Pervasive sensing enables diverse wearable event detection (WED) applications, but deploying machine learning models on resource-constrained microcontrollers (MCUs) poses significant challenges, particularly in ensuring prediction reliability under data shifts or out-of-distribution (OOD) inputs. While Uncertainty quantification methods offer a way to assess this reliability, many are computationally prohibitive for MCUs, and detecting multiple events concurrently further exacerbates resource constraints. Addressing these combined challenges, this paper presents an uncertainty and resource-aware framework designed for reliable and efficient multi-event WED on MCUs, significantly extending our preliminary work. The proposed framework achieves this by integrating Evidential Deep Learning (EDL) for efficient, single-pass uncertainty estimation with a novel cascade learning architecture. This architecture promotes resource efficiency via: (i) intra-event sharing using uncertainty-aware early exits within a staged model (shallow, medium, deep), allowing simpler samples to terminate inference earlier; and (ii) inter-event sharing using a multi-head design where multiple event detectors share a common backbone, minimizing overhead. System efficiency is further enhanced through MCU-specific optimizations, including targeted architecture search, quantization, efficient uncertainty operator implementation using standard TensorFlow Lite Micro (TFLM) operations, and library footprint reduction. We conducted extensive experiments on four distinct wearable datasets (Oesense, KWS, ECG5000, and HHAR) and two MCU platforms (STM32F446ZE, STM32H747XI), comparing the proposed framework against strong baselines including Deep Ensembles and Vanilla EDL. Results demonstrate the proposed framework’s effectiveness, achieving competitive accuracy and uncertainty performance (e.g., up to 22% lower NLL than data augmentation) while drastically reducing resource consumption, offering up to 8.64 × faster inference, up to 8.57 × lower energy use, and 55% smaller memory footprint compared to ensemble methods. The proposed framework enables the deployment of reliable, uncertainty-aware multi-event detection on a wider range of low-power MCUs.

Traditional machine learning techniques are prone to generating inaccurate predictions when confronted with shifts in the distribution of data between the training and testing phases. This vulnerability can lead to severe consequences, especially in applications such as mobile healthcare. Uncertainty estimation has the potential to mitigate this issue by assessing the reliability of a model's output. However, existing uncertainty estimation techniques often require substantial computational resources and memory, making them impractical for implementation on microcontrollers (MCUs). This limitation hinders the feasibility of many important on-device wearable event detection (WED) applications, such as heart attack detection. In this paper, we present UR2M, a novel Uncertainty and Resource-aware event detection framework for MCUs. Specifically, we (i) develop an uncertainty-aware WED based on evidential theory for accurate event detection and reliable uncertainty estimation; (ii) introduce a cascade ML framework to achieve efficient model inference via early exits, by sharing shallower model layers among different event models; (iii) optimize the deployment of the model and MCU library for system efficiency. We conducted extensive experiments and compared UR2M to traditional uncertainty baselines using three wearable datasets. Our results demonstrate that UR2M achieves up to 864% faster inference speed, 857% energy-saving for uncertainty estimation, 55% memory saving on two popular MCUs, and a 22% improvement in uncertainty quantification performance. UR2M can be deployed on a wide range of MCUs, significantly expanding real-time and reliable WED applications.

Self-supervised learning (SSL) has shown remarkable performance in computer vision tasks when trained offline. However, in a Continual Learning (CL) scenario where new data is introduced progressively, models still suffer from catastrophic forgetting. Retraining a model from scratch to adapt to newly generated data is time-consuming and inefficient. Previous approaches suggested re-purposing self-supervised objectives with knowledge distillation to mitigate forgetting across tasks, assuming that labels from all tasks are available during fine-tuning. In this paper, we generalize self-supervised continual learning in a practical setting where available labels can be leveraged in any step of the SSL process. With an increasing number of continual tasks, this offers more flexibility in the pre-training and fine-tuning phases. With Kaizen1, we introduce a training architecture that is able to mitigate catastrophic forgetting for both the feature extractor and classifier with a carefully designed loss function. By using a set of comprehensive evaluation metrics reflecting different aspects of continual learning, we demonstrated that Kaizen significantly outperforms previous SSL models in competitive vision benchmarks, with up to 16.5% accuracy improvement on split CIFAR-100. Kaizen is able to balance the trade-off between knowledge retention and learning from new data with an end-to-end model, paving the way for practical deployment of continual learning systems.

Uncertainty quantification is critical for ensuring the safety of deep learning-enabled health diagnostics, as it helps the model account for unknown factors and reduces the risk of misdiagnosis. However, existing uncertainty quantification studies often overlook the significant issue of class imbalance, which is common in medical data. In this paper, we propose a class-balanced evidential deep learning framework to achieve fair and reliable uncertainty estimates for health diagnostic models. This framework advances the state-of-the-art uncertainty quantification method of evidential deep learning with two novel mechanisms to address the challenges posed by class imbalance. Specifically, we introduce a pooling loss that enables the model to learn less biased evidence among classes and a learnable prior to regularize the posterior distribution that accounts for the quality of uncertainty estimates. Extensive experiments using benchmark data with varying degrees of imbalance and various naturally imbalanced health data demonstrate the effectiveness and superiority of our method. Our work pushes the envelope of uncertainty quantification from theoretical studies to realistic healthcare application scenarios. By enhancing uncertainty estimation for class-imbalanced data, we contribute to the development of more reliable and practical deep learning-enabled health diagnostic systems.

Ensembles of deep learning models can be used for estimating predictive uncertainty. Existing ensemble approaches, however, introduce a high computational and memory cost limiting their applicability to real-time biosignal applications (e.g. ECG, EEG). To address these issues, we propose early exit ensembles (EEEs) for estimating predictive uncertainty via an implicit ensemble of early exits. In particular, EEEs are a collection of weight sharing sub-networks created by adding exit branches to any backbone neural network architecture. Empirical evaluation of EEEs demonstrates strong performance in accuracy and uncertainty metrics as well as computation gain highlighting the benefit of combining multiple structurally diverse models that can be jointly trained. Compared to state-of-the-art baselines (with an ensemble size of 5), EEEs can improve uncertainty metrics up to 2× while providing test-time speed-up and memory reduction of approx. 5×. Additionally, EEEs can improve accuracy up to 3.8 percentage points compared to single model baselines.

Recently, sound-based COVID-19 detection studies have shown great promise to achieve scalable and prompt digital prescreening.However, there are still two unsolved issues hindering the practice.First, collected datasets for model training are often imbalanced, with a considerably smaller proportion of users tested positive, making it harder to learn representative and robust features.Second, deep learning models are generally overconfident in their predictions.Clinically, false predictions aggravate healthcare costs.Estimation of the uncertainty of screening would aid this.To handle these issues, we propose an ensemble framework where multiple deep learning models for sound-based COVID-19 detection are developed from different but balanced subsets from original data.As such, data are utilized more effectively compared to traditional up-sampling and down-sampling approaches: an AUC of 0.74 with a sensitivity of 0.68 and a specificity of 0.69 is achieved.Simultaneously, we estimate uncertainty from the disagreement across multiple models.It is shown that false predictions often yield higher uncertainty, enabling us to suggest the users with certainty higher than a threshold to repeat the audio test on their phones or to take clinical tests if digital diagnosis still fails.This study paves the way for a more robust sound-based COVID-19 automated screening system.

For long, we have studied tiny energy harvesters to liberate sensors from batteries. With remarkable progress in embedded deep learning, we are now re-imagining these sensors as intelligent compute nodes. Naturally, we are approaching a crossroad where sensor intelligence is meeting energy autonomy enabling maintenance-free swarm intelligence and unleashing a plethora of applications ranging from precision agriculture to ubiquitous asset tracking to infrastructure monitoring. One of the critical challenges, however, is to adapt intelligence fidelity in response to available energy to maximise the overall system availability. To this end, we present the design and implementation of ePerceptive: a novel framework for best-effort embedded intelligence, i.e., inference fidelity varies in proportion to the instantaneous energy supplied. ePerceptive operates on two core principles. First, it enables training a single deep neural network (DNN) to operate on multiple input resolutions without compromising accuracy or incurring memory overhead. Second, it modifies a DNN architecture by injecting multiple exits to guarantee valid, albeit lower-fidelity inferences in the event of energy interruption. The combination of these techniques offers a smooth adaptation between inference latency and recognition accuracy while matching the computational load to the available power budget. We report the manifestation of ePerceptive in designing batteryless cameras and microphones built with TI MSP430 MCU and off-the-shelf RF and solar energy harvesters. Our evaluation of these batteryless sensors with multiple vision and acoustic workloads suggest that the dynamic adaptation of ePerceptive can increase the inference throughput by up to 80% compared to a static baseline while ensuring a maximum accuracy drop of less than 6%.

Breakthroughs from the field of deep learning are radically changing how sensor data are interpreted to extract the high-level information needed by mobile apps. It is critical that the gains in inference accuracy that deep models afford become embedded in future generations of mobile apps. In this work, we present the design and implementation of DeepX, a software accelerator for deep learning execution. DeepX signif- icantly lowers the device resources (viz. memory, computation, energy) required by deep learning that currently act as a severe bottleneck to mobile adoption. The foundation of DeepX is a pair of resource control algorithms, designed for the inference stage of deep learning, that: (1) decompose monolithic deep model network architectures into unit- blocks of various types, that are then more efficiently executed by heterogeneous local device processors (e.g., GPUs, CPUs); and (2), perform principled resource scaling that adjusts the architecture of deep models to shape the overhead each unit-blocks introduces. Experiments show, DeepX can allow even large-scale deep learning models to execute efficently on modern mobile processors and significantly outperform existing solutions, such as cloud-based offloading.

Microphones are remarkably powerful sensors of human behavior and context. However, audio sensing is highly susceptible to wild fluctuations in accuracy when used in diverse acoustic environments (such as, bedrooms, vehicles, or cafes), that users encounter on a daily basis. Towards addressing this challenge, we turn to the field of deep learning; an area of machine learning that has radically changed related audio modeling domains like speech recognition. In this paper, we present DeepEar -- the first mobile audio sensing framework built from coupled Deep Neural Networks (DNNs) that simultaneously perform common audio sensing tasks. We train DeepEar with a large-scale dataset including unlabeled data from 168 place visits. The resulting learned model, involving 2.3M parameters, enables DeepEar to significantly increase inference robustness to background noise beyond conventional approaches present in mobile devices. Finally, we show DeepEar is feasible for smartphones by building a cloud-free DSP-based prototype that runs continuously, using only 6% of the smartphone's battery daily.