Real-time multi-label video classification on embedded devices is constrained by limited compute and energy budgets. Yet, video streams exhibit structural properties such as label sparsity, temporal continuity, and label co-occurrence that can be leveraged for more efficient inference. We introduce Polymorph, a context-aware framework that activates a minimal set of lightweight Low Rank Adapters (LoRA) per frame. Each adapter specializes in a subset of classes derived from co-occurrence patterns and is implemented as a LoRA weight over a shared backbone. At runtime, Polymorph dynamically selects and composes only the adapters needed to cover the active labels, avoiding fullmodel switching and weight merging. This modular strategy improves scalability while reducing latency, and energy overhead. Polymorph achieves 40% lower energy consumption and improves mAP by 9 points over strong baselines executing the TAO dataset.

High-resolution Vision-Language Models (VLMs) are widely used in multimodal tasks to enhance accuracy by preserving detailed image information. However, these models often generate an excessive number of visual tokens due to the need to encode multiple partitions of a high-resolution image input. Processing such a large number of visual tokens poses significant computational challenges, particularly for resource-constrained commodity GPUs. To address this challenge, we propose High-Resolution Early Dropping (HiRED), a plug-and-play token-dropping method designed to operate within a fixed token budget. HiRED leverages the attention of CLS token in the vision transformer (ViT) to assess the visual content of the image partitions and allocate an optimal token budget for each partition accordingly. The most informative visual tokens from each partition within the allocated budget are then selected and passed to the subsequent Large Language Model (LLM). We showed that HiRED achieves superior accuracy and performance, compared to existing token-dropping methods. Empirically, HiRED-20% (i.e., a 20% token budget) on LLaVA-Next-7B achieves a 4.7x increase in token generation throughput, reduces response latency by 78%, and saves 14% of GPU memory for single inference on an NVIDIA TESLA P40 (24 GB). For larger batch sizes (e.g., 4), HiRED-20% prevents out-of-memory errors by cutting memory usage by 30%, while preserving throughput and latency benefits.

The growing gap between the increasing complexity of large language models (LLMs) and the limited computational budgets of edge devices poses a key challenge for efficient on-device inference, despite gradual improvements in hardware capabilities. Existing strategies, such as aggressive quantization, pruning, or remote inference, trade accuracy for efficiency or lead to substantial cost burdens. This position paper introduces a new framework that leverages speculative decoding, previously viewed primarily as a decoding acceleration technique for autoregressive generation of LLMs, as a promising approach specifically adapted for edge computing by orchestrating computation across heterogeneous devices. We propose SLED, a framework that allows lightweight edge devices to draft multiple candidate tokens locally using diverse draft models, while a single, shared edge server verifies the tokens utilizing a more precise target model. To further increase the efficiency of verification, the edge server batches the diverse verification requests from devices. This approach supports heterogeneous devices and reduces server-side memory footprint by sharing a single upstream target model across devices. Our initial experiments with Jetson Orin Nano, Raspberry Pi 4B/5, and an edge server equipped with 4 Nvidia A100 GPUs indicate substantial benefits: ×2.2 higher system throughput, ×2.8 higher system capacity, and better cost efficiency, all without sacrificing model accuracy.

Conventional deep learning (DL) model compression methods affect all input samples equally. However, as samples vary in difficulty, a dynamic model that adapts computation based on sample complexity offers a novel perspective for compression and scaling. Despite this potential, existing dynamic techniques are typically monolithic and have model-specific implementations, limiting their generalizability as broad compression and scaling methods. Additionally, most deployed DL systems are fixed, and unable to adjust once deployed. This paper introduces DyCE, a dynamically configurable system that can adjust the performance-complexity trade-off of a DL model at runtime without needing re-initialization or re-deployment. DyCE achieves this by adding exit networks to intermediate layers, thus allowing early termination if results are acceptable. DyCE also decouples the design of exit networks from the base model itself, enabling its easy adaptation to new base models. We also propose methods for generating optimized configurations and determining exit network types and positions for dynamic trade-offs. By enabling simple configuration switching, DyCE enables fine-grained performance-complexity tuning in real-time. We demonstrate the effectiveness of DyCE through image classification tasks using deep convolutional neural networks (CNNs). DyCE significantly reduces computational complexity by 26.2% for ResNet 152 , 26.6% for ConvNextv2 tiny and 32.0% for DaViT base on ImageNet validation set, with accuracy reductions of less than 0.5%. • Effectively compress the computational complexity of deep learning models. • Dynamically Scale any AI model and select a complexity-performance tradeoff point for the model in run-time. • Enable early exiting on any existing deep learning models by attaching tiny exits. • Generates the best early-exit configuration in a multi-exit system for various trade-off preferences.

This paper introduces TinySaver, an early-exit-like dynamic model compression approach which employs tiny models to substitute large models adaptively. Distinct from traditional compression techniques, dynamic methods like TinySaver can leverage the difficulty differences to allow certain inputs to complete their inference processes early, thereby conserving computational resources. Most existing early exit designs are implemented by attaching additional network branches to the model’s backbone. Our study, however, reveals that completely independent tiny models can replace a substantial portion of the larger models’ job with minimal impact on performance. Employing them as the first exit can remarkably enhance computational efficiency. By searching and employing the most appropriate tiny model as the computational saver for a given large model, the proposed approaches work as a novel and generic method to model compression. This finding will help the research community in exploring new compression methods to address the escalating computational demands posed by rapidly evolving AI models. Our evaluation of this approach in ImageNet-1k classification demonstrates its potential to reduce the number of compute operations by up to 90%, with only negligible losses in performance, across various modern vision models.

Wearable Internet of Things (IoT) devices are gaining ground for continuous physiological data acquisition and health monitoring. These physiological signals can be used for security applications to achieve continuous authentication and user convenience due to passive data acquisition. This paper investigates an electrocardiogram (ECG) based biometric user authentication system using features derived from the Convolutional Neural Network (CNN) and self-supervised contrastive learning. Contrastive learning enables us to use large unlabeled datasets to train the model and establish its generalizability. We propose approaches enabling the CNN encoder to extract appropriate features that distinguish the user from other subjects. When evaluated using the PTB ECG database with 290 subjects, the proposed technique achieved an authentication accuracy of 99.15%. To test its generalizability, we applied the model to two new datasets, the MIT-BIH Arrhythmia Database and the ECG-ID Database, achieving over 98.5% accuracy without any modifications. Furthermore, we show that repeating the authentication step three times can increase accuracy to nearly 100% for both PTBDB and ECGIDDB. This paper also presents model optimizations for embedded device deployment, which makes the system more relevant to real-world scenarios. To deploy our model in IoT edge sensors, we optimized the model complexity by applying quantization and pruning. The optimized model achieves 98.67% accuracy on PTBDB, with 0.48% accuracy loss and 62.6% CPU cycles compared to the unoptimized model. An accuracy-vs-time-complexity tradeoff analysis is performed, and results are presented for different optimization levels.

Time-interleaved analog-to-digital converters (TIADCs) are widely used in communication systems due to their exceptionally high sampling rates; however, in real-world applications, the offset, gain, and time-skew mismatches in TIADCs are a significant challenge for the circuit system. This article proposes a forward error correction (FEC)-aided decision feedback blind mismatch calibration for TIADCs in the time-varying channels environment specific to the orthogonal frequency-division multiplexing (OFDM) system. In our proposed approach, we use an FEC decision feedback technique to generate a ground truth reference signal for the purpose of calibration. There are two stages. In the first stage, the offset and gain mismatches are estimated and corrected using standard techniques. In the second stage, an adaptive filter bank corrects the time-skew mismatch directly without the need for any additional calibration hardware. The coefficients of this adaptive filter are continuously adjusted in the background based on an error signal derived from the decision feedback ground truth signal. This calibration algorithm significantly reduces the bit error rate (BER) and improves the system performance. The efficacy of these approaches is validated through comprehensive simulations to attain a performance assessment, quantified by the BER, using a realistic wireless time-varying channel system configuration.

Electrocardiogram (ECG) analysis is widely used in the diagnosis of cardiovascular diseases. This paper proposes an explainable rule-mining strategy for prioritizing abnormal class detection in ECG data. The proposed method utilizes a biased-trained Artificial Neural Network (ANN) with input features derived from an ECG beat sequence and formulates a set of rules at each node of an on-demand tree-like search algorithm. The rule base at each node is derived from a linear combination of the most impactful features identified using gradient analysis in an ANN. The final derived model is an explainable rule-based system that detects abnormal heartbeats based on statistical and morphological features from ECG. The model achieves the target sensitivity, and accuracy with a low run-time complexity through a comprehensive offline rule mining process and is trained using the MIT-BIH Arrhythmia Database. The system achieves an accuracy of 93% with only nine nodes and a test sensitivity of 90% and 80% respectively for VEB and SVEB beat types, when tested on previously unseen ECG data from the INCART database. The model performance and complexity can be easily adjusted based on the real-time resource constraints of a wearable sensor. The model was deployed on an ARM Cortex M4-based embedded device and is shown to achieve a >50% reduction in sensor power consumption when only abnormal beats are wirelessly transmitted. i.e RF transmission is gated using the model output and transmission is disabled when the subject’s ECG is normal. The proposed technique is highly suited for healthcare applications because of its explainability, lower complexity, and real-time flexibility when deployed in the Internet of Things (IoT) enabled wearable edge sensors.

This paper proposes a foreground digital calibration algorithm that estimates and corrects the offset, gain, and time-skew mismatches for time-interleaved analog-to-digital converters (TIADCs) furthermore our algorithm is designed to correct for harmonic distortion introduced by the presence of a nonlinear front-end. We propose a novel simplified non-linear model in place of the more complex conventional Volterra series based structure. The mismatch estimation technique based on the Fast Fourier Transform (FFT) is proposed to estimate the various time-interleaving mismatches simultaneously. A Taylor-based technique is applied to compensate for these mismatches. We also consider the choice of an appropriate time reference for the time-skew correction algorithm by theoretical analysis. The nonlinear distortion correction technique is based on estimating and inverting an assumed$3^{\text {rd}}$order nonlinearity with a fractional delay. To do this, we design a customized filter in an offline process. Our algorithms are designed to operate in any Nyquist zone. The proposed techniques are verified by a Xilinx Zynq UltraScale+ RFSoC ZCU111 evaluation kit containing a 12-bit, 4.096 GHz TI-ADC with 8 sub-ADCs operating in the$2^{\text {nd}}$Nyquist zone. Accordingly, we observed an improvement in SFDR of 14 dB for mismatch calibration alone and up to another 12 dB with nonlinear correction enabled.

Cardiovascular diseases can be detected early by analyzing the electrocardiogram of a patient using wearable systems. In the context of smart sensors, detecting arrhythmias with good accuracy and ultra-low power consumption is required for long-term monitoring. This paper presents a novel cardiac arrhythmia classification method based on antidictionaries. The features are sequences of consecutive slopes generated from the input signal's event-driven processing. The proposed system shows an average detection accuracy of 98% while offering an ultra-low complexity. This antidictionary-based method is also particularly suited to imbalanced datasets since the antidictionaries are created exclusively from heartbeats classified as normal beats.

The accurate detection of heartbeats is of paramount importance in the current healthcare scenario as they act as an indicator for various underlying cardiac conditions and provides an indication of cardiorespiratory fitness. The article presents a novel multimodal data fusion technique using the discrete wavelet transform (DWT) and an application for fusing electrocardiogram (ECG), and photoplethysmogram (PPG) signals to improve beat detection accuracy in ambulatory monitoring using Internet of Things (IoT) sensors. The characteristics of interest from the input signals are first isolated in the wavelet domain and then combined to form a fused feature signal using a weighted average. The weights used are derived from a signal quality index calculation algorithm, suitable for periodic/quasiperiodic signals of different wave morphologies. The peak detection process to identify the heartbeat locations is carried out on the final fused signal. The research evaluates the algorithm performance when different types of noises at varying amplitudes corrupt the ECG and PPG signal inputs, affecting the signal-to-noise ratios (SNRs). The algorithm consistently exhibited a sensitivity of 99.69%, positive predictive value (PPV) of 99.64%, mean beat-to-beat interval relative error of 0.01, and an error spread (corresponding to 90th percentile of relative errors) of 0.02 in the −30 to 50 dB SNR range for all noise scenarios considered. The proposed algorithm exhibits improved detection sensitivities and PPVs under ambulatory conditions compared to state-of-the-art beat detection algorithms and can be used to accurately detect heartbeats where single-channel monitoring tends to fail in IoT devices.

This Innovative Practice Category Work-In-Progress paper describes the innovative way in which a course on hardware/software codesign was conducted fully online during a near-lockdown necessitated by the COVID-19 pandemic. A novel remote lab setup was established in a very limited time, which was used to ensure that students were able to achieve hands-on experience through a project in spite of the course being taught fully online. Students could run the development environment on their own computers and needed to access the online lab setup only for running their program on the actual hardware, which was connected to a light server with the serial and programming ports forwarded over the internet. Lab exercises were modified appropriately to fit the constraints imposed by the remote setup while not compromising on the rigor and desirable course learning outcomes. A wiki-based platform was used for the dissemination of information, scaffolding, collaboration, as well as booking of slots to access the remote lab setup. Zoom video conferencing tool was used for consultations as well as evaluations. The results are encouraging, with students satisfied with the experience gained, without having to compromise on practical knowledge. This opens up the potential to implement such remote lab-based hands-on projects in MOOCs and continuing education scenarios to enhance student learning.

Internet of Things (IoT) enabled wearable sensors for health monitoring are widely used to reduce the cost of personal healthcare and improve quality of life. The sleep apnea-hypopnea syndrome, characterized by the abnormal reduction or pause in breathing, greatly affects the quality of sleep of an individual. This paper introduces a novel method for apnea detection (pause in breathing) from electrocardiogram (ECG) signals obtained from wearable devices. The novelty stems from the high resolution of apnea detection on a second-by-second basis, and this is achieved using a 1-dimensional convolutional neural network for feature extraction and detection of sleep apnea events. The proposed method exhibits an accuracy of 99.56% and a sensitivity of 96.05%. This model outperforms several lower resolution state-of-the-art apnea detection methods. The complexity of the proposed model is analyzed. We also analyze the feasibility of model pruning and binarization to reduce the resource requirements on a wearable IoT device. The pruned model with 80% sparsity exhibited an accuracy of 97.34% and a sensitivity of 86.48%. The binarized model exhibited an accuracy of 75.59% and sensitivity of 63.23%. The performance of low complexity patient-specific models derived from the generic model is also studied to analyze the feasibility of retraining existing models to fit patient-specific requirements. The patient-specific models on average exhibited an accuracy of 97.79% and sensitivity of 92.23%. The source code for this work is made publicly available.

In this article, non-uniformly sampled electrocardiogram (ECG) signals obtained from level-crossing analog-to-digital converters (LC-ADCs) are analyzed for event-driven classification and compression performance. The signal compression results show that it is important to assess the distortion in eventdriven signals when simulating LC-ADC models, especially at lower resolutions and larger quantization steps. The effects of varying the LC-ADC parameters for the application of cardiac arrhythmia classifiers are also assessed using an artificial neural network (ANN) and the MIT-BIH Arrhythmia Database. In comparison with uniformly-sampled data, it is possible to achieve comparable classification accuracy at a much lower complexity with event-driven ECG signals. The results show the best eventdriven model achieves over 97% accuracy with 79% reduction in ANN complexity with signal-to-distortion ratio (S/D)>21dB. For S/D<; 21dB, the best event-driven model achieves 93% accuracy with a 96% reduction in ANN complexity. An open-source event-driven arrhythmia database is also presented.

Over the past several years, the electrocardiogram (ECG) has been investigated for its uniqueness and potential to discriminate between individuals. This paper discusses how this discriminatory information can help in continuous user authentication by a wearable chest strap which uses dry electrodes to obtain a single lead ECG signal. To the best of the authors' knowledge, this is the first such work which deals with continuous authentication using a genuine wearable device as most prior works have either used medical equipment employing gel electrodes to obtain an ECG signal or have obtained an ECG signal through electrode positions that would not be feasible using a wearable device. Prior works have also mainly dealt with using the ECG signal for identification rather than verification, or dealt with using the ECG signal for discrete authentication. This paper presents a novel algorithm which uses QRS detection, weighted averaging, Discrete Cosine Transform (DCT), and a Support Vector Machine (SVM) classifier to determine whether the wearer of the device should be positively verified or not. Zero intrusion attempts were successful when tested on a database consisting of 33 subjects.

Wearable Artificial Intelligence-of-Things (AIoT) devices demand smart gadgets that are both resource and energy-efficient. In this paper, we explore efficient implementation of binary convolutional neural network employing function merging and block reuse techniques. The hardware implemented in field programmable gate array (FPGA) platform can classify ventricular beat in electrocardiogram achieving accuracy of 97.5%, sensitivity of 85.7%, specificity of 99.0%, precision of 92.3%, and F1-score of 88.9% while consuming only 10.5-μW of dynamic power dissipation.

This innovative practice category work in progress paper describes a project in a final-year un-dergraduate course on implementing a neural accelerator on an FPGA for edge computing. In our university, an undergraduate course on Embedded Hardware System Design introduces students to advanced hardware design techniques with the goal of integrating the created hardware into a complete system. Students learn concepts such as high-level synthesis (HLS), logic synthesis and physical design, with an emphasis on FPGA-based designs. They also understand bus systems such as Advanced eXtensible Interface (AXI) to interconnect the various components. The concepts are put into practice through a project. A series of labs provide scaffolding to students through the course of implementing the project. These labs take students systematically through an introduction to hardware-software co-design, hardware design, creation and interfacing of custom co-processors and HLS. Important hardware design and optimization concepts, as well as managing the data interaction between hardware and software were reinforced through the project. The project also provided many students with the opportunity to be introduced to neural networks and machine learning (ML). Quantitative and qualitative results from a survey indicate that students gained a lot of knowledge and experience through the course of the project. The current form of the project streams in data from a local computer to which the FPGA is connected. Future work includes true and direct cloud connectivity and improved use cases for making it a true Internet of Things (IoT) project.

IoT wearable devices are widely expected to reduce the cost and risk of personal healthcare. However, ambulatory data collected from such devices are often corrupted or contaminated with severe noises. Signal Quality Indicators (SQIs) can be used to assess the quality of data obtained from wearable devices, such that transmission/ storage of unusable data can be prevented. This article introduces a novel and generalized SQI which can be implemented on an edge device for detecting the quality of any quasi-periodic signal under observation, regardless of the type of noise present. The application of this SQI on Electrocardiogram (ECG) signals is investigated. From the analysis carried out, it was found that the proposed generalized SQI is suitable for quality assessment of ECG signals and exhibits a linear behavior in the medium to high SNR regions under all noise conditions considered. The proposed SQI was used for acceptability testing of ECG records in CinC Physionet 2011 challenge dataset and found to be accurate for 90.4% of the records while having minimal computational complexity.

This paper presents a low power 3-lead ECG-on-Chip with real-time QRS detection and lossless data compression for wearable wireless ECG sensors. The proposed chip uses a novel approach that embeds the data compression in the heart beat (QRS) detection process leading to improved energy efficiency. The proposed technique achieves an average compression ratio (CR) of 2.15x and a peak detection sensitivity (Se) of 99.58% and positive productivity (+P) of 99.57%. The chip consumes only 960nW for QRS detection and data compression for 2-channel of ECG making it the lowest power chip.