edge ai developers

Abstract: This paper presents a federated learning (FL) framework for semantic communication (SemCom) over the multiple access channel (MAC), where multiple distributed transmitters simultaneously convey semantic information to a common receiver using end-to-end learned semantic representations. Semantic feature extraction, compression, wireless transmission, and semantic inference are jointly optimized through FL-based SemCom. Unlike conventional SemCom systems that train users independently, simultaneous semantic transmissions over the MAC enable concurrent training of multiple transmitters within the same transmission interval, thereby reducing training overhead and channel usage. Federated semantic aggregation is incorporated with shared and personalized encoder components, allowing transmitters with limited local training data to benefit from semantic representations learned from transmitters possessing larger datasets while preserving local adaptation. Evaluated using the CIFAR-10 dataset over fading channels, the proposed approach maintains high semantic inference accuracy under shared wireless access and significantly improves performance under severe data imbalance. In the meantime, simultaneous semantic transmissions can significantly reduce training overhead while preserving robust semantic performance across distributed transmitters with heterogeneous local datasets. These results highlight the potential of combining FL and SemCom to support scalable and communication-efficient edge intelligence in future wireless networks.

Abstract: Mixture-of-Experts (MoE) inference on memory-constrained edge platforms is often dominated by expert-weight movement rather than computation. Although each token activates only a small subset of experts, routing is input-dependent, so all experts must remain accessible and memory movement dominates latency. Existing approaches such as uniform pruning and expert dropping overlook two properties of MoE execution: expert combinations are highly skewed in frequency, and experts contribute asymmetrically within each combination. We present CRISP, a combination-aware framework that co-designs pruning and runtime scheduling for efficient MoE serving on memory-constrained edge platforms. CRISP reorders neurons by importance so one stored expert can support multiple execution widths, assigns differentiated execution to hot, warm, and cold combinations based on frequency and expert role, and exploits heterogeneous widths to overlap expert loading with computation. Across Mixtral-8x7B and DeepSeek-MoE-16B, CRISP improves throughput by up to $2.1\times$ at the same memory budget while limiting accuracy loss to 3.6\% relative to full-width inference. Our results highlight that efficient edge MoE serving requires jointly optimizing which expert capacity is preserved and how expert execution is scheduled.

Abstract: This paper investigates cross-layer design for differentially private federated learning over cell-free massive MIMO networks with over-the-air (OTA) aggregation. Distributed access points (APs) collaboratively serve user equipments (UEs) through user-centric clustering, where clipped local model updates are perturbed with Gaussian noise and simultaneously transmitted over a shared uplink for OTA averaging at a central processing unit. Exploiting channel hardening, we show that thermal noise inherent in wireless reception can act as a privacy-masking resource, reducing or even eliminating the need for artificial noise. Based on this observation, we formulate a joint optimization problem over AP-UE associations, UE transmit powers, and artificial noise variances to minimize aggregation variance subject to per-UE zero-concentrated differential privacy constraints derived from an ergodic privacy accounting framework. Analysis reveals a regime-dependent structure: stringent privacy requirements favor low transmit powers and sparse AP associations to exploit thermal noise for privacy, whereas relaxed requirements favor larger clusters to improve aggregation quality. The resulting mixed-integer nonconvex problem is solved via alternating optimization combining difference-of-convex power allocation and greedy coordinate-descent clustering. Numerical results show up to 89% lower aggregation variance than conventional baselines and enable 97% test accuracy on MNIST with consistent gains on CIFAR-10. Under strict privacy budgets, competing schemes remain near random-chance performance, while under relaxed budgets they continue to lag significantly behind the proposed design.

Abstract: The proliferation of orbital sensing data and the real-time demands of emerging global services are pushing low Earth orbit (LEO) satellite networks from bent-pipe relays toward intelligent onboard processing. This shift confronts a fundamental tension, where modern artificial intelligence (AI) training paradigms require intensive computation, abundant memory, and stable connectivity, all of which are scarce under the dynamic topologies, intermittent links, and severe size, weight, and power (SWaP) constraints of satellite networks. Addressing this tension demands rethinking AI workflows, from training and inference to the emerging deployment of large language models (LLMs), under a constraint-aware design philosophy. To the best of our knowledge, this paper provides the first survey that jointly covers all three, whereas prior surveys examine these aspects in isolation. We begin by identifying the deployment challenge of satellite intelligence (SI) and deriving design principles for SI. We then examine three advancing paradigms: i) distributed training, where federated learning (FL) and split learning (SL) enable satellite-ground co-training under resource limitations; ii) collaborative inference, which exploits single-satellite, multi-satellite, and satellite-ground architectures to meet real-time service demands; and iii) on-orbit LLM adaptation and serving, which tackles the frontier challenge of deploying LLMs under severe orbital constraints and resource-aware serving. Finally, we identify open challenges and future directions. By bridging the AI and satellite networks, this paper provides a research roadmap for the implementation of space-native AI technologies.

Abstract: Multi-Head Latent Attention (MLA) has been adopted by recent frontier large language models. Instead of per-head keys and values, MLA caches a compressed KV latent that is up-projected into keys and values at decode time, together with a narrow positional key that carries position information into the attention score. Existing KV-cache quantization methods are designed around the keys and values of standard attention and do not account for what MLA actually caches, which costs accuracy at low bit widths. We analyze the MLA cache both in activation distribution and in how quantization error propagates to the attention output, and show that the two cached tensors play different roles: the latent mainly supplies the values, whereas the positional key dominates the attention score. We therefore propose RAQ-MLA (Role-Aware Quantization for Multi-Head Latent Attention Caches), which fits an invertible transform to each cached tensor against the error that tensor delivers to the attention output: attention-weighted value error for the latent and score perturbation for the positional key. Without training, RAQ-MLA quantizes the full MLA cache to about 2.5 bits per element, a 6.5x increase in theoretical KV-cache capacity, while remaining nearly lossless on reasoning and coding tasks across four MLA models.

Abstract: Safe robotic task planning requires planners that can reliably achieve task goals while satisfying safety constraints. Classical planners search over explicit symbolic task representations and can provide formal guarantees with respect to encoded safety constraints, but require rigid input formats. Their computational cost also increases rapidly with the search space. In contrast, large language models (LLMs) offer flexible task interfaces and strong generalization capabilities, but they are large, power-hungry, and pre-trained with little robot-specific safety data and knowledge. Thus, they can be difficult to deploy locally on robotic systems with limited resources. To address these limitations, we present SLoSa, a neuro-symbolic compression and fine-tuning framework for training foundation-model-based robotic task planners with Small size, Low power demand, and Safety awareness. The framework first constructs a safety-aware dataset using classical planners, an advanced teacher LLM, and formal verifiers, followed by two training stages: supervised fine-tuning and reinforcement learning (RL) fine-tuning guided by a verification-based reward machine to enforce safety alignment. We evaluate SLoSa across four Planning Domain Definition Language (PDDL) domains, two household simulation environments, local edge platforms, and a desktop robotic arm. Across the PDDL tasks, verifier-guided post-training raises the success rate of compact 4B-14B planners to 70.5-76.0%, while INT4 deployment on a Jetson Orin requires 17.85-26.41 W of average power. Results in household simulators and on the physical robot further demonstrate the applicability of SLoSa across diverse robotic task settings.

Abstract: Instance-dependent early stopping (IES) reduces training computation by skipping backpropagation for examples whose losses have stabilized, but the second-order loss-difference criterion used by IES lacks a theoretical justification. We develop Theory-Guided Instance-Dependent Early Stopping (TG-IES), which analyzes this criterion under explicit assumptions and derives a noise-calibrated mastery threshold and a conditional stability bound for instance exclusion. We show that second-order differencing has a favorable bias-variance trade-off over first- and third-order differences in a specific signal-to-noise regime, and that its signal contains both gradient-drift and curvature terms. TG-IES derives its stopping threshold and patience from training-run statistics rather than using the manually selected threshold of IES. Across CIFAR-10 and CIFAR-100, TG-IES saves 53-58% of backpropagation instances in its main convolutional settings while keeping accuracy within 0.1-0.9 percentage points of full training; in the primary ResNet-18/CIFAR-10 setting, this corresponds to 35.7% FLOP-normalized savings. The method also transfers to ViT-Tiny, where it achieves smaller savings and incurs a modest accuracy cost. These results provide a theoretical basis for second-order instance-dependent early stopping and show that theory-guided calibration can substantially reduce training computation under the assumptions and experimental settings studied.

Abstract: We take the multiply-accumulate (MAC) unit of an open-source 2$\times$2 int8 systolic array taped out on Tiny Tapeout (GF180 process) and propose two replacements. The first one, \emph{hybrid}, uses a different arithmetic structure for each row of the array. It reduces the logic area of one MAC unit by 4--6\% and of the whole chip by 0.7--3.2\% after place and route. The second one, \emph{fused\_eq}, never builds the long carry chain of the accumulation. It shortens the critical path by 20\% for one MAC unit and by 14\% for the whole chip. All numbers come from the standard Tiny Tapeout flow (LibreLane) run in GitHub Actions. Correctness is checked by exhaustive simulation of all $2^{24}$ inputs, a formal equivalence proof, and the original test bench.

Abstract: The training loss of binary and ternary neural networks, expanded as a multilinear polynomial over the discrete weights, is structurally equivalent to the energy of a hierarchical high-order Ising model. We exploit this equivalence to apply Ising solvers to discrete-weight network training, while accounting for a fundamental difference in objectives: an Ising solver seeks the ground state of a fixed energy function, whereas training seeks solutions that generalize to unseen data. Our key contribution is $ΣΔ$-Sampler, an optimizer that reconciles these objectives by embedding a stochastic sampling-and-annealing step as a noisy quantizer inside a first-order $ΣΔ$ noise-shaping loop. The loop pushes noise from weight quantization, high-order frustration, and mini-batch sampling out of the low-frequency adaptation band, so that discrete weight updates follow the smooth descent trajectories embedded in the rugged loss landscape. We validate the framework on a planted quadratic unconstrained binary optimization (QUBO) benchmark with a known minimizer; on CIFAR-10, where a $13$M-parameter ternary model comes within $1\%$ of full-precision accuracy; on ImageNet-1k, where a ternary ResNet-18 comes within $0.4\%$ Top-5 accuracy of a ternary baseline; on class-incremental continual learning, where prior-task accuracy is preserved; and on MatMul-free language models up to $2.7$B parameters, where it outperforms a straight-through-estimator baseline under matched continued pre-training. Compared with standard quantization-aware training, $ΣΔ$-Sampler reduces the training memory footprint by over $50\%$, delivers an order-of-magnitude speed-up, and reduces weight-update memory traffic by four orders of magnitude, making it attractive for on-device training and fine-tuning of large binary and ternary models.

Abstract: Orthogonal transforms have emerged as an effective tool for low-bit quantization of large language models (LLMs), as they reshape activation distributions without changing the underlying model function. Existing approaches typically rely either on fixed data-agnostic transforms, such as Hadamard rotations, or on calibration-time optimization of learned rotations. We study rotation selection from first principles and show that maximizing covariance leads to spectral alignment and provides a principled criterion for constructing quantization-friendly orthogonal transforms. Based on this insight, we propose SpectrumQuant, a closed-form rotation method that estimates activation covariance from calibration data and uses it to construct a covariance-aware orthogonal transform for post-training quantization. Our formulation characterizes quantization-friendly rotations as compositions of alignment and mixing components, while recovering Hadamard matrices as a special case. Interestingly, a class of butterfly matrices belongs to this category. In the realm of LLM quantization, SpectrumQuant bridges the gap between data-agnostic structured rotations and learned rotations, providing an effective and efficient data-aware alternative for post-training quantization. Experiments on Llama and Mistral models show that SpectrumQuant is competitive with state-of-the-art rotation strategies while requiring less than one-third of the GPU memory and significantly less time than learned rotations approaches like SpinQuant and DartQuant.

Abstract: LLM agents are evolving from a single model calling tools to planners that orchestrate multiple expert models whose capabilities live in distinct weights. This shift breaks a hidden assumption of today's serving systems: that the working set of models fits in GPU memory. On the memory-constrained GPUs typical of private and edge deployments, naive whole-model swapping serializes weight transfer with computation and, without host-backed KV restoration, discards session state at a switch. We present JointServe, a serving system that treats weight shards and KV blocks as residency objects competing for one unified memory budget. JointServe exploits three structural properties of agent workloads (directed calls, gapped execution, and dual-state residency) to prepare models before they are called: an intent signal parsed from the caller's own output stream triggers prefetch during generation, partial k* residency reduces switching to incremental layers, and a preemptive transfer scheduler protects live-session KV with a guaranteed bandwidth share. A carve-out control step adjusts the weight/KV allocation as workload demands change. Evaluations show that JointServe reduces session completion time by 21.7% over the strongest whole-model swap baseline and stays usable in a crash regime where every whole-model swap baseline fails to finish (a 46-fold gap). On replayed SWE-bench, tau-bench, and OSWorld traces, it cuts session completion time by up to 49% and lifts SLO attainment from 2.8-24.5% to 76.8-100%.

Abstract: As demand for LLM inference grows and consumer GPUs grow more capable, edge deployment is becoming increasingly attractive. WebGPU is compelling in this setting because it offers cross-platform portability and browser accessibility. However, these same strengths also make high performance difficult to achieve, as an implementation that performs well in one context may not perform well in another. To address this, we present GPU Goldmines, a framework for specifying, executing, and tuning WebGPU kernels. We instantiate GPU Goldmines with kernels for key LLM building blocks: matrix multiplication, matrix-vector multiplication, and flash attention. We characterize performance across diverse devices, shapes, and browsers, and use the resulting database to accelerate subsequent autotuning and configuration selection. On shapes drawn from Llama and Qwen models, our matrix multiplication kernels outperform ONNX Runtime and llama.cpp WebGPU kernels by 7.0x and 1.41x geometric mean, respectively, and deliver end-to-end improvements of 1.48x in the prefill stage of llama.cpp for chat workloads.

Abstract: Multimodal decentralized learning enables collaboration across heterogeneous edge devices without sharing raw data, but differing modality sets create coupled computation and communication bottlenecks. Multimodal agents support cross-modal alignment while training more model components and sharing bandwidth across multiple modality overlays. Under bandwidth limits, these agents can become hot spots: transfers miss aggregation windows, and their updates become stale or underrepresented, degrading worst-agent accuracy. Asynchrony removes global blocking but does not by itself coordinate local cross-modal interaction with timely peer-update exchange. We present MODAIC (Modality-Decoupled Asynchrony with Interleaved Coordination), a wait-free system that coordinates local modality progress, peer-to-peer update delivery, and aggregation under heterogeneous compute and bandwidth. Hierarchical asynchrony lets modalities progress independently while enforcing reciprocal feature exchange; load-aware update fetching overlaps training with communication, limits concurrent serving load, and reuses late arrivals; and consensus-aware peer selection uses lightweight metadata to account for update availability, peer participation, and loss similarity without active probing. We implement MODAIC as a containerized multi-agent framework and evaluate it on local and cross-region cloud testbeds. Across AVE and KU-HAR, MODAIC improves average accuracy and multimodal worst-agent performance. On AVE at 200 Mbps per agent, it improves multimodal worst-agent accuracy by 11.0 percentage points at 1000 seconds over the strongest baseline. On the cloud testbed, it reduces time-to-accuracy by up to 30%. Ablations show that simpler load-balancing and delayed-aggregation variants leave a worst-agent accuracy gap, and stress tests show tolerance to WAN variability and partial tracker unavailability.

Abstract: Quantization lowers deployment costs of vision-language-action (VLA) policies but can compromise task completion. Across five VLA models and 4,050 real-world trials, we systematically reveal that quantization-aware training (QAT) generally improves on post-training quantization (PTQ) but leaves substantial gaps to full precision (FP) on long-horizon tasks and under visual and semantic shifts, while its imitation-based objective does not explicitly use rollout success or partial completion. To exploit these outcome signals, we introduce Quantization-Conditioned Reinforcement Learning (QCRL), an offline RL method coupling progress-based advantage estimation with quantization-aware source conditioning. FP rollouts supply reference experience beyond action-level distillation, progress-scaled terminal penalty distinguishes partial completion, and quantization-aware conditioning selects advantageous reference behavior within the quantized policy. We evaluate weight-only INT4 W4A16 on $π_{0.5}$ and GR00T N1.5 and joint weight-activation NVFP4 W4A4 on $π_{0.5}$. Using only in-distribution data, QCRL reaches 99.3% of FP's mean score across three visual-shift conditions for $π_{0.5}$ with INT4 W4A16, alongside 42.9% model-memory savings. With NVFP4 W4A4, it reaches 100.7% of the same FP baseline, alongside a $1.73\times$ speedup using NVFP4/FP8 deployment. Hardware gains are relative to FP16 on Jetson Thor.

Abstract: Hybrid federated semantic communication enables user equipment (UEs) to contribute model updates or semantic features under a shared training framework. However, heterogeneous role costs and the interaction between model aggregation and feature reconstruction (FR) remain key challenges in practical radio access network (RAN) deployment. To address these issues, we propose ARC-FedSC, a RAN-aware framework that coordinates UE role assignment and server-side controlled FR under a fixed communication budget. Within ARC-FedSC, RAN Aware Role-Bandit learns the historical contribution of UEs in different roles together with current uplink conditions, while Temporal Consistency Alignment (TCA) regulates FR according to recent model aggregation directions. Experiments on CIFAR-10 with time-varying RAN channels show that ARC-FedSC outperforms baselines across all five tested role configurations in the later training stage, with the best configurations achieving about 24.17 dB average PSNR over the last 20 rounds. Component level results further show that Role-Bandit maintains an average rolling UE coverage of 85.3%, compared with about 65.7% for baseline, while TCA reduces the probability of harmful FR events from 0.245 to 0.093, corresponding to a 62.0% reduction. Ablation and sensitivity studies further support the effectiveness and robustness of the proposed design.

Abstract: Large language models (LLMs) have achieved substantial performance gains through increases in model size, training data, and computational resources. However, traditional scaling approaches produce diminishing returns, rising financial and environmental costs, and barriers to participation for researchers operating outside large industrial laboratories. This review examines the evolution of LLM scaling theory from empirical scaling laws to compute-optimal training, with particular emphasis on parameter efficiency, token utilization, data efficiency, and resource-constrained environments. Foundational work on scaling laws is synthesized alongside later research on compute-optimal training, data pruning, efficient architectures, quantization, low-rank adaptation, and edge-oriented optimization. The literature indicates a shift from scale maximization toward more deliberate allocation of parameters, tokens, compute, and hardware resources. At the same time, important empirical, theoretical, and methodological gaps remain regarding whether scaling principles established on enterprise-grade infrastructure generalize to smaller models and constrained computing environments. This review organizes these developments into a unified framework for resource-efficient LLM training and argues that future progress should evaluate efficiency not solely through model performance, but through the relationship among performance, parameter count, computational cost, token allocation, and hardware constraints.

Abstract: In this paper, we study how to achieve one-step action generation in Robotic Foundation Models (RFMs), aiming to overcome the high inference latency of multi-step flow matching. MeanFlow provides a promising framework for this goal, yet its direct application leads to performance collapse. We discover that this stems from two distinctive dynamics exhibited in the RFM velocity field: (1) the ``local acceleration" exhibits stability early on, but surges sharply towards the end of the denoising process, and (2) the spread of its magnitudes across samples widens as denoising progresses. To address these issues, we introduce Kinematic MeanFlow (K-MF), a novel one-step action policy tailored for RFMs. Specifically, grounded in a kinematic identity, K-MF decouples the time derivative term in the MeanFlow formulation into two sub-interval terms separated by an intermediate point. This decoupled formulation enables the two terms to capture early-stage and late-stage denoising dynamics, respectively, while mitigating the error amplification across the process. As a result, our K-MF empowers RFMs to achieve one-step action generation in both training from scratch and fine-tuning paradigms across diverse tasks, while outperforming multi-step flow matching in most settings. In terms of inference efficiency, K-MF reduces action-head latency of GR00T-N1.6 by 67.5%~74.4% across L40 and Jetson Orin in eager and compiled modes, yielding end-to-end latency reductions of 30.3%~54.9%. Code will be available at https://github.com/IntelChina-AI/K-MF.

Abstract: The emergence of AI-RAN infrastructure, which equips cell sites with GPU-accelerated hardware, creates an opportunity to colocate non-RAN workloads with primary RAN processing. We explore using this spare capacity for decentralized training of foundation models (FMs), one of the most compute-intensive AI workloads. We present the first characterization of spare GPU capacity in AI-RAN systems at both micro-scale--across transmission slots within a cell site--and macro-scale--across sites. Our analysis finds that 40-85% of GPU capacity is unused; although this capacity is temporally bursty at individual sites, it is spatially complementary across sites. To safely and efficiently harness these resources, we present Weaver, a system that opportunistically trains FMs alongside latency-critical RAN workloads without degrading RAN performance. Weaver adopts a RAN-first design: a spare-compute controller integrated into the MAC scheduler uses compute-aware scheduling to smooth RAN GPU demand and exposes more usable spare GPU capacity. A two-level elastic training framework then adapts to dynamic, heterogeneous spare capacity within and across sites. Experiments on an O-RAN-aligned system prototype show that Weaver creates up to 4.9x more usable spare compute and utilizes up to 83% of the available spare capacity. On a multi-site testbed, Weaver improves training throughput by 2.1-3.7x over baseline approaches.

Abstract: Fine-tuning anguage models (LLMs) with first-order optimizers requires a memory several times larger than that required for inference. Memory-efficient zeroth-order optimization (MeZO) sidesteps this cost by estimating gradients from forward passes only. However, for BitNet architectures, a family of LLMs with ternary {-1,0,1\} weights and 8-bit activations, fine-tuning requires updating full-precision latent weights, and thus the memory footprint of MeZO no longer matches that of inference. A promising solution is to finetune only a subset of the latent weights, but existing sparse zeroth-order (ZO) methods either ignore the ternary structure or require first-order gradient information to build a sparse mask, which is at odds with the purpose of ZO fine-tuning. We propose TerMeZO, a sparse MeZO scheme that exploits the geometry of the ternary quantizer itself to identify the latent weights that are more likely to change values during fine-tuning, at no additional data or memory cost. Our convergence analysis shows that TerMeZO can converge faster than full-parameter MeZO, owing to its optimized reduction of the fine-tuning effective dimension. We run extensive experiments on BitNet models ranging from 1B to 3B parameters, spanning classification, instruction-following, and mathematical reasoning tasks. TerMeZO matches or exceeds the performance of full-parameter MeZO while substantially reducing the fine-tuning memory footprint.

Abstract: Onboard vision-language models could enable satellites to answer queries directly, but exhaustive tiled inference over high-resolution imagery is slow and energy-intensive. We identify answer-invariant token redundancy (AITR): image tiles and vision tokens that can be removed without changing the final answer. We present Rift, a two-stage system that performs query-conditioned tile pruning followed by elastic prefill to reduce token budget. We evaluate it on LLaVA-1.5 7B running on Jetson AGX Orin. Compared with exhaustive tiled inference, Rift reduces energy by 78% and latency by 69%, while increasing accuracy from 45% to 73%.

Abstract: Sink-aware deployment may identify important first-token attention heads before a model is quantized, then reuse that map at the edge. We test when this shortcut is safe for 4-bit NF4 weight-only post-training quantization (PTQ). Our Sink Topology Consistency (STC) metrics separate global rank preservation, top-$k$ set overlap, and layerwise sink-mass shift, and distinguish per-input sensitivity from calibration-map transfer. Across Qwen2.5-0.5B, Qwen2.5-1.5B, and Llama-3.2-1B, global bf16-to-4-bit ranks remain high at 4,096 tokens ($ρ_s \geq 0.980$), yet top-$k$ Jaccard overlap is only 0.619-0.793, corresponding to 76.5-88.5% membership retention. The global statistic also masks local failures: terminal Qwen layers shift by 6.2-7.9x their model means, whereas Llama-3.2-1B shows low, nearly uniform drift. Under a C4-to-LongBench shift, cross-domain overlap degrades more than the within-domain precision comparison for both Qwen models, but not for Llama-3.2-1B. Matched-domain 4-bit recalibration reaches 90% of a split-half stability plateau at the smallest tested $n=8$ for both Qwen models and $n=32$ for Llama-3.2-1B, though not as a sharp threshold; for the two Qwen models, updating only selected layers does not reach the full-map stability criterion. On Jetson Orin NX, the 16-sample workload takes seconds for the two models with valid on-device sink measurements. The practical message is precise: global rankings often transfer, but discrete head sets, layer-local policies, and cross-domain calibration should be revalidated after quantization.

Abstract: Diffusion-based Vision-Language-Action (VLA) models achieve strong performance in embodied tasks, but their iterative sampling imposes heavy computational and memory-access cost, blocking real-time deployment on edge platforms. Existing acceleration methods either require expensive training (e.g., distillation, flow matching) or degrade perception via statically scheduled pruning and caching, ignoring the dynamic workload variance of robotic interactions. This paper presents FASA (Feedback-Aware Sampling Adaptation), a training-free runtime framework that treats real-time multimodal feedback as a control signal for the denoising pipeline: an interaction-driven range adaptor modulates the global sampling-step budget based on visual and gripper-force feedback, and a proprioception-aware step adaptor pinpoints the optimized step within the adapted range. This co-designed framework allows the underlying hardware architecture to adaptively match the workload demands of different execution phases. Comparative evaluations across several benchmarks show that the inference speed can be increased by up to 1.45$\times$ while maintaining competitive success rates, providing a novel dynamic runtime architecture paradigm for deploying heavy generative embodied AI workloads onto resource-constrained computing platforms.

Abstract: Convolutional Neural Networks (CNNs) are increasingly deployed in safety-critical edge applications, where soft errors can silently corrupt inference outputs and lead to unsafe decisions. Such applications typically rely on resource-constrained embedded GPUs, requiring fault detection and mitigation techniques that add minimal compute, memory, and latency overhead while integrating seamlessly with the standard GPU inference pipeline. Existing algorithm-based fault tolerance techniques rely on matrix augmentation and per-operation checksum verification, imposing substantial overhead that is prohibitive for CNN inference on embedded GPUs. In this work, we propose carry-through checksum, a fundamentally new scheme for soft-error detection in CNN inference on embedded GPUs. The method embeds dedicated carry-through filters into the convolutional layers, which compute a checksum from the CNN's own operations and propagate it through inference, enabling end-to-end error detection with a single output verification. Experimental results on multiple CNN architectures show that the proposed method detects 95.86% and 86.56% of critical faults for FP32 and FP16, respectively, at almost no additional per-image overhead. Detected faults are mitigated through re-execution, incurring only 2.27% run-time overhead across the entire test set on an NVIDIA Jetson Orin NX GPU.

Abstract: Vision language action (VLA) policies continue to grow in parameter count, making deployment on resource-constrained robot platforms difficult. The central goal is to reduce the number of LLM-side parameters retained in the deployed policy while preserving downstream task performance. Our approach, AdaDE, adapts selected dense feed forward blocks into mixture of experts (MoE) layers and derives expert retention masks from router statistics during fine tuning. The Dense2MoE conversion preserves the original dense FFN function at initialization, so expert deactivation can start without a separate recovery stage. Instead of using a fixed shutdown rule, expert masks are updated dynamically from router usage statistics, with staged training and expert protection to avoid early collapse. With 40% of the LLM parameters deactivated, AdaDE retains 95.1% average success in LIBERO and 42.0% average success across all 50 RobotWin2.0 tasks. These results suggest that dense to MoE adaptation with dynamic expert deactivation is a practical direction for reducing active VLA model size without severe performance loss.

Abstract: Dynamic-vision-sensor (DVS) cameras generate events on a per-pixel basis with a $μ$s-level temporal resolution, calling for new algorithm-hardware co-design approaches compared to standard frame-based vision. While event-driven graph neural networks (EV-GNNs) emerge as a promising algorithmic solution, they raise new HW challenges by mixing dense-regular compute operations and sparse-irregular memory accesses. We present ETHEREAL, the first EV-GNN accelerator that scales to 640$\times$480 resolutions, thanks to a neighbor-parallel spline convolution engine and a 2D/3D-split memory hierarchy with a novel region-of-interest spatiotemporal caching mechanism. Measurement results demonstrate end-to-end inference with 25.6$μ$s latency and 1.7$μ$J energy per event on state-of-the-art workloads

Abstract: Compute-in-memory (CIM) accelerators for spiking neural networks (SNNs) offer a promising solution for achieving $μ$s-level inference latency and ultra-low energy in edge vision applications. However, their limited flexibility at both circuit and system levels restricts their deployment across diverse workloads. This work introduces FlexSpIM, a digital CIM architecture supporting arbitrary operand resolution and shape within a unified storage for weights and neuron states (i.e., membrane potentials). These circuit-level capabilities enable a layer-level hybrid weight- and output-stationary dataflow, maximizing operand reuse and reducing costly on- and off-chip data movement during SNN execution. Measurement results from a fabricated FlexSpIM prototype in 40-nm CMOS demonstrate competitive 1-bit-normalized energy efficiency and higher throughput compared with prior fixed-precision digital CIM-based SNN accelerators, while providing bitwise resolution reconfiguration. Evaluated on the IBM DVS gesture dataset, FlexSpIM achieves 95.8% accuracy while enabling up to 45% energy and 52% latency reductions in large-scale systems compared with fixed stationarity approaches.

Abstract: Large language model (LLM) inference increasingly spans models that differ in size, capability, price, and provider. This shift creates two costs for developers. One is the integration cost of choosing among and switching between many models. The other is the inference cost of rebuilding a KV cache when it is unavailable or incompatible with the selected model. We define and analyze the Unified AI Gateway as a system setting for an edge-deployed AI traffic hub. It coordinates model routing, KV cache management, and compute placement across end devices, edge resources, and cloud model services. At request time, the gateway jointly selects a target model, an execution site, and a KV cache action under task-quality, latency, cost, and resource constraints. In parallel, background cache-management actions optimize KV cache placement, replication, retrieval, and lifecycle decisions for subsequent requests. We synthesize existing evidence on KV cache reuse, compression, cross-model mapping, distributed storage, and transfer, and discuss the remaining challenges of integrating these capabilities into one system. Across eight typical workload profiles, our workload-level analytical simulation reports TTFT speedups of 1.25$\times$--13.28$\times$ and input-cost benefits of 1.20$\times$--6.16$\times$.