partial regions, causes these methods to exhibit noticeable blurring or artifacts. To this end, we propose MrFlow, a training-free multi-resolution acceleration strategy for pretrained flow-matching models built upon a staged low-to-high-resolution pipeline. MrFlow first rapidly generates the main structure at low resolution, then performs super-resolution in the pixel space using a lightweight pretrained GAN-based model, subsequently injects low-strength noise to enable high-frequency resampling, and finally refines the details at high resolution. Quantitative and qualitative results on FLUX.1-dev and Qwen-Image show that MrFlow exploits the quadratic token reduction and reduced step requirement of low-resolution sampling to achieve 10x end-to-end acceleration while keeping OneIG within a 1% gap relative to that before acceleration, significantly surpassing other training-free acceleration strategies, and requiring no training or runtime dynamic identification whatsoever. MrFlow can further be directly combined orthogonally with pre-trained timestep distillation strategies, achieving even higher generation acceleration of up to 25x.
arXiv:2607.01654v1 Announce Type: new Abstract: We investigate volumetric reconstruction for compressive sensing light-sheet microscopy (CS-LSM), where fast volumetric imaging is achieved by encoding multiple axial planes into each camera exposure. To recover the underlying volume from highly multiplexed measurements, we propose a plug-and-play (PnP) framework that flexibly incorporates any user-specified denoiser into the reconstruction process. Building on a slice-based formulation, we further introduce an axial-coupled model that exploits correlations between adjacent slices to improve volumetric continuity. For efficient computation, we derive a Woodbury-based update for the data-consistency step in both the slice-based and axial-coupled formulations, and employ a Gauss-Seidel sweep for the denoising step in the axial-coupled model. Under a weakly convex regularization assumption, we establish subsequential convergence of the proposed algorithm. Experiments on synthetic and real zebrafish-heart data demonstrate that the proposed framework successfully recovers cellular structures from compressed measurements, and provide practical insights into the comparative performance of commonly used denoisers within the PnP framework under the CS-LSM setup.
arXiv:2606.01997v2 Announce Type: replace Abstract: Thermalization of light, where photons acquire a temperature and chemical potential analogous to a material gas, remains a striking yet experimentally elusive manifestation of quantum statistical physics. To date, it has been realized only in carefully engineered photonic environments that enforce repeated absorption-emission cycles. Here we show that such thermalization can emerge in a radically simpler setting: an open scattering medium. Using a pumped fluorescent dye solution doped with colloidal particles, we demonstrate that multiple scattering alone suffices to trap photons long enough to drive them to thermal equilibrium. The emitted radiation follows a Bose-Einstein distribution with a finite chemical potential, independently tunable via optical pumping, while its temperature is set by the host medium. A clear spectroscopic signature of thermalization is observed as a plateau at the sample temperature over a finite spectral range. Our results establish disordered scattering media as a generic platform for photon thermalization, extending this fundamental phenomenon beyond resonant cavities and opening new routes towards cavity-free photonic thermodynamics and thermal light sources operating under ambient conditions.
arXiv:2606.05778v4 Announce Type: replace Abstract: Traditional Image Aesthetic Assessment (IAA) methods mainly rely on regressing absolute Mean Opinion Scores (MOS). However, such a paradigm overlooks the inherently dynamic nature of human aesthetic perception, which relies on subconscious comparison against implicit visual references. Consequently, the lack of causal reasoning regarding aesthetic differences prevents models from learning generalizable aesthetic principles, thus limiting their generalization across diverse scenarios. In this work, we rethink the IAA task and propose Relative Edit-induced Difference Aesthetic learning (RED-Aes), a novel framework that leverages controllable image editing models to simulate the human aesthetic reasoning process. Instead of fitting absolute score distributions, RED-Aes explicitly learns the visual factors that drive aesthetic changes. To support this paradigm, we construct the RED-20k dataset, which comprises editing-based image pairs, quantitative aesthetic differences, and Chain-of-Thought (CoT) reasoning. Furthermore, we introduce a three-stage training strategy guided by a relative ranking consistency reward, optimizing the model solely via relative supervision. Extensive experiments demonstrate that RED-Aes achieves state-of-the-art performance on multiple public benchmarks, exhibiting superior generalization capabilities.
arXiv:2606.06324v2 Announce Type: replace Abstract: LLM agents increasingly rely on agent harness: the runtime infrastructure around the base model that defines execution environments, tool interfaces, context, lifecycle orchestration, observability, verification, and governance. Existing self-improving agents and automatic harness evolution methods mainly improve agents through runtime supervision, prompt optimization, workflow search, or harness modification based on final outcomes. However, they often fail to diagnose where the responsible evidence lies in failed trajectories and which harness implementation mechanism causes the unreliable behavior, resulting in broad, indirect, or poorly scoped changes. This paper proposes HarnessFix, a trace-grounded and diagnosis-driven framework for repairing agent harnesses. HarnessFix compiles raw execution traces and harness artifacts into a Harness-aware Trace Intermediate Representation (HTIR), which normalizes fragmented trajectory evidence, captures step-level data-flow and control-flow relations, and aligns runtime steps with the harness artifacts that shape their behavior. It then attributes failures to responsible steps and harness artifacts, and consolidates recurring diagnoses into repair-oriented flaw records. Finally, HarnessFix maps these records to scoped repair operators, generates patches under flaw-specific repair specifications, and accepts them through regression-aware validation. We evaluate HarnessFix on four popular benchmarks, and results show that it improves the performance over the initial harnesses by 6.3% to 18.4%, significantly outperforming human-designed and self-evolution baselines. HarnessFix highlights the value of treating failed trajectories not only as feedback signals, but also as structured evidence for diagnosing and repairing the harness mechanisms behind agent failures.
arXiv:2606.07439v2 Announce Type: replace Abstract: We present a 65-nm risk-aware multimodal Bayesian inference engine for privacy-preserving, fully on-device skin lesion screening under uncontrolled at-home conditions. The proposed compute-in-memory architecture performs in-word Mixture-of-Gaussian sampling, improving uncertainty modeling beyond conventional unimodal Bayesian neural networks. This added probabilistic expressiveness increases equal-risk operating coverage by 1.4x, improves robustness to user-data perturbations by >1.5x, enhances process-variation resilience by 5.5x, and improves balanced accuracy by 1.8% over state-of-the-art unimodal Bayesian neural networks. Hardware robustness is further supported by calibration-free Gaussian random-number generation using complementary process variation, achieving 16.3 fJ/sample and 168.6 GSa/s/mm^2 efficiency. These results demonstrate a practical, energy-efficient, and risk-aware edge-AI solution for privacy-conscious medical screening.
arXiv:2607.01657v1 Announce Type: new Abstract: Visual recognition models often fail when deployed in new environments. Domain Generalization (DG) addresses this by learning representations that remain invariant to environment-specific variations. Recent approaches increasingly rely on large vision-language models, assuming that preserving their expressive visual representations improves robustness. However, we show that such visual expressiveness can instead propagate spurious cues that tie representations to the training environments, hindering invariant learning. We therefore discard visual guidance and instead treat the language embedding space as the primary source of domain invariance, naturally acting as an information bottleneck that preserves core semantics while suppressing domain-specific variations. Extensive experiments across diverse backbones exhibit state-of-the-art performance and further analyze what makes guidance effective for robust generalization. These findings shift the focus of DG from improving representations to designing supervision that enforces invariance.
arXiv:2607.01664v1 Announce Type: new Abstract: Substrate-agnostic perspectives are currently attracting increased attention. For example, it has become customary to refer to agnostic biosignatures to reflect the range of alternative extraterrestrial biospheres and to account for the deeper philosophical dependence of candidate biosignatures on the underlying theory of life. Analogously, one can formulate a concept of agnostic technosignatures, reflecting that the more we expand the search for technosignatures, the more we invite theories of technology that undo the terrestrial bias. For this reason, this paper argues that there exists a strong theoretical justification for an integrated study of technosignatures and biosignatures, articulated in a unified perspective on substrate-agnostic ecologies. The paper introduces the concept of substrate-agnostic ecology as an abstraction unconstrained by terrestrial circumstances, anchored instead in a functional understanding of agent-ecology coupling provided by niche construction theory.
arXiv:2607.02365v1 Announce Type: new Abstract: We compute the weight distribution of the third-order Reed--Muller code RM(3,11) of length 2048. The weight enumerator is assembled from the coset weight enumerators of f+RM(2,10), evaluated for representatives of all 3691560 nonzero GL(10,2)-orbits of Boolean cubic forms in ten variables. The computation rests on a structural theorem: a nondegenerate Boolean cubic form admits a nondegenerate hyperplane restriction, except for a single orbit in each odd dimension. The same pass determines the second-order nonlinearity of every cubic form: the relative covering radius of RM(2,10) in RM(3,10) is 408, attained on 179 orbits. This raises the best known lower bound on the covering radius of RM(2,10) from 400 to 408. A complementary heuristic search shows that the relative covering radius of RM(6,10) in RM(7,10) is at most 32, improving the previous bound of 50.
arXiv:2603.27313v2 Announce Type: replace Abstract: Disturbance observer-based control has shown promise in robustifying robotic systems against uncertainties. However, tuning such systems remains challenging due to the strong coupling between controller gains and observer parameters. In this work, we propose MetaTune, a unified framework for joint auto-tuning of feedback controllers and disturbance observers through differentiable closed-loop meta-learning. MetaTune integrates a portable neural policy with physics-informed gradients derived from differentiable system dynamics, enabling adaptive gains across tasks and operating conditions. We develop an adjoint method that efficiently computes the meta-gradients with respect to adaptive gains backward in time to directly minimize the cost-to-go. Compared to existing forward methods, our approach reduces the computational complexity to be linear in the data horizon. On quadrotor control tasks, MetaTune achieves competitive or improved tracking performance while reducing gradient computation time by more than 50\%. In PX4-Gazebo hardware-in-the-loop simulation, the learned policy transfers zero-shot and reduces tracking RMSE by about 15--20\% in aggressive flight and up to 40\% under strong disturbances.
arXiv:2606.07571v2 Announce Type: replace Abstract: Key-value (KV) caching for shared prefixes is essential for high-throughput large language model (LLM) serving, but it faces critical challenges in emerging diffusion language models (DLMs). In DLMs, bidirectional attention means that updating any token dynamically alters the entire context and its corresponding KVs. Thus, existing caching techniques developed for LLMs, which assume that KVs remain invariant once computed, corrupt the shared prefix KVs. Our experiments show that applying these techniques to DLMs causes model accuracy to collapse to near zero. To unlock high-throughput DLM serving, we propose bidirectional prefix caching, bicache, the first KV caching technique for shared prefixes in DLMs. bicache is designed based on key observations from our comprehensive analysis: shared prefix KVs remain stable and reusable in shallow layers, while the depth of shallow layers depends on the fraction of shared prefix tokens in each request. Thus, bicache dynamically identifies a safe layer depth for reusing shared prefix KVs and eliminates redundant computation. Evaluations demonstrate that bicache significantly improves serving throughput by 36.3%-98.3% compared to existing techniques without accuracy collapse (only 0-1.8% difference).
arXiv:2606.08998v3 Announce Type: replace Abstract: Agentic AI systems can behave differently across runs: the same request may produce a different plan, a different tool call, a different code edit, or a different final answer. Such variability arises from several layers that are often conflated. At the core of many current agents is a foundation model, a large pretrained model adaptable to many downstream tasks, embedded in an orchestration loop that plans, calls tools, observes results, and updates state. One explicit intrinsic source of variability in such systems is token generation: the model computes scores over possible next tokens, the scores are converted into probabilities, and a decoder may sample tokens using a pseudo-random number generator. A small sampled token difference can then cascade downstream into a different tool call, code path, search query, or agent state. Other sources of variability are extrinsic to token sampling, including changing environments, live data, serving infrastructure, batch effects, and numerical details. By separating these layers, this tutorial clarifies what it means to call agentic AI systems stochastic, when such variability can be reproduced under matched conditions, and why deterministic execution need not imply identical behavior in deployed settings.
arXiv:2606.10822v3 Announce Type: replace Abstract: Aerial search and rescue missions require fast and reliable victim detection under uncertain and rapidly changing environments. Deterministic deep learning models can produce overconfident false positives, forcing unmanned aircraft systems to perform costly verification maneuvers that reduce search coverage and increase rescue delay. Bayesian neural networks provide uncertainty-aware detection, but their sampling overhead is challenging for battery-constrained edge platforms. This work presents a FeFET-based Bayesian inference engine with a write-free central limit theorem Gaussian random number generator embedded in a compute-in-memory macro. By summing currents from a randomly selected subset of minimum-sized, programmed-once FeFETs, the proposed architecture eliminates energy- and endurance-intensive write operations during inference while maintaining scalable Gaussian sampling. The CLT-GRNG consumes 640 aJ per sample, providing a 560x energy-efficiency improvement over prior BNN accelerators, while the CIM tile achieves 185 TOPS/W/mm2. Evaluated on aerial search and rescue detection, the Bayesian model improves uncertainty calibration and robustness under environmental corruption, reducing risk and enabling low-confidence detections to be filtered before costly verification. These results demonstrate an energy-efficient and uncertainty-aware edge AI engine for autonomous search and rescue systems.
arXiv:2606.11357v2 Announce Type: replace Abstract: With the growing demand for on-device LLM inference, edge SoCs increasingly integrate NPUs to improve performance and energy efficiency under tight power and thermal budgets. However, practical LLM deployment on current client NPUs remains difficult: widely used quantization formats such as AWQ do not map cleanly onto many existing NPU software stacks, which are often proprietary and expose limited low-level control. In this work, we present TileFuse, a close-to-metal mixed-precision kernel library for AMD XDNA2 NPUs that targets GEMM/GEMV-based operators in quantized LLM inference. TileFuse brings practical low-bit formats such as AWQ-style W4A16 and W8A16 directly onto XDNA2, rather than forcing the model to be reshaped around an NPU-specific quantization scheme. TileFuse co-designs weight layout, metadata placement, mixed-precision microkernels, and array-level dataflow. Specifically, it fuses unpacking, dequantization, and GEMM/GEMV execution into a single kernel flow, introduces an interleaved pre-tiling layout that supports GEMM dimensions up to 32K, and redesigns GEMV dataflow to utilize the full 4x8 AIE array. Across kernel-level evaluations, TileFuse improves performance by up to 121.6% for GEMM and 281% for GEMV over full-precision baselines, while delivering more than 2x performance and energy-efficiency gains over strong iGPU baselines on GEMM. In end-to-end LLM experiments on Ryzen AI laptops, TileFuse achieves up to 2.0x lower prefilling latency with more than 64.6% lower energy consumption. Together, these results show that XDNA2 is a practical target for AWQ-style edge LLM inference and that native NPU support for off-the-shelf quantization can make NPUs substantially more usable in real client deployments.
arXiv:2606.13723v2 Announce Type: replace Abstract: Intersection-over-Union (IoU), as a pivotal metric for evaluating the spatial alignment between candidate proposals and ground-truth annotations, directly determines the quality of positive sample sets and the training efficacy of visual detection models. Through theoretical modeling and analysis, we uncover a non-sensitive region on the IoU response curve, within which samples yield nearly identical IoU scores despite distinct geometric overlaps. To overcome this limitation, we introduce a set of morphological similarity metrics covering area, shape, and aspect ratio, to refine the positive sample assignment process, thereby ensuring more discriminative and reliable matching. A supplementary matching score is derived via mean-based aggregation of these multidimensional similarities, compensating for the intrinsic limitation of IoU in representing structural correspondence. Theoretically, incorporating morphological similarity reshapes the response distribution of the matching function, yielding both effective directional gradients and polygon-like iso-response contours, which tightly confine high-response regions around each ground-truth instance and substantially enhance the precision of positive sample selection. Experiments based on the YOLOv9 framework demonstrate consistent performance gains on both NEUDET and GC10- DET datasets. Notably, the proposed approach is fully plug-and-play and incurs zero additional inference overhead, thereby ensuring deployment efficiency for industrial visual inspection.
arXiv:2606.13732v2 Announce Type: replace Abstract: The proliferation of recursive training on synthetic data can alleviate data scarcity but risks model collapse, where repeated training erodes distributional tails and homogenizes outputs. Data selection is widely viewed as a remedy, yet its reliability depends critically on the reference distribution used by the verifier. We show that in low-resource verification regimes, where each verifier observes only a small, fragmented, and biased slice of the target manifold, selection itself becomes biased. This situation naturally arises in low-resource data silos such as healthcare consortia or proprietary financial institutions, where raw data cannot be pooled and local references are inherently incomplete. As a result, selection preferentially retains samples aligned with the local manifold while pruning globally relevant tail modes, turning from a safeguard against collapse into a mechanism that precipitates it. We theoretically prove that such siloed selection accelerates collapse and induces power-law diversity decay. As an initial mitigation, we construct Wasserstein proxy references from multiple silos without sharing raw data. Empirical results confirm that local-reference selection fails on skewed distributions, whereas collaborative proxy references mitigate diversity degradation, suggesting that recursive synthetic-data pipelines require particular caution when real-data coverage is fragmented or scarce.
arXiv:2606.18411v2 Announce Type: replace Abstract: Cholesky-QR is among the fastest algorithms for computing the thin QR factorization of tall-and-skinny matrices on GPUs, relying entirely on BLAS-3 operations. However, it is numerically unstable: forming the Gram matrix squares the condition number, causing breakdown when $\kappa_2(\boldsymbol{A}) \gtrsim 10^8$. We present MRCQR (Mixed-Precision Randomized Cholesky-QR), a stable GPU algorithm that addresses this limitation. MRCQR uses a subsampled randomized trigonometric transform to construct a preconditioner $\boldsymbol{R}_s$ that reduces $\kappa_2(\boldsymbol{A}\boldsymbol{R}_s^{-1})$ to near unity with high probability, then applies Cholesky-QR in double precision to the preconditioned matrix. The key insight -- supported by perturbation analysis -- is that the preconditioner requires far less accuracy than the final result: single (FP32) precision suffices when $\kappa_2(\boldsymbol{A}) \lesssim 10^8$, and half (FP16) when $\kappa_2(\boldsymbol{A}) \lesssim 10^4$. MRCQR produces an explicit orthogonal factor $\widehat{\boldsymbol{Q}}$ satisfying $\|\boldsymbol{I} - \widehat{\boldsymbol{Q}}^\top\widehat{\boldsymbol{Q}}\|_2 = \cal O(\mathbf{u})$ ($\mathbf{u} \approx 10^{-16}$, double-precision unit roundoff) for condition numbers up to $10^{16}$, far beyond the $10^8$ limit of CholQR2. Experiments on an NVIDIA H100 GPU show that MRCQR (FP16) outperforms rand-cholQR by $1.4$--$1.8\times$ across all tested column counts and is $1.8$--$13.5\times$ faster than cuSOLVER geqrf, while the FP16 sketch (used when $\kappa_2(\boldsymbol{A}) \lesssim 10^4$) is $2\times$ cheaper than FP64 at no accuracy cost.
arXiv:2606.23041v2 Announce Type: replace Abstract: Multimodal Large Language Models (MLLMs) have achieved remarkable success in visual understanding but remain constrained in visual generation due to the fundamental feature discrepancy between semantic perception and pixel-level reconstruction. Bridging this gap requires overcoming two core challenges: endowing semantic encoders with high-fidelity reconstruction capabilities, and effectively aligning generative models with semantic spaces without relying on external teachers. To this end, we propose a novel unified multimodal framework featuring \textbf{S}emantic-\textbf{P}ixel self-alignment and \textbf{A}daptive \textbf{R}outing (\textbf{SPAR}). First, to reconcile semantic perception with pixel-level reconstruction, we introduce an asymmetric dual-stream unified tokenizer. A lightweight semantic stream anchors discriminative features, while a Transformer-augmented pixel stream recovers fine-grained visual details into a unified compact latent space. Second, to eliminate external dependencies, we propose a self-aligned generation paradigm that natively leverages this optimized tokenizer as an internal alignment teacher for the diffusion model. Furthermore, to facilitate flexible multimodal interaction within this unified space, we introduce Dynamic Token Routing, which enables each token to adaptively aggregate multi-layer MLLM features based on its distinct semantic demands. Extensive experiments demonstrate that SPAR establishes the state-of-the-art for unified architectures, achieving exceptional generation and reconstruction quality while preserving foundational visual understanding capabilities.
arXiv:2606.23370v2 Announce Type: replace Abstract: Device-side Large Language Models (LLMs) have grown explosively, offering stronger privacy and higher availability than their cloud-side counterparts. During LLM inference, both the model weights and the user data are valuable, and attackers may compromise the OS kernel to steal them. ARM TrustZone is the de facto hardware-based isolation technology on mobile devices, used to protect sensitive applications from a compromised OS. However, protecting LLM inference with TrustZone incurs significant overhead to both the secure inference and the normal aplications, due to two challenges: the inflexible resource isolation and the inefficient secure resource management. To address these challenges, this paper presents FlexServe, a fast and secure LLM inference system for mobile devices. The key idea is to decouple the access permission from the management permission of secure resources, so that the normal-world OS cannot access them but can still manage them as usual. First, FlexServe introduces a Recallable Resource Isolation mechanism to construct Recallable Secure Memory (Flex-Mem) and a Recallable Secure NPU (Flex-NPU). They can only be accessed by the secure world, but can be efficiently allocated and reclaimed by the normal-world OS. Based on them, FlexServe further introduces a FlexServe Framework to run secure LLM inference in the secure world. It works together with the normal-world OS to perform cooperative secure memory management. We implement a prototype of FlexServe and compare it with two TrustZone-based strawman designs. The results show that FlexServe achieves average TTFT speedups of 10.05X over the strawman and 2.44X over an optimized strawman.
arXiv:2606.24564v5 Announce Type: replace Abstract: Reconstructing realistic, physically plausible garments from a single image remains a fundamental challenge. Template-free methods capture surface geometry but lack explicit sewing structure for simulation; while programmatic systems are simulation-ready but constrained by predefined templates. This reveals a fundamental representation gap between geometric reconstruction and structured garment construction. We present PatternGSL, a structured garment representation in the form of a template-free and learnable specification language that encodes complete sewing patterns, including panel boundaries, parameterized seams, and explicit stitch topology, in a compact and standardized form. PatternGSL preserves the physical rigor of pattern-based models while removing template dependence, elevating sewing structure as a first-class target for generative modeling. We further propose a vision-language framework that predicts PatternGSL specifications directly from a single image and decodes them into garments using lightweight deterministic validity handling, without optimization-based refinement or manual cleanup. In addition, we introduce PatternGSLData, the first large-scale image-to-GSL paired dataset comprising 300K samples with complete sewing pattern annotations, enabling supervised VLM training for structured garment reconstruction. Experiments demonstrate improved pattern accuracy over prior baselines, explicit sewing-structure recovery, reliable cloth simulation, and pattern-level editing through the same deterministic decoding pipeline. Code and data-processing scripts will be released at https://lagrangeli.github.io/PatternGSL/.