mobile device developers

Abstract: Modern automatic speech recognition (ASR) systems have achieved remarkable accuracy by scaling model depth and capacity, but at the cost of increased memory, energy, and computational demands. On edge devices where ASR is often most needed, such as watches and glasses, deploying such large models is infeasible due to extremely constrained resources. This raises a fundamental question: can we achieve high representational power in deep ASR models without scaling up parameterization? In this work, we revisit the role of depth and identify layer wise representational dynamics, in which most layers learn functionally similar transformations. Motivated by this insight, we propose a block-recurrent ASR architecture that replaces parameterized depth with a small set of recurrent blocks, each consisting of weight-shared layers, thereby preserving effective depth while drastically reducing model size. Through representation-guided grouping and knowledge distillation, block-recurrent models retain the accuracy of large reference models while using an order of magnitude fewer parameters. Across Open ASR Leaderboard benchmarks, a model with only two shared blocks recovers 97.4% of the accuracy of large models on average. The compact backbone further enables efficient on-device personalization through lightweight MoE-LoRA adaptation, allowing user-specific models to match or surpass the accuracy of general-purpose ASR systems with significantly lower inference cost. By reusing weights across recurrent steps, the proposed architecture reduces memory-access overhead and inference energy consumption. Together, these results show that high quality ASR can be achieved without large parameterization, and that structured recurrence provides a principled path towards ultra-compressed, energy-efficient, and personalized speech recognition on the edge.

Abstract: Large language models (LLMs) increasingly run on edge devices alongside tasks that continuously process sensor data. In many applications, these foreground tasks prepare observations as inputs to the LLM, whose outputs guide subsequent processing or actions. The observations must be processed by foreground deadlines, and LLM feedback must arrive while its source observations remain useful. Yet the CPU and accelerators share a unified-memory system, so LLM inference can delay foreground processing through memory-bandwidth contention even when compute resources remain available. We present EdgePace, a bandwidth-aware runtime that protects foreground deadlines while sustaining timely LLM feedback. EdgePace creates safe suspension points across prefill and decode, monitors foreground memory demand and stalls at low cost, and schedules inference according to contention risk, deadline slack, and output freshness. We evaluate EdgePace on a OnePlus 13 and three NVIDIA Jetson platforms using multi-camera processing and robotic target tracking. Relative to unmanaged inference, it reduces foreground deadline misses by 24.1%--96.1% while retaining 64.9%--98.5% of LLM throughput. All measured LLM outputs arrive within their task-defined freshness windows. On OnePlus 13, for example, camera deadline misses fall by 91.3% while 90.4% of LLM throughput is retained.

Abstract: DeepInfer-Mac runs the 750B-class DeepSeek-V4.1-Flash model on a MacBook Air with 24 GB of unified memory. A mixed low-bit conversion occupies 338 GiB on the internal SSD. About 7.6 GiB of weights remain resident; routed experts and Engram rows are loaded when needed. This separation makes the model executable, but it leaves each generated token waiting for a succession of storage reads. We organize expert weights so that reads fill reusable GPU buffers directly, and couple cache allocation with the lifetime of those buffers. In end-to-end runs, the system generates English, Chinese and code at 0.21-0.72 tokens per second, with a largest sampled process footprint of 16.73 GiB across the tested requests. Matched decode comparisons show a 21.0% throughput gain from next-layer prefetching over resident-expert overlap alone, at the cost of additional reads. Longer prompts remain costly. We describe the design, provide the implementation and measurement scripts, and identify the limits of this operating point: a model far larger than memory can run locally, while storage and prefill still govern the user's wait.

Abstract: Users often exhibit changes in interaction behavior before explicitly seeking assistance, enabling systems to anticipate future assistance use. This is especially important in high-stakes environments, where delayed assistance can have operational, financial, or safety consequences. However, conventional Large Language Models (LLMs) are often unsuitable for continuous on-device behavioral prediction because of their computational and memory demands. Behavioral patterns are also highly individual and device-specific, motivating local processing that preserves privacy and data sovereignty. Small Language Models (SLMs) offer a promising alternative, but their ability to reason reliably over low-level sequential interaction events remains underexplored. This study introduces Evidence-Grounded Local Behavioral Prediction (ELBP), a four-layer architecture combining deterministic interaction processing, dual-view behavioral representation, SLM-based prediction, and evidence verification with abstention. Three instruction-tuned models of 3B parameters or fewer were evaluated using raw, structured, dual-view, and complete ELBP configurations. The best-performing model, Qwen2.5-3B with ELBP, achieved a full-set F1-score of 0.80 and PR-AUC of 0.84, representing a 15.9\% relative improvement over raw-event prompting and 17.6\% over XGBoost, while full-set ECE decreased by 43.2\% relative to raw-event prompting. With evidence verification and abstention, selective F1 increased to 0.84 at 91\% coverage and evidence support reached 96\%. These results show that structured behavioral representation and evidence-verified selective prediction can improve the reliability of behavioral forecasting in SLMs.

Abstract: Edge-cloud split inference runs the early layers of a multimodal large language model (MLLM) on a device and sends intermediate features to a server to complete the task. While keeping raw images local is often perceived as a privacy benefit, an honest-but-curious server can still infer sensitive attributes, such as age, gender, and race, from the received features. Existing protections intervene at the input, transmitted representation, or model's responses, but their native evaluations rely on final-response refusals or fixed-classifier failures, which underestimate an adaptive server's capabilities. To address this, we propose a label-budgeted attribute-inference framework that measures representation-level privacy through a recovery curve: the share of the unprotected leakage the server recovers as a function of its labeled-sample budget, from zero (a classifier trained on unprotected features) to a full labeled set. Our evaluation across two popular MLLMs reveals two critical findings. First, in unprotected pipelines, intermediate features leak all three attributes regardless of the split depth. Second, even when protection mechanisms are deployed across different stages, sensitive information remains recoverable under limited label budgets. For example, Safe-LLaVA removes biometric content from the model's responses, but from the features it transmits in split inference, 20 labeled samples per class recover 75-85% of the unprotected leakage. This work provides a label-budgeted measure of feature-level privacy for split MLLM inference and motivates defenses that act on the transmitted representation.

Abstract: Search is a central component of Automatic Speech Recognition (ASR) systems and has been deeply investigated for classical ASR modeling based on Hidden Markov Models (HMMs). However, in work on current sequence-to-sequence ASR architectures it is often simplified to a small fixed-size beam search without external Language Model (LM). In this work, we present a comparative analysis of search space complexity and efficiency for Connectionist Temporal Classification (CTC) and neural transducer models under different pruning strategies and LM configurations. We show that these end-to-end (E2E) models work with a small search space most of the time but exhibit large temporary peaks in hypothesis count during periods of uncertainty. Score-based pruning dynamically adapts the search to model uncertainty, leading to a significantly lower Real-Time Factor (RTF). Our results highlight the limitations of search with low fixed-size beam pruning and provide practical guidelines for efficient search in E2E ASR.

Abstract: Visual perception should be iterative and task-directed, with representations refined in response to the question and the need for further processing. While Vision-Language Models (VLMs) increasingly support extended reasoning, their visual tokens are still typically acquired in a single, task-agnostic encoder pass. This creates an evidence-acquisition bottleneck: models may reason longer without refining the visual evidence needed to answer the question. We propose LoopLook, a lightweight encoder-side recurrent perception framework that equips pretrained VLMs with adaptive visual depth. LoopLook recurrently revisits a middle-to-late window of the vision encoder, injecting question representations at each loop to guide visual refinement. A learned stopping mechanism adaptively determines the recurrent depth, while a token-wise Mixture-of-Looks (MoL) combines representations from different loops, allowing each visual token to selectively integrate information acquired across recurrent steps. Across four backbones and eight benchmarks, LoopLook consistently improves performance over the baseline, yielding gains of up to 8.6% on MMStar and 8.1% on HallusionBench. Our analysis shows that LoopLook progressively refines visual attention and adaptively allocates depth according to task difficulty. It also better preserves task-relevant visual information under reduced input resolution, while increasing time to first token (TTFT) by only 5.1-6.5%. By bringing recurrence from reasoning back into perception, LoopLook transforms visual perception from a one-shot glimpse into an adaptive process of recurrent refinement.

Abstract: Personalized federated learning must decide which clients should share representations. We introduce HDC-FL (Head-Derived Clustering for Personalized Federated Learning), which reads that decision off a signal shared-encoder training already produces: the client-specific classifi- er heads. At periodic refresh rounds the server projects the cached heads into a normalized low-rank descriptor space, groups the descriptors, builds one encoder prototype per group from the cached client encoders, and returns to each client a soft mixture of those prototypes paired with its own unchanged head. Client-specific heads are never averaged or distributed to oth- er clients; the shared-encoder trajectory is left intact; and no second optimization trajectory, additional local objective, or post-training fine-tuning stage is required. Across 11 CIFAR-10 and CIFAR-100 configurations covering label-subset, Dirichlet and rotation heterogeneity with ResNet-18 and MobileNetV2 encoders, HDC-FL is competitive with the strongest personalized and clustered baselines in macro-averaged mean accuracy and attains the numerically highest macro-averaged worst-client and bottom-20% accuracy among the evaluated methods. The tail results distinguish methods that look similar when only their means are compared. A controlled donor-swap diagnostic further finds positive donor-ranking associations between head similarity and encoder compatibility in all three settings tested, most clearly under covariate shift.

Abstract: Consider a standard multi-hop wireless radio network in which nodes relay messages for one another. While the contents of messages may be sensitive and thus encrypted, the very fact that two radios communicated can leak sensitive information about both the physical location of the origin and the physical location of the destination. We investigate the trade-off between protecting the privacy of a user's current location and the privacy of their location of interest, and how the user should choose dummy destinations to obfuscate their queries.

Abstract: Learning inertial odometry for a new platform usually requires ground-truth trajectories. We show that an inertial odometry model can instead be trained from scratch by a myopic teacher that integrates the IMU over each short window independently, a scheme we call integration self-supervision.Interestingly, although supervised only by short-window velocity changes, the student learns to directly predict velocity, and does so more accurately than its integration teacher. Pose graph distillation then refines the model with sequence-consistent targets from a pose graph over many windows. Training on a new platform therefore no longer requires collecting ground-truth trajectories, but only calibrating the IMU and recording data during normal operation. Across six datasets spanning pedestrians, drones, quadrupeds, and a humanoid, the same recipe with no velocity or position labels trains models that come within $0.07$~m/s average velocity error of a same-architecture supervised baseline.

Abstract: Integrated sensing and communication (ISAC) has been recognized as a critical technology for the sixth generation (6G) wireless networks. Under monostatic sensing mode, achieving a balance between communication and sensing performance via optimal beamforming is a critical research problem. In multiple input multiple output (MIMO) systems, such beamforming problems are of substantial computational complexity due to the large number of antennas. Existing methods exploit exact subspaces of lower dimensions to speed up the beamforming process where the complexity is proportional to the number of communication users and sensing targets instead of antennas. In this paper, it is demonstrated that the beamforming problem can be solved in an even lower-dimensional subspace with only a small sensing performance loss. Numerical results compare the sensing performance and computational complexity of ISAC beamforming problem in the exact subspace and smaller subspaces.

Abstract: Recent speech brain-computer interfaces (BCIs) increasingly rely on large models and complex multi-stage pipelines for neural decoding. However, in speech BCI, where training data are scarce, noisy, and expensive to collect, the benefits of model scaling remain unclear. This work revisits a fundamental question: are large models truly necessary for accurate neural decoding? We propose BEST (Brain-Encoding-Speech-Text), a two-stage framework for efficient neural-to-text decoding. Instead of relying on large generative models or phoneme-centric pipelines, BEST introduces hierarchical intermediate representations and a compact speech-oriented decoder that directly maps neural activity to text. This design preserves essential structure while significantly reducing computational cost, enabling robust generalization in low-resource settings and practical deployment under constrained resources. On the Brain-to-Text '24 and '25 benchmarks, BEST achieves word error rates (WERs) of 6.19% and 1.62%, respectively, using only 3.85% of the parameters required by comparable methods. Beyond these results, we provide three key insights: (1) decoder scaling yields non-monotonic gains; (2) cross-modal alignment is more critical than model size; and (3) overfitting degards sequence-level performance. Representation analysis further reveals a consistent neural-audio-text alignment pathway.

Abstract: Retrieval-augmented generation (RAG) on edge devices enables personalized large language model (LLM) inference while preserving privacy. However, RAG's inherent length asymmetry makes the prefill phase the dominant bottleneck. While precomputing the KV caches offline alleviates the prefill cost on server systems, the multi-gigabyte caches over the narrow interconnect create a severe I/O wall for on-device systems. To mitigate this, we propose MemLLM, an end-to-end in-storage processing (ISP) architecture that executes the complete attention pipeline based on precomputed KV caches within the SSD. MemLLM utilizes per-channel accelerators to fully utilize the aggregated NAND channel bandwidth and accelerates all KV cache-related computations on the storage-side to minimize PCIe transfers. To efficiently handle the sparse updates required by RAG, MemLLM introduces a fused recomputation-query kernel paired with a hardware-based filter logic to halve the NAND read volume and eliminate write-back penalties. Furthermore, MemLLM's global SRAM-based stateful reduction engine enables a completely DRAM-less dataflow, thereby removing the internal memory bottlenecks and bounding the PCIe egress to an $O(1)$ footprint. A full-stack software stack allows asynchronous, zero-copy transfers from the ISP to the host and provides transparent execution within modern ML frameworks. Evaluated across eight real-world RAG datasets and six on-device LLM models, MemLLM achieves an average end-to-end speedup of 4.06$\times$ over the GPU baseline.

Abstract: How long should each client train before communicating? Federated learning usually answers with a fixed, hand-tuned number of local epochs, yet an extra epoch helps only while the loss it removes outweighs the client drift it adds. We cast local training as finite-horizon optimal stopping: a client trains while the next epoch's expected loss reduction exceeds its priced squared drift. We prove that this one-stage look-ahead threshold is optimal whenever stopping decisions are nested, derive closed forms for geometric and power-law returns, and show that an estimated rule loses utility only through wrong-signed decisions. Epoch-level evidence supports the model's structure (geometric diminishing returns and additive drift) and reveals a simple regularity: the stage of training, not the client, sets the budget. The resulting rule, FedStop, is never more than one point below tuned FedAvg in mean accuracy. Where its budgets vary, it raises worst-decile accuracy by 4-8 points over tuned FedAvg and an adaptive-interval baseline, and every one of these gains remains significant after a Holm correction. Against the client-side stopping rules ALT and FLASH, it has the highest CIFAR-10 worst decile on every seed and is level elsewhere. It stays stable where per-client early stopping collapses, and each stopping decision takes constant time.

Abstract: Existing federated data condensation methods face privacy risks, as uploaded synthetic data may preserve sensitive information and remain vulnerable to reconstruction attacks. Moreover, their generation often incurs considerable client-side computational overhead, limiting deployment on resource-constrained devices. To address these issues, we propose FedSumUp, which employs a learnable lossy compressor on clients to generate compact and visually interpretable representations. By controlling information capacity, FedSumUp reduces privacy leakage while allowing users to directly assess the privacy-utility tradeoff. We further introduce a dual-objective optimization mechanism to preserve task-relevant semantics while suppressing reconstruction risks. The compressor is trained on the server and only performs lightweight inference on clients. Since the server cannot access raw data, we develop a data-free closed-loop optimization strategy that optimizes the compressor by minimizing the discrepancy between original and twice-compressed representations after structure-preserving upsampling. Experiments on BloodMNIST, CIFAR-10, and ChestMNIST demonstrate reconstruction PSNR as low as 8.56 dB (with synthetic PSNR as low as 5.45 dB), client runtime reductions up to 96.5% (from 778s to 27s on ChestMNIST), and competitive classification accuracy (89.72% ChestMNIST, 82.24% BloodMNIST, 65.63% CIFAR-10), while remaining inherently robust to non-IID data distributions.

Abstract: Vision-language models (VLMs) are moving onto edge devices, but little is known about how much energy they draw on single-board computers with NPU or GPU accelerators, or where that energy goes. We measure total-system energy, latency, and thermal behavior of Qwen2-VL-2B-Instruct and its text-only sibling on a Raspberry Pi 5 with a Hailo-10H NPU and on a Jetson Orin Nano Super, on both CPU and accelerator, and separate the fixed energy of each request from the energy that grows with output length. Because the accelerators use different quantization and serving software, we compare deployment stacks rather than silicon. Three findings stand out. First, NPU offloading makes the Pi a much more efficient VLM platform: it cuts energy per generated token by about two-thirds, leaves the host CPU largely free, and avoids the thermal throttling that the CPU backend reaches in sustained use. Second, an image adds both fixed and per-token energy, but at typical response lengths generating the answer costs far more than processing the image. Third, the Jetson GPU favors short responses through its lower fixed cost per request, whereas the Pi NPU favors longer ones through its lower cost per output token; their energy curves cross near 40 output tokens.

Abstract: Feature distillation is usually expected to add to logit distillation. We find that this expectation does not hold for two common feature losses in compact multilingual automatic speech recognition (ASR), and analyze why through their gradients. The gradient of a projection-based loss is orthogonal to the task gradient, while that of Centered Kernel Alignment (CKA), a similarity-based loss, is weakly aligned with it. CKA lowers the word error rate (WER) on non-English languages, but at the expense of data-rich English. Three loss-level remedies (one trained, two probed) all fall short, and the CKA gradient overlaps with the subspace English uses in the shared encoder by 97.5% to 99.5% (about 81% by chance). Guided by this diagnosis, we propose CAFE (Confined-Adapter Feature distillation), which confines the CKA gradient to adapters that English bypasses, addressing the trade-off in the parameters rather than in the loss. On the Open ASR Leaderboard, CAFE lowers the average WER from 8.55% to 8.02%, beats an adapter-only capacity control at 8.51%, and makes English more stable across checkpoints. At 254M parameters, it trails the 0.6B teacher by 2 WER points and is competitive with much larger models across 48 test sets in 24 languages. We use 564k hours of training data and state-of-the-art models to keep the findings relevant to practice.

Abstract: Compressing the key-value (KV) cache is essential for serving long-context large language models (LLMs) under memory constraints. While current scalar quantization methods achieve significant compression, they typically face a quality-fidelity trade-off, either suffering from performance instability at 2-bit precision or necessitating 3-bit keys to maintain model quality. In this work, we demonstrate that a uniform low-bit k2v2 configuration (2-bit keys and 2-bit values) can preserve near-baseline quality by using vector quantization (VQ) based on the $E_8$ lattice for KV keys. We show that structured VQ exploits correlations across adjacent dimensions to achieve signal-to-quantization-noise ratios (SQNR) higher than those of independent scalar quantization at the same key payload rate. To ensure practical utility, we adapt a table-based asymmetric scoring algorithm that avoids explicit codebook dequantization during attention computation and preserves decode-throughput parity with scalar baselines. Across five LLM families, our $E_8$ k2v2 approach yields \textasciitilde$6\times$ KV compression, roughly 20\% smaller than scalar k3v2 (3-bit keys and 2-bit values), while retaining near-baseline long-context perplexity. In our full-suite perplexity evaluation, $E_8$ improves over scalar k3v2 in 80\% of model--dataset--context settings. These results indicate that $E_8$ lattice VQ improves quality per byte while keeping throughput at parity, suggesting that structured key quantization is a useful primitive for KV-compression stacks. Overall, our work provides a low-overhead primitive for expanding the quality-memory frontier in long-context LLM inference.

Abstract: Multi-source semantic information fusion enables edge networks to support global inference using observations collected by distributed terminals. However, limited communication and computation resources restrict semantic uploading, leaving the edge server with incomplete tokens that differ in modality, semantic granularity, arrival time, and channel condition. The value of these tokens also depends on the complementary or redundant information provided by other terminals. In this paper, we propose a multi-source and multi-dimensional semantic information fusion framework that coordinates terminal uploading and edge-side fusion under resource constraints. We develop Hyper-Slot to combine semantic content with available token attributes through iterative token-to-slot assignment and slot-to-token feedback, producing a compact representation for global inference. To encourage useful semantic uploading, we further design a contribution-based incentive mechanism that uses Hyper-Slot assignments to evaluate information coverage and cross-type complementarity. A bi-level Stackelberg-Nash game determines the server reward pool and terminal transmission-quality strategies while accounting for computation and communication energy costs. Experiments on a tri-modal extension of COCO2014 and image-text retrieval on Flickr30K show that Hyper-Slot improves mAP by 3.66 percentage points and RSum by 15.9 points over the strongest evaluated baselines, respectively, while reducing the number of parameters by more than 80% compared with a naive concatenation-based fusion baseline.

Abstract: Processing long visual token sequences from high-resolution images makes multi-step reasoning computationally expensive for multimodal Large Language Models (MLLMs). Existing one-shot pruning and aggregation methods compress visual tokens into a fixed context before decoding. However, visual evidence needs can shift as reasoning unfolds, making it difficult for a fixed compressed context to retain all the details needed across stages. To address this challenge, we propose ViMoD, a lightweight framework that maintains a compact visual context while preserving access to original fine-grained evidence as reasoning needs evolve. Deformable Aggregation of Region-wise Tokens (DART) learns content-adaptive groups and aggregation capacities, constructing compact Coarse representations linked to recoverable original Fine tokens. Temporal Routing for Adaptive Contextual Evidence (TRACE) integrates decoding history to anticipate upcoming evidence needs and select, retain, or replace active Fine-token groups. Selected Fine tokens augment the persistent Coarse context in the frozen backbone, enabling stage-specific evidence access without continuously attending to all visual tokens. On Qwen3-VL-4B, ViMoD outperforms all evaluated baselines on all eight reasoning benchmarks at a 20% target visual token budget, improving the mean normalized score by 39.0% over the strongest evaluated one-shot baseline. These gains are achieved with only 0.0546% additional trainable parameters relative to the frozen backbone.

Abstract: Polarization imaging provides physical cues beyond intensity imaging but typically requires specialized hardware. Recent methods infer polarization from RGB-like inputs, yet predict only normalized Stokes components or relative descriptors, from which the radiometric scale needed for full Stokes reconstruction has been divided out. We introduce PolarScale, a benchmark that makes this scale an explicit prediction and evaluation target. Built on existing trichromatic full-Stokes measurements, PolarScale takes the per-scene normalized total-intensity image $s_0$ (a scene-referred linear image, not a consumer sRGB photograph) and asks models to predict normalized Stokes components, AoLP/DoLP/DoCP, and a per-scene scale. Because the scale is divided out of the input, it is not physically identifiable; PolarScale therefore evaluates dataset-conditioned semantic scale estimation against a constant-scale control, together with angular, self-consistency, and physical-bound metrics. Across seven restoration-based and generative backbones and three prediction strategies, the strongest restoration models estimate the scale with 3.6-4.3% mean relative error versus 5.7% for the constant control and violate physical bounds on fewer than 0.25% of pixels, whereas two generative baselines collapse to a near-zero scale; explicit descriptor supervision improves descriptor accuracy (23.66 vs. 18.88 dB PSNR for MAE). Predicted full-Stokes representations improve diffuse/specular separation, material segmentation, and glare classification, although in diffuse/specular separation the learned scale performs only on par with the constant control.

Abstract: We present a fully offline speech-to-speech translation pipeline that runs on a Jetson Nano (4 GB) and corrects its own weak translations without retraining. A Whisper-tiny ASR feeds an Opus-MT translator; multilingual BERT cosine similarity acts as a Quality Estimation (QE) gate, triggering a secondary-pass correction when confidence falls below a pre-defined threshold $τ$. We compare three correction methods: QE reranking (M1), Minimum Bayes-Risk decoding (M2), and constrained beam search (M3). On 1,012 FLORES-200 sentences (English-Spanish), M2 at $τ=0.90$ produces statistically significant improvements over greedy decoding on BLEU (+0.67, p<0.001), ChrF (+0.51, p<0.001), and COMET (+0.0020 at N=3, p=0.002); M1 yields no significant gains, and M3 is significantly worse than baseline (p>0.99). Our central finding is that QE functions effectively as a gate but poorly as a ranker: removing the QE model from candidate selection (M1$\to$M2) does not hurt quality and frees 680 MB from the critical path. Using a gain-to-edit ratio adapted from the post-editing-effort literature, we further show that smaller candidate pools (N=3) yield more surgical corrections with better semantic adequacy, while larger pools (N=10) maximise lexical reward. We release the system and demonstrate live translation across six language pairs.

Abstract: Language-model agents are evolving into long-running services that interact with models, tools, computers, mobile devices, and distributed environments. Existing agent frameworks simplify reasoning and tool invocation. However, cloud-centric designs face three limitations: centralized execution increases failure impact, scaling pressure, and compute cost; extending agents across computers, mobile devices, and edge environments requires a unified execution abstraction with permission control; and long-running executions require consistent lifecycle management across failures, recovery, results, usage, and settlement. We present Nexus, a cloud-edge platform that treats each invocation as a persistent task. Nexus uses an OpenWrt-based runtime for distributed serving, run-scoped delegation for authorized access to Computer and Mobile environments, and persistent records to track execution, outputs, failures, recovery, usage, and charging across cloud and edge components. We evaluate Nexus on controlled, cross-device, and model-driven workloads. All ten Computer-Android workflows succeed, and all six revocation tests block subsequent writes while preserving prior authorized reads. Under worker loss, journaling eliminates duplicate appends (six to zero per task), adding 0.933 s mean normal-path overhead. Across 24 matched task pairs, Nexus completes 24 tasks versus Dify's 22 and is a median 3.88 s faster on jointly successful pairs. In a separate workload, Nexus operates under a smaller tested incremental-runtime memory ceiling than Dapr (16 versus 64 MiB), although Dapr achieves lower successful-call latency. These results demonstrate how locality, operation-scoped authority, and persistent result identity support cloud-edge agent services with workload-dependent costs.

Abstract: Fluid antenna systems (FAS) have recently emerged as a promising technology for future wireless networks due to their ability to access a large number of channel realizations using a single radio-frequency chain. This capability creates new opportunities for physical-layer security beyond conventional beamforming and artificial-noise approaches. In this paper, we propose a novel Fluid Antenna Secure Channel-driven Constellation (FASCC) framework for downlink communications, in which information is conveyed through channel-domain constellation points generated by FAS port selection and phase adaptation. Two complementary FASCC schemes are developed, depending on the availability of eavesdropper's channel state information. Numerical results demonstrate that the proposed FASCC framework achieves significant secrecy gains in terms of bit error rate and mutual-information secrecy performance, even when Eve is significantly closer to the base station than Bob.

Abstract: Recent progress in multimodal, high-dimensional learning has enabled foundation models to process heterogeneous, large-scale data. However, at test time, acquiring all features or modalities can be prohibitively costly and often redundant. Sequentially selecting informative modalities is therefore critical, yet challenging when the downstream task or prediction target is unknown. To this end, we introduce ECHO-$k$, a task-agnostic and self-supervised learning principle for modality acquisition: we use a deep model's internal pretrained representations (e.g., from a foundation model) as proxy targets that summarize cross-modal information. We provide theoretical guarantees in a stylized linear setting that motivate a reinforcement learning (RL) policy for sequential modality selection. Across task-agnostic and label-free acquisition baselines, ECHO-$k$ consistently improves budgeted downstream performance across diverse foundation-model backends. Our method provides a principled route to cost-aware test-time deployment, with implications for any multimodal system where measurements are expensive or time-constrained, and downstream tasks unknown a priori.

Abstract: Emerging multi-agent LLMs demand privacy-preserving edge deployment, yet current inference systems struggle with these collaborative workflows. Specifically, the memory-bound decode phase causes severe bus contention on unified memory architectures (UMA), paralyzing naive CPU-GPU co-execution. Furthermore, speculative decoding in multi-agent workloads faces extreme variance in drafting difficulty, alternating between complex reasoning and predictable structured generation. Compounded by frequent tool-induced stalls, this highly fragmented execution severely underutilizes hardware and defeats traditional static batching. We present EdgeAgent, a cross-layer inference system explicitly co-designed for edge UMA and multi-agent workloads. At the micro-architectural level, it bypasses rigid graph-compiler constraints to enable zero-copy UMA-aware tensor parallelism, utilizing asymmetric memory layouts to fully saturate both CPU and GPU compute units. At the scheduling level, it dynamically allocates draft budgets based on real-time sequence predictability to bound bandwidth waste. Concurrently, an asynchronous suspend-and-yield mechanism actively evicts stalled agents, ensuring continuous hardware saturation during unpredictable tool invocations. Extensive evaluations on an Apple M4 SoC demonstrate that the UMA-aware execution alone contributes a 1.29x speedup over batched speculative decoding. Adding the agent-aware scheduling lifts the full EdgeAgent system to a 1.77x speedup under extreme tool-use latencies.

Abstract: Advancements in high-resolution numerical weather prediction (NWP) and data assimilation (DA) have shaped the developments in deep learning (DL) architectures emulating atmospheric dynamics. Emulators for weather forecasting exhibit forecast quality comparable to physics based models at forecast horizon scaling from few days to subseasonal time scales. The emulators are driven by hardware-accelerated matrix multiplication in autoregressive inferences, significantly reducing the computation time and resources required for NWP. Optimization of the matrix multiplication processes in GPU architectures provides opportunities to scale towards high-resolution domain, and offers implementation of out of the box solutions. Post-training quantization (PTQ) has been demonstrated across multiple DL architectures to accelerate and increase the number of computations in unit time while consuming less power, enabling applications on edge hardware. In this study, we investigate the effect of PTQ on pre-trained AI emulators for global-scale weather forecasting. We implement PTQ algorithms in Deep Learning Weather Prediction (DLWP) and FourCastNet (FCN) models as a proof of concept for geophysical fluid dynamics applications. We systematically investigate the effect of PTQ on emulator inferences over short-range forecast horizons. Evaluation of PTQ configurations using simulated quantization hints at qualitatively meaningful forecasts over short-time horizons. These results provide a first benchmark of PTQ for autoregressive weather emulators and a basis for quantization-based optimization of DL models for dynamical systems.

Abstract: Agentic AI systems increasingly rely on cloud-hosted large language models for planning, tool use, and iterative execution, raising concerns about API cost and data exposure. Advances in personal AI devices enable agents to execute locally, but limited resources on device may affect task success and performance. Existing benchmarks are inadequate for systematically characterizing these trade-offs across devices, workloads, and deployment architectures. We present AgBench, a benchmark suite and open artifacts for reproducible evaluation of agentic AI on personal devices. Using AgBench, we evaluate local, hybrid, and cloud execution across agentic workloads, examining task success, latency, cloud API cost, and data exposure. Our results, drawn from over 162.07 million data points, show that personal AI devices can complete many agent tasks locally, but local-only execution generally has lower task success and longer completion times than cloud-only execution, especially as concurrency increases. Local-only execution eliminates cloud model API costs and sensitive-information exposure to cloud agents. Hybrid execution can improve task success, but its cloud cost and data exposure depend on how agents divide work and share information. No single architecture performs best across task success, goodput, cloud cost, and data exposure; deployment choices should reflect the intended workload and device capabilities. AgBench is available at https://anonymous.4open.science/r/AgBench-2777.

Abstract: Vision Transformers (ViTs) incur quadratic self-attention cost in the number of tokens. Most token-reduction methods adapt token identities within a prescribed layer-wise compression schedule, or search a static mask offline, and thus limit online adaptation of when and how much to prune. We propose DORA (Dynamic Online Reinforcement Agent), which learns an input-adaptive pruning policy itself for frozen ViTs. At each eligible block, a hierarchical actor decides whether to prune, how many tokens to remove, and which tokens to remove from each image's evolving representation. Because early deletions change the states observed by later decisions, DORA formulates pruning as a finite-horizon Markov decision process. Complete-prefix shadow evaluations convert final-prediction fidelity into localized per-step credit, while closed-loop accuracy feedback adjusts the fidelity penalty toward a shared accuracy-drop target. A privileged critic and all shadow computations are training-only. Deployment retains the frozen backbone and a lightweight actor that applies hard deletion and packed variable-length FlashAttention, converting token reduction into measured speedups. On ImageNet-1K with DeiT-Base, DORA reduces FLOPs by 38.4% relative to the uncompressed backbone within one percentage point of accuracy loss. Averaged across four ViT-type backbones at matched accuracy, DORA uses 13.2% fewer FLOPs and achieves 32.4% higher throughput than the corresponding per-backbone baseline means. Under zero-shot transfer to ImageNet-A, these gains widen to 20.3% and 45.6%, respectively.

Abstract: Signal-level speech privacy mechanisms suppress linguistic content while preserving acoustic information needed by downstream sensing applications. However, their privacy settings are typically evaluated/selected offline and remain fixed during deployment, even though speech-content leakage can vary substantially across utterances and speakers. Adapting privacy protection at runtime requires estimating how much speech remains recoverable, but conventional measures such as WER or PER require ground-truth transcripts and therefore cannot be computed online. We present CLEAR, a reference-free approach for estimating speech-content leakage at runtime using disagreement among heterogeneous ASR systems. Our key insight is that independently trained ASRs exhibit consistent behavior when linguistic content remains recoverable and increasingly disagree as privacy transformations obscure speech. Using configurable speech-suppression mechanism, we show that cross-ASR disagreement closely tracks transcript-grounded leakage across privacy operating points, achieving a correlation of 0.8. We further characterize the latency-accuracy trade-off of heterogeneous ASR subsets and use their hypotheses to identify potentially exposed words. These capabilities enable privacy to be treated as a runtime property rather than a fixed configuration: CLEAR can communicate residual speech exposure to users and provide feedback for dynamically adjusting privacy aggressiveness while retaining acoustic utility for downstream sensing tasks.

Abstract: Automatic speech recognition models are audited for demographic fairness at full precision, yet the models that ship to production have been quantized, pruned, and distilled. We ask whether post-training weight compression, which alters model weights rather than the audio signal or its feature representation, redistributes error burden across demographic groups. Across the Whisper family on Fair-Speech, Common Voice 25, and AfriSpeech-200, 50% Wanda pruning of Whisper-large-v3 sharply widens the Black/AA-vs-Asian temporal-taxation differential on Fair-Speech: the absolute word-error-rate gap between the worst- and best-served groups more than doubles; at an assumed cost of five seconds of correction effort per transcription error this is a rise from 30 to 64 seconds of correction time per minute of speech. This +111% relative increase is invariant to the assumed per-error cost, survives an audio-quality control, and is only partly mitigated by beam-search decoding, which still leaves an +86% increase. At edge model size, INT4 HQQ quantization compounds catastrophic transcript loops on West African accents by factors of five to seven. Distillation, by contrast, narrows demographic gaps in 21 of 27 evaluated settings (teacher-student pair, precision, and dataset), with the exceptions concentrated on a single model pair. We cast the temporal-taxation construct of Choi and Choi (2025) as a quantitative metric, and show that single-snapshot fairness audits on full-precision models do not capture the deployment-time burden that compression places on already-marginalized speakers.

Abstract: KV cache memory has become a major deployment bottleneck for video generation and world models, which motivates low-bit quantization study for efficiency. Existing 2-bit KV cache quantization methods can achieve nearly lossless performance on video benchmarks such as VBench, however, we find that they still cause severe temporal flickering and visual degradation. Meanwhile, deeper investigates show that Key quantization produces smaller reconstruction errors than Value, but surprisingly leads to much larger output degradation. We trace this discrepancy to attention: small Key perturbations can change the attention logits, i.e., QK^\top, and shift the temporal-spatial tokens selected by Queries. These observations motivate us to explicitly preserve attention logits and temporal-spatial token selection during KV cache quantization to alleviate the visual degradation problem. To address this issue, we present QuantWM, a training-free and strictly causal 2-bit KV cache quantization framework. QuantWM introduces two complementary techniques to mitigate the attention shifts. Firstly, quantization-sensitivity-aware clustering (QSAC) jointly considers historical Query sensitivity and residual ranges to select INT2-friendly Key centroids, which reduces quantization errors in channels that are more critical to attention. In addition, principal-subspace attention compensation (PSAC) restores the remaining Key errors along the dominant Query subspace using low-rank projections, which provides a direct and efficient correction to stabilize attention logits. Extensive experiments on Causal-Forcing, LingBot-World-v2, HY-World 1.5, Matrix-Game-2 and Longcat-Video demonstrate that QuantWM significantly improves visual quality and temporal consistency, while outperforming existing methods across image and video quality metrics with up to 6.20x KV cache memory compression and limited additional overhead.

Abstract: Large vision-language models (VLMs) can be efficiently deployed under stringent memory and latency constraints through post training quantization (PTQ). However, most PTQ methods are designed for unimodal large language models (LLMs). These methods treat quantization errors as isotropic perturbations under the Euclidean assumption, which provides weak guidance on directions most sensitive to quantization in VLMs. Consequently, directly adapting unimodal PTQ approaches or solely employing modality-specific scaling often leads to uneven bit-width distribution and inconsistent performance in low-bit settings. To address these challenges, we propose Riemannian Geometry-Sensitive Quantization (RGSQ), which formulates quantization as a reconstruction problem under a unified Fisher-Riemannian metric. RGSQ identifies modality-specific sensitive directions via Riemannian manifold mappings built from modality-partitioned empirical Fisher factors and fused into a modality-aware Kronecker-structured metric. We then apply geometry-aligned rotations to reorient the local tangent frame, steering low-bit perturbations toward loss-insensitive axes. Finally, we apply a whitening transformation that maps the Riemannian objective to an equivalent Euclidean form, enabling standard unimodal PTQ methods to evaluate multimodal quantization error under their original assumptions. Across an extensive and diverse set of mainstream VLM benchmarks, RGSQ achieves the highest accuracy and stability under extremely low-bit settings (W2A8 and W3A8). It outperforms VLM-aware baselines, such as MBQ and MQuant, by up to 5.9% and surpasses single-modality improvements by up to 8.6%.

Abstract: Physical AI requires models to ground visual and linguistic understanding in real-world environments while accounting for environmental constraints and execution feedback. We introduce MachEmbodied-VLM (ME-VLM), a unified vision-language model with two variants, 4B and 35B-A3B, that brings together embodied cognition and multimodal agent capabilities. Our work emphasizes physical perception and spatiotemporal reasoning, together with planning, interaction, and outcome assessment in both digital and physical environments. We construct training data spanning embodied and multimodal agent tasks, including execution observations and feedback to support outcome assessment and decision refinement. The training pipeline comprises embodied capability injection, separate reinforcement learning of embodied and multimodal-agent experts, and multi-teacher on-policy distillation that consolidates their complementary capabilities into a single model. Experiments show competitive performance on both embodied and agent benchmarks, as well as on autonomous-driving and embodied-navigation tasks. For edge deployment, visual token compression, W4A8 quantization, and hardware--software co-optimization enable on-device inference of the 4B variant on the M100, reducing prefill latency from 400 ms to 188 ms. Project Page: https://machembodied.com/ME-Brain/ME-VLM.html Code Repository: https://github.com/MachEmbodied/ME-VLM

Abstract: Voice-controlled interaction in industrial settings is hampered by acoustic noise, which severely degrades audio-only speech recognition. Audio-visual speech recognition (AVSR) addresses this by fusing lip-motion cues with the audio stream, but state-of-the-art pipelines rely on three-dimensional convolutions, recurrent units, and attention modules that exceed the budget of typical edge devices. We present NAVIR, an end-to-end AVSR system targeting the BrainChip Akida neuromorphic processor, which natively supports only sequential two-dimensional convolutional inference. The pipeline factorises spatial and temporal encoding into separate AkidaNet-based modules: a per-frame visual encoder, a temporal video encoder, and a spectrogram audio encoder, fused by a lightweight predictor head and decoded by constrained beam search. Models are trained with connectionist temporal classification on noise-augmented audio and then fine-tuned with quantization-aware training. On the GRID benchmark, the quantized audio-visual model reaches 14.0% word error rate (WER) under noise on the unseen-speaker split and 3.3% WER on the overlapped-speaker split, against 22.5% and 11.8% for audio-only baselines, and it attains 98.6% command accuracy at 1.5% WER on a task-specific industrial-command corpus. Operation-count analysis indicates a 13-fold energy advantage of the spiking formulation over its artificial neural network counterpart at 27.6% mean firing rate. On-board measurements show roughly 5-fold lower energy per inference than a Raspberry Pi central processing unit on the lip-reading model, and over 100-fold lower than a laptop graphics processing unit, while sustaining 14.5 inferences per second. To the best of our knowledge, this is the first complete multimodal AVSR pipeline running on neuromorphic hardware of this class.

Abstract: Self-attention is central to modern Transformer architectures, but its dense dot-product formulation makes it difficult to identify which internal directions are structurally important and which can be modified without disrupting the model. SVD-Inspired Attention (SVDA) addresses part of this problem by introducing a learned diagonal spectrum into the query-key score interaction, making latent attention directions explicitly inspectable through indicators such as spectral entropy, effective rank, sparsity, alignment, selectivity, and perturbation response. This paper examines the transition from diagnostic interpretation to operational intervention. A diagnosis--intervention--verification framework is proposed, and one intervention is evaluated: spectral energy retention in the attention-score pathway. Across FashionMNIST, CIFAR-10, CIFAR-100, and Food-101, the $ρ=0.90$ prescription removes 24.5--53.7\% of score directions, reduces parameters by 2.6--4.3\%, and reduces estimated MACs by 2.8--5.4\%. The paired mean accuracy change of the dimension-reduced model ranges from $-0.03$ to $+0.05$ percentage points over three seeds. These results support SVDA as an intrinsically interpretable attention mechanism whose learned spectrum exposes an operational coordinate system for deterministic and verifiable modification of attention-score formation.

Abstract: Vision Transformers allocate most parameters to multi-layer perceptrons (MLPs) for channel mixing, while token interactions usually rely on quadratic multi-head self-attention (MHSA). Linear attention reduces sequence complexity to O(N), but remains coupled with the same fixed-activation MLP as softmax Transformers. Kolmogorov-Arnold Networks (KANs) instead place learnable univariate maps on edges, yet prior vision KANs keep MHSA or omit attention entirely. We introduce LKAT (Linear Kolmogorov-Arnold Transformer), an isotropic ViT encoder that couples chunkwise Gated Linear Attention (GLA) with a two-layer KAN feed-forward, and we provide an I/O-aware fused RBF-KAN kernel for the radial-basis grid maps. Under a shared DeiT-style recipe we compare LKAT with ViT, ViT-5, and MLP-Mixer. LKAT-B exceeds ViT-B/16, ViT-5-B, and Mixer-B/16 on ImageNet-100. Tiny/Small/Base LKAT variants scale consistently on CIFAR-10/100, and ImageNet-100 pretraining transfers to CIFAR fine-tuning. The results support gated linear attention and KAN-based radial-basis functions as complementary inductive biases for mid-scale visual representation learning.

Abstract: Visual signals require compact yet sufficient representations for robust downstream prediction. Convolutional sparse coding (CSC) provides an explicit mechanism for suppressing redundant components while preserving signal content, but its sparsity coefficient is typically fixed and manually selected. We propose an adaptive convolutional sparse coding framework for robust visual signal representation. Specifically, we unfold the CSC optimization with the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) and treat the sparsity coefficient as a differentiable variable jointly learned with the network parameters. From the information bottleneck perspective, this coefficient controls the trade-off between information retention and compression: the sparsity term promotes compact representations, while the reconstruction term together with task loss preserves task-relevant signal content. We further introduce a label-free post-training strategy that adjusts the compression strength for corrupted inputs with the main network parameters fixed. Experiments on CIFAR and ImageNet demonstrate competitive clean-data recognition and greatly improved robustness under different input perturbations.

Abstract: Self-supervised learning (SSL) has improved speech representations, yet performance degrades in pathological domains such as electrolaryngeal (EL) speech, and the computational footprint of SSL models limits their applicability in real-time, on-device deployment. We propose a multi-teacher knowledge distillation framework to train a lightweight, streaming content encoder that generalizes across healthy (HE) and EL speech. Two teachers are distilled progressively: a frozen SSL model providing discrete phonetic cluster targets from HE speech, and an EL-fine-tuned speech recognition model supplying continuous bottleneck feature targets. Evaluated via downstream speech recognition, our approach reduces the EL word error rate to 21.2%, compared to 39.3% for the strongest zero-shot SSL baseline. Among causal convolutional, Transformer, Conformer, and Mamba-based student architectures, a Mel-Conformer achieves the best combination of EL accuracy and computational efficiency. The final encoder contains 21.9,M parameters and runs at a real-time factor of 0.30 under ONNX Runtime on a single CPU core.

Abstract: Parameter-efficient fine-tuning (PEFT) has recently emerged as a pivotal research direction for adapting pre-trained point cloud transformers to diverse downstream tasks. Although existing methods achieve excellent fine-tuning performance with high parameter efficiency, they ignore inference efficiency. To tackle this problem, a novel PEFT method termed position anchor tuning (PAT) is proposed in this paper. As multi-head attention (MHA) and feed-forward network (FFN) are computation-heavy blocks in pre-trained transformers, PAT decreases their computational cost through token aggregation-expansion pairs. Each pair comprises a token aggregation module (TAM) and a token expansion module (TEM). For MHA and FFN blocks, TAMs extract representative tokens from their input tokens based on position anchors in 3D space. These extracted tokens, rather than the original input tokens, are processed by the blocks, thereby reducing the number of tokens involved in computation. Then, TEMs propagate the learned representations back to the original input tokens. Since TAMs are solely responsible for capturing task-specific representations, base-sharing low-rank adaptation (BSLoRA) is further introduced to enable them to learn such representations effectively with only a small number of trainable parameters. Extensive experiments on widely used benchmarks demonstrate that PAT performs comparably to state-of-the-art methods while incurring significantly lower computational overhead and fewer trainable parameters.

Abstract: Sparse DNN accelerators specialize in ANN or SNN execution, leaving energy or latency on the table when workload characteristics vary within a layer. Hybrid accelerator designs that switch modes at layer or tile granularity suffer from low PE utilization since one core type idles whenever the other is active. NeuroFlex is the first accelerator to assign every output element independently to ANN or SNN execution mode with zero accuracy loss. We extend integer-exact ANN-SNN equivalence from layers to individual output elements, thereby enabling mode switching with no conversion error. An offline cost-guided scheduler scores each element by its marginal energy-delay trade-off and packs work across PEs, achieving 97-99% PE utilization compared to 40-45% for layer-wise hybrids. NeuroFlex reduces EDP by 57-67% over a strong ANN-only baseline and delivers up to 2.5x speedup over a dual-sparse SNN-only baseline. Our cost-guided scheduler improves throughput by 16-19% over random element assignment across vision, language, and transformer workloads.

Abstract: Recent advances in multimodal large language mod- els (MLLMs) have opened new opportunities for edge intelligence by enabling reasoning across heterogeneous sensor modalities, such as vision, text, and telemetry data. However, deploying these capabilities on resource-constrained edge platforms remains challenging due to the substantial computational, memory, and communication demands of modern MLLMs. Rather than transmitting raw sensor observations or partitioning neural networks at intermediate layers, Edge Multi-Modal Intelligence (EMMI) communicates a compact representation between edge devices and server resources, enabling communication-efficient edge MLLM inference. To achieve this, EMMI performs modality-specific encoding, cross-modal representation fusion, and learned compression at the edge, transmitting only a compact latent representation to server-side resources for high-capacity MLLM reasoning. This representation-centric design reduces communication overhead, preserves local data privacy, and provides a fixed-size interface between heterogeneous edge devices and server-side MLLMs. Evaluation on a representative multimodal benchmark demonstrates that EMMI can reduce the communication payload by 32x while maintaining comparable downstream accuracy, resulting in up to a 3.4x reduction in estimated end-to-end inference latency under bandwidth-constrained edge conditions.

Abstract: KV cache is evolving from a serving optimization into an external memory substrate for long-term LLM agents. In a shared multi-user deployment, however, reusable KV blocks introduce a missing access-control question: semantic relevance alone cannot determine whether a memory block is authorized for the current physical user. We propose Bio-MemArt, a biometric-aware KV-cache memory framework for multi-user LLM agents. Bio-MemArt attaches a normalized biometric template to each stored KV memory block, filters the shared memory pool with the current user's biometric probe, and then runs the original MemArt retrieval and KV reuse pipeline only inside the authorized candidate pool. This design preserves latent-space retrieval, direct cache reuse, and decoupled position encoding while adding physical-user access control to shared KV memory. We evaluate Bio-MemArt under Owner and Non-owner query conditions on long-term dialogue QA with face and palmprint benchmarks. Across face benchmarks, the average owner and non-owner biometric success rates are 95.71% and 0.86%; across palmprint benchmarks, they are 97.60% and 2.00%. In the efficiency study, average prefill tokens drop from 18,781.96 under full-context prompting to 28.57 with Bio-MemArt, showing that biometric gating preserves the low-token operating regime of KV-cache memory.