Ebook: Intelligent Environments 2024: Combined Proceedings of Workshops and Demos & Videos Session

Intelligent Environments (IEs) enhance physical spaces by integrating advanced technologies from information and communication, sensing and actuating, artificial intelligence, robotics, and human-computer interfaces, among other relevant areas. These enhancements enrich user activities, enable effective management of environmental features, and foster awareness of each technology’s capabilities. The multidisciplinary research community dedicated to IEs explores and delves into the foundational concepts and core ideas of IEs across various contexts and addresses critical challenges. These include enhancing healthcare and well-being, ensuring reliability and privacy, and integrating IEs with advanced computational models such as deep learning and large language models.

The 20th International Conference on Intelligent Environments (IE 2024) took place in Ljubljana, Slovenia, from 17th June to 20th June 2024. This year’s conference included the following workshops:

∙ 13th International Workshop on Intelligent Environments Supporting Healthcare and Well-being (WISHWell 2024)

∙ 13th International Workshop on the Reliability of Intelligent Environments (WoRIE 2024)

∙ 4th International Workshop on Artificial Intelligence and Machine Learning for Emerging Topics (ALLEGET 2024)

The workshops received a significant number of submissions, and after a rigorous review process, only a select number of papers were accepted for presentation and inclusion in the proceedings.

We are particularly proud to acknowledge the longevity and continued relevance of the WoRIE and WISHWell workshops, now in their 13th editions. Additionally, the ALLEGET workshop provided a platform to discuss the latest developments in deep learning and large language models within the context of IEs, highlighting the innovative spirit of our community.

We are also happy to include two submissions from the demos & videos session, highlighting examples of practical implementations of the novel technologies discussed at IE 2024.

Finally, we would like to express our gratitude for the work, time and effort invested by the organizers of the workshops and other events within IE 2024. The conference, including the workshops and the demos & videos session held within it, is ultimately made possible by the authors of the high-quality papers. We thank them for their contributions to the overall enduring success of IE 2024.

The proliferation of sensing device technologies, and the growing demand for data intensive IoT applications, are paving the way to the next wave of transformation in IoT computing systems architecture. The goal today is to design, implement and deploy a seamless interconnection of IoT, edge and cloud resources in one computing system, to form a compute continuum, also referred to as edge-to-cloud. In this talk, compute continuum refers to the deployment and execution of self-adaptive machine learning-based applications employing IoT sensors. Because of their distributed nature over constrained resources devices, these applications leverage the cloud infrastructure for learning tasks while exploiting edge devices for inference tasks on data coming from local IoT sensors. But the next wave of development is already underway; it will involve designing edge-to-edge platforms where learning takes place locally. A coordination platform is used to exchange intelligence between the edges. This talk will be organised as follow: (1) why and what is continuum computing? (2) A comparative study of continuum computing solutions, (3) a cloud continuum platform deployed on edge-to-edge infrastructure and supporting distributed federated learning (FL) applications and (4) an example of a FL based IoT application, for smart grid and renewable energies, deployed on an open source distributed continuum computing solution. This talk is the result of two European projects: SWARM and LASAGNE.

This study proposes a method for home activity recognition solely from the cumulative power consumption data of individual circuits obtained from HEMS distribution boards, recorded every 30 minutes. The proposed method targets seven activities: waking up, going to bed, cooking, laundry, dishwashing, bathing, and personal hygiene, aiming to estimate which activity occurred in each 30-minute time slot. Initially, it identifies the circuits most closely related to each activity. For activities identifiable by the ON/OFF status of appliances, it uses the presence or absence of power consumption in the corresponding circuit to recognize them. For other activities, it constructs models to estimate their presence using machine learning based on specially designed features. Furthermore, it adapts to inter-household differences using transfer learning. We conducted experiments using one year’s HEMS data from 17 households through collaboration with a cooperative company. As a result, we confirmed that it could recognize each of the seven activities with an average F1 score of 0.86. Furthermore, we confirmed that the recognition accuracy of each activity could be improved by performing transfer learning.

Wi-Fi sensing, which can identify human’s action, number of that, and individuals using Wi-Fi signals, is a method that can be introduced easily. The reason for this is possible to operate low cost and protect privacy. However, with only Wi-Fi signal sensing, there are few problems of decreased identification accuracy due to difference in the location of device and in silhouette. Therefore, in this paper, we propose WiAudina to achieve robust identification for environmental changes using Wi-Fi signals with acoustic information from footsteps. In this proposed method, then changing clothing, an average improves about 10% of identification accuracy compared to only Wi-Fi signals or only acoustic information.

Autonomous vehicles (AVs) represent a transformative shift in transportation, promising faster transit, enhanced safety, and reduced accidents. Leveraging the Vehicular Ad-Hoc Network (VANET) for communication among vehicles and roadside units, AVs exchange critical information to optimize driving conditions. However, the constant communication necessitates robust security measures to safeguard both the network and the vehicles themselves. This paper delves into the various cyber threats facing AVs and proposes countermeasures to mitigate them. A comparative analysis identifies prevalent attacks such as Denial of Service (DoS), Sybil, Spoofing, Replay, and Blackhole attacks as the most prominent in AV environments. Subsequently, protocols aimed at thwarting these common attacks are examined. However, it is noted that these protocols may fall short in cases where physical tampering compromises the vehicle’s systems. To address this vulnerability, a novel countermeasure involving cryptographic key management for system access control is proposed and discussed. Additionally, the reliability of each countermeasure is evaluated to ensure robust protection against evolving threats.

Last year witnessed a growing interest from the Business Process Management research community in analyzing activities carried out in sensorized environments using techniques originally intended for business processes. However, activities conducted in such scenarios differ significantly from typical processes in terms of repetitiveness and predictability. This raises the issue of assessing the suitability of state-of-the-art modeling formalisms and mining techniques to represent them, especially when humans are involved. In this paper, we present the results of a user study conducted with this specific goal. Specifically, we analyze the opinions of a group of experts regarding different representation formalisms and mining algorithms, drawing conclusions about the usefulness of such models in smart environments.

Acclimatisation of post-larvae shrimp is critical to guarantee adequate growth in freshwater shrimp farming. Acclimatisation aims to adapt shrimp from their natural habitat (seawater) to the freshwater that is in the pools where they will grow later. Acclimatisation is a challenging process because of the more frequent required monitoring, the potential harm that out-of-limit indicators could cause, and the rapid correction needed to return the monitoring indicators to their specification ranges. This research proposes an IoT-based system integrated with advanced statistical tools in the form of control charts to support the monitoring of the acclimatisation process. The proposal is to be implemented in a shrimp farming company and addresses the negative effects of the existing manual monitoring (operators taking readings and annotating in notebooks manually), the need to monitor acclimatisation proactively (with control charts) and, eventually, reduce the current mortality rate of the process. The paper reports on the design, future implementation, validation and integration with advanced statistical models, challenges, and eventual benefits of the proposed IoT-based system.

Emotions are an essential constituent of well-being. They can be recognized using contact-free sensors such as cameras, based on facial expressions and physiological parameters, such as changes in temperature. We conducted an early evaluation of emotion recognition from RGB cameras using two datasets and high-lighted challenges such as subject-specific relationship between facial expressions and emotions, as well as inconsistent expressions during the same emotional state. Additionally we confirmed the feasibility of measuring subtle changes in temperature between facial regions correlating to different emotional states, using a thermal camera. Finally we proposed ideas for future improvements relating to transfer learning and cross-dataset data curation, which could allow for improvements in performance leading towards practical implementation of well-being monitoring.

One of the major day-to-day problems confronting NHS health workers in the United Kingdom is the one that conflicts with their emotional and mental wellness. This is because of various factors, namely, huge workloads because of higher number of patients, which currently leads to long waiting times, overstretched health facilities, understaffing, stress experienced during the recent COVID-19 pandemic, workplace abuses, bullying, harassments, discrimination, and stigmatization based on race, gender, sexual preferences, and exposure to occupational hazards, amongst others. The aim of this research is to propose a smart health informatics architecture that will be useful in promoting emotional and mental wellness among National Health System (NHS) Workers in the United Kingdom. The aim of this prPoject was achieved through the following objectives: (i) survey of current causes of emotional and mental health challenges and needs to solve the problem, amongst NHS workers in England. (ii) designing of a web-based app to assist healthcare professionals cater for emotional and mental challenges (iii) evaluating the web-based app and (iv) proposing a novel Smart Health Architecture for emotional and Mental Wellness for National Health System (NHS) Workers in England, United Kingdom. The significance of this smart architecture cannot be over emphasized. This newly proposed smart health architecture will assist NHS staff to overcome emotional and mental health challenges, especially during, before and after major global pandemic.

Cervical dystonia, a neurological disorder characterized by involuntary muscle contractions in the neck, presents significant challenges in symptom monitoring due to its fluctuating nature. Current clinical assessments are limited by time constraints and infrequent patient visits. This paper presents the feasibility study of a novel approach to continuous symptom monitoring using a pervasive monitoring system equipped with 9DOF sensors. The system aims to capture the kinematic properties of head movement in patients with cervical dystonia, providing detailed data outside the clinical setting. We outline the development of this system, including the integration of sensors with a mobile application for real-time data logging. Our methodology involves a set of predefined movement protocols, established by practitioners, to standardize data collection. The system’s performance is evaluated using both the sensor data and a reference system, OptiTrack, to ensure accuracy and reliability. Preliminary results confirm the viability of the proposed approach for tracking cervical-dystonia specific head movements.

In this study, we explore the integration of ChatGPT with the Insieme platform, a robust electronic and mobile health system designed as an Italian and Slovenian project. This integration provides a novel way in which users access medical information, offering online support from healthcare professionals and enabling interactions with a sophisticated virtual assistant that utilizes cutting-edge natural language processing technologies. Our paper delves into the specific features of the Insieme platform, presenting a comprehensive explanation of the virtual assistant’s implementation. The incorporation of ChatGPT into this medical platform introduces new solutions and challenges stemming from integrating a chatbot and an integral medical platform, potentially transforming the landscape of the Slovenian healthcare system. Furthermore, we examine the broader implications of this technology in enhancing patient care and optimizing healthcare workflows. Our working prototype provides perspectives on the evolution and future prospects of digital health solutions.

Nowadays, efficient management of energy consumption is crucial for the sustainability of our cities, and overall of our planet. Approaches investigated so far, mostly adopt complex approaches, often based on deep learning, which have an important footprint. This study focuses on the importance of using simpler methods to predict energy consumption in smart buildings, emphasizing a methodological approach that prioritizes simplicity, transparency, and computational efficiency, especially when data is scarce. It emphasizes that even the prediction of energy consumption at the scale of a building, which is sometimes ignored due to computational complexity, is feasible and can make a big difference. By using simple analytical models combined with outlier detection, this research contributes to the field by showing how we can still gain valuable insights with limited data. Therefore, this study provides a practical and scalable way to improve energy efficiency and sustainability in buildings, which has a significant contribution to energy management practices.

The rise of big data and multimodal content—from text and images to videos and audio—demands sophisticated data management and analysis techniques, particularly in intelligent environments. This paper addresses the challenges of managing and analyzing multimodal data within complex Entity-Relationship (ER) models, often structured in normalized forms to enhance efficiency and scalability. We delve into relational learning, emphasizing the integration and manipulation of data tables in third normal form, and explore how these are processed through feature extraction techniques into Machine Learning (ML) models.

Central to our study is the use of the propositionalization algorithm, Wordification, a method that facilitates the straightforward application of general ML algorithms and supports specialized algorithms like PropDRM and PropStar, designed for multi-relational data mining. This research aims to showcase the efficacy of our automated Wordification method for feature extraction, contrasting it with traditional manual approaches.

Our findings indicate that propositionalization significantly simplifies the application of ML algorithms to relational learning, enhancing the processing of multimodal data in intelligent environments. This method significantly improves efficiency and scalability, proving beneficial for rapid deployment and the enhancement of social intelligence through more effective data analysis.

Recent advancements in computer-aided drug discovery revolutionize healthcare by integrating virtual screening (VS) and artificial intelligence (AI). Virtual screening enables the efficient screening of vast chemical libraries in silico, reducing the number of compounds requiring physical testing in the lab before drug synthesis or repurposing. An essential aspect of successful virtual screening is the representation of chemical compounds. While traditionally represented as feature vectors, leveraging convolutional neural networks (CNNs) to interpret chemical structures as images has emerged as a promising approach, harnessing the learning capabilities of CNNs. One potential application of CNNs is in creating classifiers capable of accurately distinguishing between drugs and decoys. These classifiers could serve as a foundation for developing generative adversarial neural networks (GANs), facilitating the synthetic generation of potential non-toxic drugs. This study, which attempts to serve as a basis for future work in the field of smart health, assesses a selection of pre-trained CNNs for their efficacy in classifying drugs associated with diabetes, cancer, and malaria. To enhance model training, a data augmentation phase has been incorporated, introducing variations to the initial images to impart rotational invariance to the learning process. Results indicate that DenseNet201 exhibits superior accuracy, albeit with considerable computational time requirements. Surprisingly, excluding data augmentation significantly improves predictive performance across all models, challenging the initial assumptions. Consequently, applying pre-trained CNNs for drug classification is contingent upon specific conditions, necessitating carefully considering augmentation strategies for optimal outcomes.

Large language models have gained extensive research interest in the past few years. They have demonstrated remarkable ability to process and generate human-like text, and have improved performances on various natural language processing tasks. This paper is focused on the prompting techniques and knowledge augmentation techniques for text style transfer tasks. Text style transfer involves the transformation of a given sentence in a stylistically different manner while preserving its original meaning. It requires models to understand and manipulate different aspects such as politeness, formality, and sentiment. This paper provides an overview of several methods for prompting large language models for text style transfer and presents an overview of several methods for knowledge augmentation with a discussion about potential use for text style transfer. Preliminary results on formality transfer using the T5 model are presented to evaluate prompting and knowledge augmentation techniques. The results show that using knowledge augmentation techniques improves the performance compared to models without augmentation, while zero-shot prompting techniques are less effective. This emphasizes the necessity of fine-tuning and incorporating knowledge augmentation for enhanced model performance.