Vision impairment increases risks such as social isolation, mobility challenges, and falls. Wearable Augmented Reality (AR) devices with Artificial Intelligence (AI) can enhance sensory perception by enabling real-time recognition of obstacles, assisting visually impaired individuals during street navigation, aiming to reduce the risk of falls. In this paper, we propose a Federated Learning (FL) framework for outdoor obstacle detection using resource-constrained edge AR devices. As such, our proposed framework is designed to optimize energy efficiency by partially fine-tuning a generic pre-trained model originally developed for visually impaired assistance. Our framework is evaluated on a testbed with NVIDIA Jetson Nano, demonstrating that FL improves accuracy over standalone models on each AR device, while achieving an accuracy comparable to centralized approaches, but without the need to transfer the local raw data on each AR device to a central server/cloud to alleviate privacy concerns.

Machine learning (ML) generally requires a substantial amount of data to reach or surpass human-level performance. However, data collection and annotation by experts are known to be costly and time-consuming, which often leads to suboptimal performance for ML algorithms. One approach to tackle this challenge is to adopt patient-annotated data on each patient’s device in a federated learning (FL) setting. However, this approach comes with certain challenges. For instance, in the case of epilepsy monitoring, patient-annotated data is known to involve inaccuracies, i.e., patients may lose consciousness and annotate a seizure with substantial delay compared to the seizure onset. To address this challenge, we propose an FL framework for epileptic seizure detection with noisy patient-annotated data. We evaluate our approach in the case of epileptic seizure detection and show that our proposed method achieves up to 32.63% higher accuracy, 32.95% higher specificity, and 22.28% higher F1 score compared to the model trained on the noisy dataset.

We investigate the intricate relation between robustness of a Deep Neural Network (DNN) model, a typical safety property, and membership inference, a prominent attack on privacy. To this end, we introduce the notion of Patient Membership Inference in the context of personalized health and precision medicine where personalized models are often adopted. Given a set of patients and a model trained using the data of one of them, Patient Membership Inference aims at identifying the patient whose data was used for training. For this, we leverage on the specificities of robustness of the model when considering data from different patients. In contrast to the classical membership inference, where the task is to determine whether a certain sample has been part of the training set, patient membership inference does not assume access to training data. As such, patient membership inference also demonstrates that access to training data is not necessary for membership inference and that membership inference is possible even for well-generalized models, not suffering from overfitting. We evaluate and demonstrate that robustness may be used to infer membership in the context of two healthcare application domains, i.e., epileptic seizure and cardiac-rhythm abnormality detections.

Interpretability has become a crucial component in the Machine Learning (ML) domain. This is particularly important in the context of medical and health applications, where the underlying reasons behind how an ML model makes a certain decision are as important as the decision itself for the experts. However, interpreting an ML model based on limited local data may potentially lead to inaccurate conclusions. On the other hand, centralized decision making and interpretability, by transferring the data to a centralized server, may raise privacy concerns due to the sensitivity of personal/medical data in such applications.In this paper, we propose a federated interpretability scheme based on SHAP (SHapley Additive exPlanations) value and DeepLIFT (Deep Learning Important FeaTures) to interpret ML models, without sharing sensitive data and in a privacy-preserving fashion. Our proposed federated interpretability scheme is a decentralized framework for interpreting ML models, where data remains on local devices, and only values that do not directly describe the raw data are aggregated in a privacy-preserving fashion to interpret the model.

The human brain performs tasks with an outstanding energy efficiency, i.e., with approximately 20 Watts. The state-of-the-art Artificial/Deep Neural Networks (ANN/DNN), on the other hand, have recently been shown to consume massive amounts of energy. The training of these ANNs/DNNs is done almost exclusively based on the back-propagation algorithm, which is known to be biologically implausible. This has led to a new generation of forward-only techniques, including the Forward-Forward algorithm. In this paper, we propose a lightweight inference scheme specifically designed for DNNs trained using the Forward-Forward algorithm. We have evaluated our proposed lightweight inference scheme in the case of the MNIST and CIFAR datasets, as well as two real-world applications, namely, epileptic seizure detection and cardiac arrhythmia classification using wearable technologies, where complexity overheads/energy consumption is a major constraint, and demonstrate its relevance. Our code is available at https://github.com/AminAminifar/LightFF.

Today, wearable technology is frequently used for continuous monitoring of physiological indicators in the health-care domain. However, mobile-health and wearable devices are generally used in ambulatory settings, hence vulnerable to noise. This interferes with the accuracy of Machine Learning (ML) models running on such systems and their decision-making procedures. To address this issue, we first need to identify the presence of noise. In this paper, we propose RecogNoise to detect noisy segments in Electrocardiography (ECG) recordings using heartbeat detection algorithms and ML. We evaluate our approach based on the MIT-BIH arrhythmia database and three types of noise, i.e., Electrode Motion (EM) , Baseline Wander (BW), and Muscle Artifact (MA), with different Signal to Noise Ratios (SNRs). We show that RecogNoise can detect noisy segments with an F1-score of 86.9% and an accuracy of 88.3%.

Stress has become a prevalent issue affecting individuals’ physical and mental well-being. Detecting stress is the first crucial step to managing it and preventing it from causing other health issues. In this paper, we present a new method to improve the performance of detecting stress, using a comfortable to wear sensor, namely Photoplethysmography (PPG), which is embedded virtually in all smartwatches. To this end, we use PPG sensor data from the publicly available wearable stress and affect detection dataset (WESAD). Using new denoising processes, segmentation methods, and key feature extract, we achieve 95.55% accuracy in detecting stress using the Support Vector Machine (SVM) algorithm. Simplifying the process alongside improved accuracy in this paper facilitates smartphone usage as a real-time stress detection, which we plan as future work.

Machine Learning (ML) algorithms are generally designed for scenarios in which all data is stored in one data center, where the training is performed. However, in many applications, e.g., in the healthcare domain, the training data is distributed among several entities, e.g., different hospitals or patients' mobile devices/sensors. At the same time, transferring the data to a central location for learning is certainly not an option, due to privacy concerns and legal issues, and in certain cases, because of the communication and computation overheads. Federated Learning (FL) is the state-of-the-art collaborative ML approach for training an ML model across multiple parties holding local data samples, without sharing them. However, enabling learning from distributed data over such edge Internet of Things (IoT) systems (e.g., mobile-health and wearable technologies, involving sensitive personal/medical data) in a privacy-preserving fashion presents a major challenge mainly due to their stringent resource constraints, i.e., limited computing capacity, communication bandwidth, memory storage, and battery lifetime. In this paper, we propose a privacy-preserving edge FL framework for resource-constrained mobile-health and wearable technologies over the IoT infrastructure. We evaluate our proposed framework extensively and provide the implementation of our technique on Amazon's AWS cloud platform based on the seizure detection application in epilepsy monitoring using wearable technologies.

Today, in many real-world applications of machine learning algorithms, the data is stored on multiple sources instead of at one central repository. In many such scenarios, due to privacy concerns and legal obligations, e.g., for medical data, and communication/computation overhead, for instance for large scale data, the raw data cannot be transferred to a center for analysis. Therefore, new machine learning approaches are proposed for learning from the distributed data in such settings. In this paper, we extend the distributed Extremely Randomized Trees (ERT) approach w.r.t. privacy and scalability. First, we extend distributed ERT to be resilient w.r.t. the number of colluding parties in a scalable fashion. Then, we extend the distributed ERT to improve its scalability without any major loss in classification performance. We refer to our proposed approach as k-PPD-ERT or Privacy-Preserving Distributed Extremely Randomized Trees with k colluding parties.