Milos Manic

Sleep is increasingly acknowledged as a cornerstone of public health, with chronic sleep loss implicated in preventable injury and deaths. Obstructive sleep apnea (OSA) affects over one billion people worldwide but remains widely under-diagnosed due to dependence on polysomnography (PSG), an overnight, hospital-based intrusive procedure. As an adjunct to the clinical diagnosis of OSA, this paper presents a low-cost, smartphone-based Generative AI agent framework for sleep apnea detection and sleep coaching at the bedside. Powered by an on=device Generative AI model, the four agents of this framework include a classifier, an analyser, a visualiser, and a sleep coach. The key agent activities performed are sleep apnea detection, sleep data management, data analysis, and natural language sleep coaching. The framework was empirically evaluated on a subject-independent hold-out set drawn from a dataset of 500 clinician annotated clips collected from 10 clinically diagnosed OSA patients. Sleep apnea detection achieved an accuracy of 0.89, precision of 0.91, and recall of 0.88, with nightly Apnea–Hypopnea Index (AHI) estimates strongly correlated with PSG-based clinical scores. The framework was further assessed on the performance metrics of computation, latency, memory, and energy usage. The results of these experiments confirm the feasibility of the proposed framework for large-scale, low-cost OSA screening, with pathways for future work in federated learning, noise robustness, and broad clinical validation.

Generative Artificial Intelligence (GenAI) is rapidly transforming industries and organizations through automation and augmentation. Personalization of human–system interaction is a key area that can be significantly advanced through the effective implementation of GenAI. GenAI, positioned as an intermediary between humans and systems, can transform the human experience from the pre-defined, conventional notions of personalization into a dynamic and integrated hyperpersonalization capability. This article presents the zero-shot-to-head-shot hyperpersonalization (Z2H2) framework, which aims to achieve this through the effective adoption of GenAI techniques. It is a domain-neutral framework of three incremental stages named zero-shot, few-shot, and head-shot that gradually increase the level of hyperpersonalization of the human–system interaction. The framework is further represented in a layered system design and the Z2H2 Data Modality Matrix (ZDMM), which systematically maps data types, AI capabilities, and personalization objectives for each stage. The capabilities of the framework are demonstrated in an educational setting, followed by an empirical evaluation using the Open University Learning Analytics Dataset (OULAD). Although this dataset only contains demographic and aggregated clickstream data, which is a subset of attributes relevant to the entire framework, the gradual development of zero-shot-to-head-shot hyperpersonalization is effectively demonstrated and validated on these student interactions.

As Deep Learning (DL) algorithms become more widely adopted in healthcare applications, there is a greater emphasis on understanding and addressing potential privacy risks associated with these models. The purpose of this study is to investigate the privacy vulnerabilities of the Convolutional Neural Network (CNN) classifiers for Electroencephalogram (EEG) data in the Brain-Computer Interfaces (BCIs). Specifically, it focuses on the Membership Inference Attack (MIA), which seeks to determine if data from an individual were used in model training. The novelty of this work lies in its empirical analysis of MIA, specifically by addressing two key challenges that are less common in other domains: 1) datasets that are heterogeneous and 2) spatial-temporal design choices. Motivated by these challenges, we investigate the susceptibility to MIA based on: 1) specifics of the training dataset (number of participants, demographics), and 2) specifics of the CNN (such as architecture, regularization). Our experiments revealed that an adversary with limited knowledge of the model and its training process can compromise the privacy of training participants, noting that the same attack is not effective against deep learning models trained on image and tabular datasets. Some of our findings are: 1) training on diverse participant datasets improves the privacy of the most participants but increases risks of memorization and vulnerabilities for underrepresented groups; 2) regularization is less effective in defending against the MIA for EEG data CNN classifiers when compared to other types of input data; 3) depth and width of model architecture has no impact on membership attack effectiveness. We hope that the presented insights will assist future researchers develop more privacy-aware deep learning based BCI systems.

The accelerated development of Artificial Intelligence (AI) capabilities and systems is driving a paradigm shift in productivity, innovation and growth. Despite this generational opportunity, AI is fraught with significant challenges and risks. To address these challenges, responsible AI has emerged as a modus operandi that ensures protections while not stifling innovations. Responsible AI minimizes risks to people, society, and the environment. However, responsible AI principles and practice are impacted by ‘principle proliferation’ as they are diverse and distributed across the applications, stakeholders, risks, and downstream impact of AI systems. This article presents a systematic review of responsible AI principles and practice with the objectives of discovering the current state, the foundations and the need for responsible AI, followed by the principles of responsible AI, and translation of these principles into the responsible practice of AI. Starting with 22,711 relevant peer-reviewed articles from comprehensive bibliographic databases, the review filters through to 9700 at de-duplication, 5205 at abstract screening, 1230 at semantic screening and 553 at final full-text screening. The analysis of this final corpus is presented as six findings that contribute towards the increased understanding and informed implementation of responsible AI.

The rapid adoption of artificial intelligence (AI) systems, such as predictive AI, generative AI, and explainable AI, is in contrast to the slower development and uptake of robotic AI systems. Dynamic environments, sensory processing, mechanical movements, power management, and safety are inherent complexities of robotic intelligence capabilities that can be addressed using novel AI approaches. The current AI landscape is dominated by machine learning techniques, specifically deep learning algorithms, that have been effective in addressing some of these challenges. However, these algorithms are subject to computationally complex processing and operational needs such as high data dependency. In this paper, we propose a computation-efficient and data-efficient framework for robotic motion intelligence (RMI) based on vector symbolic architectures (VSAs) and blockchain-based smart contracts. The capabilities of VSAs are leveraged for computationally efficient learning and noise suppression during perception, motion, movement, and decision-making tasks. As a distributed ledger technology, smart contracts address data dependency through a decentralized, distributed, and secure transactions ledger that satisfies contractual conditions. An empirical evaluation of the framework confirms its value and contribution towards addressing the practical challenges of robotic motion intelligence by significantly reducing the learnable parameters by 10 times while preserving sufficient accuracy compared to existing deep learning solutions.

Machine learning and deep neural networks (DNNs) have been proposed as a tool to identify anomalies in computer network communications. However, due the obfuscated nature of off-the-shelf machine learning models, their output often does not provide enough information to isolate the source of the anomaly to take corrective measures. In this article, we introduce the network transformer (NeT), a DNN model for anomaly detection that incorporates the graph structure of the communication network in order to improve interpretability. The presented approach has the following advantages: first, enhanced interpretability by incorporating the graph structure of computer networks; second, provides a hierarchical set of features that enables analysis at different levels of granularity; second, self-supervised training that does not require labeled data. The NeT model was evaluated on a set of anomalous scenarios executed in a real industrial control system. The presented approach successfully identified the anomalies, the devices affected, and the specific connections causing the anomalies, providing a data-driven hierarchical approach to analyze the behavior of a cyber network.

Text classifiers are Artificial Intelligence (AI) models used to classify new documents or text vectors into predefined classes. They are typically built using supervised learning algorithms and labelled datasets. Text classifiers produce a predefined class as an output, which also makes them susceptible to adversarial attacks. Text classifiers with high accuracy that are trained using complex deep learning algorithms are equally susceptible to adversarial examples, due to subtle differences that are indiscernible to human experts. Recent work in this space is mostly focused on improving adversarial robustness and adversarial example detection, instead of detecting adversarial attacks. In this paper, we propose a novel approach, explainable AI with integrated gradients (IGs) for the detection of adversarial attacks on text classifiers. This approach uses IGs to unpack model behavior and identify terms that positively and negatively influence the target prediction. Instead of random substitution of words in the input, we select the top p% words with the greatest positive and negative influence as substitute candidates using attribution scores obtained from IGs to generate k samples of transformed inputs by replacing them with synonyms. This approach does not require changes to the model architecture or the training algorithm. The approach was empirically evaluated on three benchmark datasets, IMDB, SST-2, and AG News. Our approach outperforms baseline models on word substitution rate, detection accuracy, and F1 scores while maintaining equivalent detection performance against adversarial attacks.

The United Nations Sustainable Development Goal 11 aims to make cities and human settlements inclusive, safe, resilient and sustainable. Smart cities have been studied extensively as an overarching framework to address the needs of increasing urbanisation and the targets of SDG 11. Digital twins and artificial intelligence are foundational technologies that enable the rapid prototyping, development and deployment of systems and solutions within this overarching framework of smart cities. In this paper, we present a novel AI approach for hypervector approximation of complex manifolds in high-dimensional datasets and data streams such as those encountered in smart city settings. This approach is based on hypervectors, few-shot learning and a learning rule based on single-vector operation that collectively maintain low computational complexity. Starting with high-level clusters generated by the K-means algorithm, the approach interrogates these clusters with the Hyperseed algorithm that approximates the complex manifold into fine-grained local variations that can be tracked for anomalies and temporal changes. The approach is empirically evaluated in the smart city setting of a multi-campus tertiary education institution where diverse sensors, buildings and people movement data streams are collected, analysed and processed for insights and decisions.