Papers for

embedded device engineers

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Fault injection attacks weaken reinforcement learning security in hardware

Fault Injection Attacks on Q-Learning: Analysis, Vulnerabilities, and Countermeasures

Abstract: Hardware fault injection attacks threaten machine learning systems, yet their impact on reinforcement learning (RL) has received little attention despite the use of RL in safety-critical applications. This paper presents a security analysis of Q-Learning implementations under voltage and clock glitching attacks on a ChipWhisperer platform, evaluating five variants: Standard, Robust, Resilient, Linear, and Deep Q-Learning. In standard Q-Table implementations, faults corrupt reward computation, action selection, and Q-value updates during learning, and change the selected action during inference. Robust Q-Learning does not prevent these attacks: NaN and infinite values injected during learning persist in the Q-table, and its greedy policy uses the same code as standard Q-Learning. This paper also evaluates a Resilient Q-Learning variant that applies triple modular redundancy with majority voting. It corrects corruption of a single Q-table copy, but inputs shared by the copies, such as the reward, and the voting code itself remain outside its protection. The analysis extends to Linear and Deep Q-Learning, where faults corrupt stored weights during learning and change actions during inference. Moving beyond benchmark environments, this work evaluates an online traffic light controller in a microscopic traffic simulator, where a single injected infinite value degrades accumulated vehicle waiting time by 148x while corrupting under 4% of the Q-table, and where triple modular redundancy corrected the faults injected in our tests, although glitching could still halt the device. Altogether, this work provides a basis for understanding and securing RL deployments against fault injection attacks.

Thu 12 NovCryptography and Security
The gist
Machine learning models that teach themselves to make decisions can be attacked by messing with their hardware, causing them to make faulty choices. The authors studied how these attacks affect a common decision method called Q-Learning on a hardware hacking platform. They found that many types of Q-Learning can be tricked by changing key numbers during learning or decision-making, causing large errors. They also showed that some error-correction methods help but do not fully stop these attacks. This work helps understand how to protect AI systems that learn by trial and error when running on physical devices.
Open → 2611.16151v1

Neuromorphic hardware improves speech recognition with input-dependent gating

Input-dependent Neuromodulation Enhances Mixed-Signal Neuromorphic Hardware

Abstract: Gating mechanisms allow a fixed neural network to adapt its computation to the current input by dynamically modulating its effective parameters. In the brain, neuromodulation uses this principle to move neural circuits between dynamical regimes, and deep learning exploits input-dependent gating, e.g., in gated recurrent networks and self-attention. Input-dependent modulation has also recently been shown to improve spiking neural networks, with particularly strong benefits in small, parameter-limited ones. This makes neuromodulation especially promising for neuromorphic processors at the edge, where area, memory and energy budgets constrain network size, while circuit physics limits parameter precision. Here we demonstrate input-dependent, within-trial neuromodulation of a spiking recurrent neural network on a mixed-signal neuromorphic processor. An off-chip modulator reads the on-chip network's input and activity and adjusts its population-level gains while the chip's analog circuits compute with the physics of the silicon. Neuromodulation raises binary spoken-digit classification accuracy by $+17.3 \pm 10.4$ percentage points, tested across five hardware network instantiations.

Fri 30 OctNeural and Evolutionary ComputingEmerging Technologies
The gist
Small brain-inspired computer chips, called neuromorphic processors, are good at running neural networks efficiently but face limits in size and precision. This paper shows that adding a way for the chip to adjust how it processes data depending on what input it sees greatly improves accuracy in recognizing spoken digits. The authors demonstrate this by using a small off-chip module to guide the chip’s internal parameters dynamically during operation. This method raised the chip’s speech recognition accuracy by over 17 percentage points across multiple test setups.
Open → 2610.41265v1

AI agent workflows improve PyTorch deployment on device accelerators

AIPC: Evidence-Guided Agent Workflows for PyTorch Project Deployment

Abstract: Deploying and optimizing a PyTorch inference project on a dedicated accelerator, especially an on-device NPU, requires more than using a framework device API as with CUDA. The project must be converted into hardware-runtime artifacts and adapted to the target toolchain and application interface. Differences in model formats, quantization flows, operator support, and toolchain interfaces create cross-layer problems spanning PyTorch/ONNX models, toolchains, runtimes, and device execution. LLM Agents can invoke tools to address such problems, but may drift because of insufficient domain knowledge, large configuration and repair spaces, or poor path selection. We propose AIPC (AI Porting & Conversion), an Agent method for deploying PyTorch inference projects. AIPC combines structured Skills with toolchain, runtime, and profiling evidence to drive cross-layer diagnosis and repair, organizing model adaptation, artifact building, on-target execution, accuracy repair, and performance optimization into one deployment flow. We evaluate AIPC on the on-device NPU path of Qualcomm QAIRT across vision, speech, LLM, and diffusion models. In two optimization cases, 6/7 Skill-arm Whisper-Small runs meet predefined accuracy and latency criteria versus 0/6 control runs, while YOLO-Worldv2 achieves a 2.791x end-to-end speedup versus 1.599x for the control. Under a pre-fixed deployment contract, Skill, tool permissions, and check conditions, some runs achieve zero in-run human intervention (0-HITL). We also examine Qwen3.8-27B as a locally deployable model for driving an Agent workflow.

Mon 26 OctSoftware EngineeringArtificial IntelligenceMachine Learning
The gist
Deploying AI models on specialized hardware like on-device NPUs is complicated because models must be converted and adjusted for different hardware and software layers. The authors propose AIPC, a system that guides AI agents with detailed information and skills to help fix and optimize these models automatically. Their approach organizes the whole deployment process including adapting models, building runtime files, running tests, and tuning speed and accuracy. Tests on various AI models showed that AIPC improved accuracy, speed, and reduced human involvement during deployment. They also explored using a large language model to help automate deployment workflows locally.
Open → 2610.34890v1

Transformer models maintain load forecast accuracy with highly limited weights

Extreme Low-State Quantized Transformer for Electricity-Load Forecasting

Abstract: Aggressive low-state weight quantization can substantially reduce neural-network representational flexibility, making predictive performance difficult to preserve when only a few discrete weight states are available. We investigate this regime for Transformer-based electricity-load forecasting using 3-, 5-, 7-, and 16-state weight quantization across feed-forward-only and attention-plus-feed-forward quantization. Two tasks predict either the next hour or the next 24 hours from the preceding 48 hours across two co-primary electricity-consumption datasets. Direct post-training quantization (PTQ) increased mean normalized root-mean-square error (NRMSE) relative to full precision by 6.25 and 7.04 percentage points in Datasets A and B, whereas quantization-aware fine-tuning (QAT-FT) recovered approximately 90.2% and 94.5% of these penalties. Quantization-aware training from random initialization (Scratch-QAT) also produced competitive low-state solutions, with particularly favorable behavior at three states. We further introduce accumulated-update thresholding, which accumulates optimizer-proposed weight changes before commitment. Relative to paired Scratch-QAT controls, this reduced committed shadow-weight updates by approximately 99.7% and discrete state transitions by approximately 77--80%, while mean NRMSE changed by only +0.042 and +0.197 percentage points. These results show that useful Transformer forecasting can be retained under highly discrete weight representations while strongly suppressing target-state activity, supporting a prospective route toward compact, low-activity edge forecasting using multilevel synaptic devices.

Tue 20 OctMachine Learning
The gist
Predicting electricity use is important but usually needs complex computer models. The authors show that Transformer models can still predict future electricity demand well even when their internal numbers are limited to very few choices. They tested different ways to train and fine-tune these simple models and found methods that keep predictions accurate. This approach could help create small, efficient devices that forecast energy use on-site with fewer computing resources.
Open → 2610.28784v1

Smart meters safeguard data integrity using Merkle trees in secure environments

Resource-constrained Remote Attestation of Smart Meters Using Merkle Trees in a Trusted Execution Environment

Abstract: Smart meters are pivotal to the functioning of smart grids, as they measure and transmit energy consumption data that includes sensitive information about household energy usage patterns. While this data offers significant financial benefits, it can also reveal detailed insights into residents' daily routines, making the security of smart meters a critical concern. This paper introduces a novel remote attestation method for smart meters that ensures the integrity and privacy of data exchanged between smart meters and central servers while detecting unauthorized tampering. Our approach structures the communication data into a Merkle Tree-a hash data structure that facilitates efficient data integrity and authenticity verification. All operations involved in constructing the Merkle root and hashing critical data are executed in the secure environment of a Trusted Execution Environment (TEE) using OP-TEE, thereby maintaining the integrity and security of the smart meter's data. Comparing our experimental results with the related works highlights the superiority of our method in terms of resource efficiency, security robustness, and the feasibility of implementation on smart meters and resource-constrained devices, thereby enhancing the overall security and reliability of smart grid systems.

Mon 12 OctCryptography and SecurityHardware Architecture
The gist
Smart meters collect important energy use data but that data can reveal private household habits, so it needs to be protected. The authors created a way to check and protect smart meter data using a special kind of data structure called a Merkle Tree, which helps quickly verify if data is untouched. This method runs inside a secure part of the meter called a Trusted Execution Environment (TEE), keeping operations protected from tampering. Their tests show this approach uses fewer resources and is strong enough to work on meters with limited computing power.
Open → 2610.16373v1

Power side channels expose training data in embedded AI devices

Power Side-Channel Membership Inference Attack on Embedded Machine Learning

Abstract: Membership inference attacks (MIAs) threaten the privacy of machine learning (ML) training data by determining whether a sample was used to train a target model. Existing MIAs rely on model outputs, ranging from prediction probabilities to predicted labels, an assumption that can be restrictive for on-device ML systems with limited or inaccessible outputs. However, suppressing model outputs does not eliminate the data-dependent computations that produce them, which may remain observable through physical side channels. We present PSCMIA, a power side-channel membership inference attack against embedded ML models that can infer membership directly from power traces without requiring prediction probabilities or even the predicted labels. We evaluate PSCMIA across multiple datasets (MNIST, FMNIST, CIFAR10, CINIC10), fully connected (FC) and convolutional neural network (CNN) architectures, and two embedded platforms (STM32F3, XMEGA). PSCMIA achieves ROC-AUC values of up to 0.907 on FC models. For CNN models, the ROC-AUC gap between PSCMIA and probability vector-based shadow MIA ranges from 0.006 to 0.116. Across the FC and CNN evaluations, PSCMIA outperforms label-only MIA in 11 of 16 model-dataset-hardware configurations, demonstrating that physical execution can expose membership information even when conventional model outputs are unavailable through unintended power side-channel leakage.

Wed 7 OctCryptography and SecurityMachine Learning
The gist
Machine learning models are often trained on private data, and attackers can figure out if a specific piece of data was used in training by looking at the model’s outputs. This paper shows that even if you hide the model’s output, attackers can still tell if data was used for training by analyzing the tiny changes in power consumption when the model runs on small devices. The authors created an attack called PSCMIA that listens to these power signals to reveal training membership information. This works across different devices and types of neural networks, showing a new privacy risk in embedded AI systems.
Open → 2610.10909v1

Quantizer-aligned recalibration improves quantized vision transformer accuracy

Test-Time Adaptation of Quantized ViTs via Single-Pass Quantizer-Aligned Recalibration

Abstract: Post-training quantization is a standard route to fitting vision transformers (ViTs) into edge compute and memory budgets, yet quantized models become especially brittle under distribution shift. Test-time adaptation (TTA) addresses such shifts without labels, but most existing approaches are poorly aligned with the constraints of quantized inference. Prevailing TTA methods recover accuracy through backpropagation, while backprop-free methods often still incur overhead from extra forward passes or parameter updates, and lightweight feature- or logit-level methods recover only part of the loss. Across these approaches, a quantization-specific failure mode that amplifies the drop is not directly targeted: under shift, activations occupy frozen quantizers' calibrated ranges differently, distorting their code distribution. We propose Quantizer-Aligned Recalibration (QuAR), a single-pass TTA method tailored to quantized ViTs that neither backpropagates nor updates any model parameters. QuAR recalibrates activations at the input to a frozen quantizer, mapping the test stream's running per-channel statistics back toward the source calibration. On ImageNet-C with ViT-B, QuAR achieves the highest mean accuracy among state-of-the-art backprop-free TTA methods at 3-, 4-, 6- and 8-bit weight/activation precision, outperforming the strongest baseline by 2.28 points at 8 bits and 4.00 at 3 bits, with 46% lower latency and a memory overhead of only 0.17 MB (0.01% of peak inference memory). Analysis and diagnostics trace the gain to a reduced per-channel mismatch at these quantizers, which restores the code distribution the baselines leave unchanged or distort further. A single fixed configuration remains ahead across continual streams, non-i.i.d. label shift, seven out-of-distribution suites, and three other backbones.

Tue 6 OctComputer Vision and Pattern RecognitionArtificial Intelligence
The gist
Vision transformers are models used for image tasks that can run on small devices by using quantization, which shrinks their size but can also make them less accurate when images change a bit from the training data. The authors found that existing methods to fix this problem either slow things down or don’t fully restore accuracy. They propose a new way to adjust the model’s internal settings without changing its parameters or needing extra processing. This adjustment, called Quantizer-Aligned Recalibration, improves accuracy and speed while using very little extra memory.
Open → 2610.08358v1

Vision language models shape robot actions under timing limits

Beyond LLM Serving: Characterizing Vision-Language-Action Workloads for Embodied AI System Design

Abstract: Vision-language-action (VLA) models translate multimodal observations into low-level robot actions. During robot operation, each control period sets an inference deadline, and overruns leave the robot acting on stale observations, reducing task success. Meeting this deadline motivates on-device or nearby edge execution, where a single robot requires batch-1 inference outside the design point of LLM serving systems. Although VLA architectures combine familiar vision-language, autoregressive, and diffusion-style components, their runtime behavior in this batch-1 control setting remains uncharacterized. We characterize four representative VLA models on an edge GPU server and two onboard SoCs, using single-inference profiling and 43,200 closed-loop episodes. Action tensor dimensionality determines whether a stage is memory- or compute-bound, platform balance can shift that bottleneck, and GPU frequency scaling yields a platform-dependent energy-latency sweet spot. In closed-loop operation, overlapping inference with action execution creates an accuracy-speed-energy tradeoff, and no configuration is Pareto-dominant across deployment SLOs. These results guide joint design of VLA model architectures, hardware, and runtime policies.

Sun 4 OctHardware ArchitectureRobotics
The gist
Robots need to quickly understand what they see and say to act in real time. The authors studied how vision-language-action models perform on different hardware when robots must make decisions under strict time limits. They found that how big the action data is changes whether memory or computing speed matters most and showed that energy use and speed must be balanced carefully. The study helps design smarter robots by matching the model, hardware, and timing strategies.
Open → 2610.05062v1

Multicamera dataset enables low cost screen detection and pointing estimation

DABACO: A Multi-Camera Dataset and Benchmark for Screen Localization and Pointing Estimation

Abstract: Screen localization and pointing estimation are key to low-cost interactive devices. Yet developing and evaluating these algorithms requires realistic data: synthetic captures cannot fully reproduce the optical distortion, rolling shutter, motion blur, and display processing of a physical acquisition, and most existing datasets provide static images rather than the video needed to assess continuous pointing. In this paper, we introduce the DABACO Dataset. Developed within the DABACO (Dispositivo Apuntador de BAjo COste, or Low-Cost Pointing Device) project, this dataset supports the development and evaluation of screen detection and camera-based pointing algorithms for embedded systems. It comprises video sequences captured with multiple low-cost camera sensors, including monochrome global-shutter and color rolling-shutter modules, on embedded platforms such as the Raspberry Pi 4B and ESP32-S3. We present a systematic annotation pipeline combining temporary visual watermarking, optical-flow tracking, manual verification, and marker removal. Both the original marked captures and the marker-free images, together with explicit corner annotations and modification masks, are released to support auditing and the study of potential reconstruction bias. In addition, we release an open-source evaluation toolkit with two reference baselines: a classical screen-detection pipeline based on edge and contour geometry, and a closed-vocabulary, off-the-shelf YOLOv8 segmentation model evaluated without dataset-specific training. Both are evaluated on a general-purpose computer using Intersection over Union, corner error, and pointing error, establishing initial reference results for future embedded implementations. Overall, DABACO addresses a gap in existing resources and is intended to support the development and evaluation of new low-cost screen-localization and pointing systems.

Fri 2 OctComputer Vision and Pattern RecognitionMultimedia
The gist
Detecting screens and where someone is pointing at them using cameras is important for making affordable interactive devices. The paper introduces the DABACO Dataset, which contains video clips from multiple inexpensive cameras set up on small computers. This dataset helps researchers and developers test and improve methods for finding screens and estimating pointing gestures, using real video rather than just pictures or synthetic data. The authors also provide tools to evaluate how well these methods work, along with some example approaches as a starting point.
Open → 2610.03928v1

Edge spiking neural networks learn locally and in parallel during sensing

Learning While Inferring: Local and Parallel Learning for Edge SNNs across Sensing Modalities

Abstract: Edge intelligence requires models to sense continuously in real time and to keep adapting on-device, all under tight compute, energy, and memory budgets. Although spiking neural networks (SNNs) enable efficient event-driven inference, standard surrogate-gradient backpropagation (BP) serializes updates and blocks ongoing inference. We investigate Bidirectional Spike-Based Distillation (BSD) as an on-device learning principle that lets edge SNNs learn while inferring. BSD couples a stimulus-driven forward network with an independent target-driven reverse network and aligns their intermediate representations through local objectives. Because the two pathways have disjoint computation graphs until alignment, forward inference, reverse inference, and stage-wise updates can run concurrently. On 25 benchmarks spanning the five sensing modalities of SOUL, BSD stays within 3.8 percentage points of matched BP baselines on average, while only its forward branch is needed at deployment. The learned representations also transfer well to few-shot class-incremental learning without replaying past data. By removing the dense floating-point backward chain, BSD reduces the projected training latency to $0.72\times$ and the estimated training energy to $0.36\times$ that of BP.

Fri 2 OctNeural and Evolutionary Computing
The gist
Devices that process sensory data close to where it's collected need to learn and adapt continually without using too much power or memory. The authors explored a new way for spiking neural networks to update their learning while still making predictions, unlike standard methods that have to pause to learn. Their approach uses two separate networks working at the same time to teach each other locally, letting the device learn continuously and efficiently. This method performs almost as well as traditional ones, uses less energy during training, and can adapt to new tasks with less data.
Open → 2610.03149v1

Hybrid attention models run faster and use less energy on edge NPUs

Empowering Hybrid Attention Models on NPUs

Abstract: Hybrid attention models have emerged as a crucial architecture for Large Language Models (LLMs) (e.g., the Qwen3.5 and Kimi series). Their memory and computational efficiency make them highly attractive for on-device inference, forming a promising synergy with edge Neural Processing Units (NPUs). However, naive execution of these hybrid models on edge NPUs fails to deliver these benefits, often bottlenecking the prefill stage due to severe memory-system inefficiencies and architectural mismatches within the linear attention (LA) layers. We present HA-NPU, the first system to enable efficient hybrid attention LLM inference on edge NPUs without modifying the underlying algorithms. HA-NPU enhances execution efficiency by reorganizing the dataflow of the LA components across three levels: (1) At the core level, it partitions workloads by the head dimension and fuses dependent operators, eliminating cross-core global memory accesses; (2) At the operator level, it reorders execution to consume intermediate tensors immediately, drastically minimizing local-buffer pressure; (3) At the tensor level, it employs dataflow-aware layout planning to minimize transformation overhead between matrix and vector processing units. Compared to competitive baselines, HA-NPU achieves up to 35.95$\times$ LA kernel speedup and 36.14$\times$ energy reduction, delivering up to 2.03$\times$ faster end-to-end request latency. The source code will be made publicly available at https://github.com/yinyuanzhang/HA-NPU

Sat 26 SeptDistributed, Parallel, and Cluster ComputingArtificial Intelligence
The gist
Large language models use a technique called hybrid attention to understand language efficiently, but running these on small edge devices with special chips called NPUs has challenges. The authors found that simply running these models on NPUs is slow and wasteful because of memory inefficiencies and mismatched hardware usage. They created a system called HA-NPU that reorganizes how the model's data and tasks flow through the hardware, making it much faster and more energy-efficient. This lets devices like smartphones or IoT gadgets run large language models more effectively without changing how the model itself works.
Open → 2609.32114v1

Dynamic depth speech enhancers match static models in quality and speed

Does per-frame early exit pay? A compute-matched study of dynamic depth for on-device speech enhancement

Abstract: Deep learning-based speech enhancement is increasingly deployed on-device in hearing aids, headsets, and earbuds. Most of these devices, however, can only accelerate static int8 graphs, so a depth-varying network must be implemented as several graphs, orchestrated by a policy. In this paper, we supervise every intermediate depth of one causal model, then we fine-tune its output heads to guarantee that deeper outputs are never worse than shallower ones. Using this training protocol, we can derive a family of static models that are more Pareto-efficient than their equivalently-sized counterparts trained from scratch on the same budget. Specifically, we achieve up to 0.11 higher PESQ for equivalent compute, and match the best PESQ at 30% less compute. We then quantize the models to int8 and measure the latency-quality frontier on an STM32N6 microcontroller. On VoiceBank-DEMAND, the dynamic enhancer lies on the same frontier as the static models, rather than trading quality for dynamic execution. Running the policy on the companion Cortex-M55 takes only 26 $μ$s per frame, while splitting the enhancer into separate NPU graphs adds 2.2% latency overhead. The cost of dynamic execution is therefore small.

Thu 24 SeptMachine LearningSound
The gist
On-device speech enhancers, like those in hearing aids or earbuds, often run fixed-speed models. The authors trained one deep learning model to work well at multiple depths, allowing it to run faster or slower dynamically without losing quality. They tested these models on small microcontrollers and found that dynamically choosing the model depth per audio frame does not harm performance or add much delay. This means devices can adjust computation on the fly, saving power while maintaining sound quality.
Open → 2609.29867v1

Hybrid key establishment secures session keys with downgrade protection

Transcript-Bound Combiners for Downgrade-Resilient Hybrid Post-Quantum Key Establishment: Definition, Proof, and Embedded-Device Cost

Abstract: Hybrid key establishment runs a post-quantum key-encapsulation mechanism (KEM) alongside a classical Diffie-Hellman primitive, so that the session key stays secure while either component resists attack. This design is now standardized in the Transport Layer Security protocol, Secure Shell, and the Internet Key Exchange, with the standardized module-lattice KEM (ML-KEM) as the post-quantum component. A hybrid KEM secures the derived key, but not the integrity of the negotiation that selects which primitives are used. Full protocols authenticate that negotiation through a handshake transcript; a hybrid KEM deployed as a standalone drop-in primitive, or inside a minimal handshake without transcript authentication, inherits no such guarantee, and an active attacker can strip the post-quantum option. We ask what the key schedule alone must contain to make downgrade resilience a local property of the combiner. We give a game-based definition at the combiner layer and prove a two-sided separation: a combiner that ignores the transcript is downgraded with certainty, whereas one that binds the session key and the confirmation tag to a hash of the transcript blocks every such attempt, up to a term negligible for a 256-bit transcript hash. We also give an explicit strongest-link security bound. Using a calibrated cost model composed from published Cortex-M4 measurements, transcript binding adds one hash per party - about 11.8% of handshake computation but only 1.5% of radio-inclusive energy - and adds no messages or bytes on the wire. Every reported number is produced by a released harness that passes a 30-check validation gate.

Fri 18 SeptCryptography and Security
The gist
Hybrid key establishment combines old and new encryption methods to keep session keys safe, even if one method is broken. The authors found that if the process doesn’t check the entire conversation (transcript), attackers can remove the new, quantum-proof encryption option. They show that including a hash of the transcript in the key generation stops this attack with very high confidence. Their approach adds a small computational cost but no extra messages, making it practical for devices like embedded systems.
Open → 2609.21273v1