a Marie Skłodowska-Curie Innovative Training Network
Here you will soon find more info on our trainings, courses and other research resources
Current smart home approaches primarily attempt to nudge eco-friendly activities by providing users with real-time electricity usage feedback or gamified challenges but leave the user alone with the actual change of routines. However, there is further potential in using CHT and smart metering data for helping consumers overcoming the barriers of practice change given by their socio-material environment to conduct more sustainable practices. To contribute to this challenge, the objectives of this PhD project are: (1) Identification of consumption practices and domestic routines that have a significant CO2 footprint (2) identification of opportunities to embed smart technologies that foster and become integral parts of future energy efficient and sufficient routines; (3) drive a sustainable interaction design approach to script CHT fostering low-energy outcomes and supporting energy efficient and sufficient consumption practices; (4) participatory development of domestic rearrangements and tools for eco-routinization (5) conducting appropriation studies and investigating spill-over and rebound effects and (6) continuous validation of the findings on living lab participants and virtual lab data including load comparison with the UN’s sustainability development goals 2030 agenda.
(1) Classification of consumption practices and domestic routines with regard to their socio-material structure and their CO2 footprint. (2) List of strategies and measures to change socio-material structures fostering eco-friendly routines. (3) Design case studies about practice-based eco-design. (4) Grounded theory on the appropriation of eco-design in daily life.
Hosting Institution: University of Siegen
Supervision: Prof. Dr. Gunnar Stevens, Prof. Dr. Volker Wulf
Smart metering data is mainly used to nudge eco-friendly activities by providing users with real-time electricity usage feedback. New ML methods offer a great potential to uncover wasteful consumption practices. However, many methods work as black boxes, which are difficult for users to understand. As a result, there is a lack of trust, limited motivation, and shortcoming to increase the energy literacy of end consumers. To contribute to this challenge, the objectives of this PhD project are: (1) To study decision processes of end-consumers with regard to planning and taking eco-friendly investments; (2) to identify and develop pro-environmental investment costs models for the domestic context that takes end-users demands and mental models into account; (3) develop explainable ML models using smart home sensors and load monitoring techniques to identify inefficient appliances and potentials for renovation (4) using external data provided by manufacturers, retailers, and service providers to calculate/predict costs and expected savings of renewing/renovating measures and creation of test data (e.g., based on load profiles) and validation of the model with the respective datasets; (5) participatory development of a decision-support and recommendation system with living lab households; (6) evaluation of the usability and quality of the recommendation system; (8) continuous validation and comparison of the findings and recommendations with low-energy household targets from the EU.
(1) Description of retrofitting practices, the decision processes, and mental models. (2) Definition of user-centered investment costs models. (3) Accuracy results on wasteful energy consumption caused by appliances and domestic infrastructure. (4) Data base of acquisition cost and expected operation cost/saving of appliances and renovation measures. (5) Assessment of the feasibility and effects of the approach on the basis of a prototypical reference implementation and its evaluation.
The goal of this PhD is to investigate different probabilistic and data visualisation methods to eliminate undesirable behaviour of ML models, specifically targeting CHT and energy efficiency at home: (1) using cloud computing and big data technologies process and analyse data from a large number of households to understand patterns of behaviour; (2) identify undesirable outcomes related to CHTs via research in living labs; (3) develop probabilistic models for avoiding any instances of undesirable behaviour (e.g., bias) in AI regression and classification tasks; (4) apply the developed methods in home energy demand setting (5) Iteratively refine the methods via living lab and virtual lab testing.
(1) A dataset of clustered patterns of behaviour in home related to energy efficiency; (2) a methodology for identifying biased outcomes that affect CHT and a list of identified biases in the living labs including their generalisation; (3) new bias-free probabilistic ML models; (4) Implemented & optimised code (as a scalable software platform) embedded into CHT including data management and privacy issues (5) Living and virtual lab result analysis; (6) business case study.
Hosting Institution: Plegma Labs
Supervision: Nikos Ipiotis, Athina Katsaris, Prof. Dr. Nikolaos Doulamis
In collaboration with AI teams, AI-enabled CHT design and technology appropriation: (1) Explore methodologies for the AI design and implementation of smart grids in households (2) involve explicitly different user groups in the design process, with the aim of forming sustainable practices in relation to different types of services and AI technologies (3) Apply existing knowledge in the HCI and AI fields to the area of smart grids and make it accessible to relevant target groups; facilitate for the design the touch points between households, AI CHT, and smart grids (4) Using participatory design methodology, design innovative concepts for the interaction touch points between households and AI, identifying needs, levels, and types of responsibility, transparency and accountability of AI systems.
(1) A design methodology for recognising the diversity of different user groups and practices in smart grid households (2) A scientific article reporting and discussing empirical results from the project (3) Course material to be included on undergraduate levels at university educational programmes; (4) ART AI-enabled technology design recommendation for different diversity groups
Link to the application:
on KTH Job portal
Hosting Institution: KTH Royal Institute of Technology
Supervision: Prof. Dr. Cecilia Katzeff, Prof. Dr. Daniel Pargman
The research question tackled by this ESR will be how to evaluate ‘responsibility’ of AI? Which metrics to use? Whose values to include, how to interpret and prioritise these values? The work will be based on the latest developments in information science, namely deep reinforcement learning, causal inference, and graph-based data representation as well as social interactions between users and technologies at home. Various model interpretability tools will be explored, investigating the level and type of responsibility of AI by identifying major harm caused by AI systems, due to their data-driven nature, including type/amount of data collected and used to train algorithms and features extracted (what do they reveal and to whom). The goal is to come up with a suite of metrics that collectively capture how far the AI system goes in ensuring bias-free, accountable, and interpretable outcomes.
The PhD researcher will evaluate the above potential harms, explore existing metrics in how far they capture ethical implications of technologies, and provide recommendations for analysis and evaluation of ethical implications of the AI systems. To establish evaluation strategies functionally-, human-, and application-grounded evaluation will be used to determine the model accuracy and assess “quality of explanation”.
The objective measures will be tested using subjective user experience (end-user and stakeholder feedback from focus groups and in the living lab environment).
(1) Report on ethical implications of current technological solutions identifying developments, opportunities and challenges; (2) a suite of evaluating strategies how far AI technology goes in ensuring ‘responsible’ and ‘ethical’, ‘bias-free’ outcomes; (3) detailed evaluation methodology based on deliberate workshops, crowd-sourcing and living labs; (4) recommendations for future evaluation approaches and metrics.
Link to the application:
on EURAXESS, on Strathclyde University Job Portal
Hosting institution: Strathclyde University
Supervision: Dr. Vladimir Stankovic, Dr. Lina Stankovic
Recently, a number of scientific papers have shown that it is possible to extract very detailed information about a household’s routine, with no knowledge besides its smart meter data. Whilst this can be useful if the householder, and data owner, has agreed to it, there are also implications of trust, identity and privacy violation if not the case. The amount and value of information, i.e., utility, that can be extracted depends on the granularity of the data; especially at risk are measurements at sampling rates of 1-60secs or higher. This has led to recent policy changes in granularity of data available to third parties, including energy suppliers, to ensure privacy of users. Additionally, in order to ensure trustworthiness of inference on the data, explainability of the AI inference algorithms is critical. This PhD will focus on the privacy, utility and trustworthiness nexus within the context of sustainability inferences and recommendations from smart home technologies. This can include obfuscation methods for protecting low rate (> 0.02 Hz) meter data with the constraint that it is still possible to extract billing information, as well as other more detailed information as co-agreed with the data user. Differential privacy has been proposed to protect appliance usage information in smart metering but can suffer from complexity. Prior to developing the tools, it is important to understand the tasks one may wish to perform on data (e.g., the inferences one may wish to derive from data; the recommendations one may wish to derive from data; level of explainability to maximise trust) along with the tasks one may wish to prevent to be performed on data (e.g., certain inferences that stakeholders may deem to be privacy-invasive).
ESR2 will address the challenging task of co-designing information extraction methods, such as NILM, with stakeholders such that the inference or recommender systems are safe, reliable, trustworthy, legally compliant and ethically sound. Particular objectives are: (1) Develop a deep understanding of utility, i.e., what information (e.g., activities and behaviour patterns) can be extracted from household data (e.g., smart meter, demographics) at different sampling rates and by which AI methods; (2) Develop data obfuscation ML methods bearing in mind utility and privacy trade-offs; (3) Co-design with a range of data owners, from individual householders to building management services, inference methods or recommender systems that address utility, privacy and trust requirements (4) Test and break the designs with full awareness of Responsible Research and Innovation (RRI).
(1) Report on what information can be mined and how, clustering categories of information (utility); (2) Novel approaches for changing the statistical properties of data that ensure that trade-off utility and privacy (utility vs privacy); (3) a suite of explainable information mining methods on ‘secured’ data co-designed with living labs, virtual labs, focus groups and industry that meet RRI criteria (utility vs privacy vs trust).
Supervision: Dr. Lina Stankovic, Dr. Vladimir Stankovic
The research question tackled by this ESR is how to design future smart home technologies to provide algorithmic interpretability, decision-making accountability, and bias-free algorithmic outcomes. In particular, motivated by the fact that social science models need to be responsible, accountable and accurate, the PhD project will exploit causal inference and data modelling and representation approaches to respond to dynamics of collected data to understand the reasoning behind the AI outcomes. The work will focus AI in smart homes with the objectives:
(1) Developing causal inference approaches to understand how different actions and processes affect AI outcomes; (2) incorporating qualitative data into the model; (3) studying effects of various input statistics on the model and identifying ambiguity in the model; (4) studying implications of model interpretability on the AI outcomes.
(1) Mathematical causal inference models; (2) novel data representation models based on coding qualitative data: (3) application in the context smart homes and result interpretation to inform low-carbon technologies.
The objective of this PhD fellowship is to investigate different, machine learning-guided methodologies for a flexible energy approach towards distributed energy-related assets integration in flexibility markets. The ultimate objective will be to integrate the best approaches into an IT solution that i) aggregates and analyzes data from smart home-related Internet of Things (IoT) platforms and smart devices, therefore establishing a comprehensive basis for precise energy disaggregation, ii) considers the outcome of this analysis along with the household habits and comfort levels towards forecasting the energy demand and flexibility, iii) calculates the available capacity of local storage and the forecasted generation of small-scale (residential) renewable sources, and iv) combines the above outcomes towards determining optimal consumption patterns when sourcing a household’s energy flexibility in flexibility aggregation and management schemes and novel energy markets.
The expected results include:
Link to the application:
Intracom SA Telecom Solutions (and Ph.D. enrollment at the National Technical University of Athens, Greece)
Dr Ilias Lamprinos (Intracom Telecom), Prof. Pavlos Georgilakis (NTUA), Prof. A. Doulamis (NTUA)
AI approaches have a high potential to augment CHTs, for instance by automatically detecting and highlighting behaviour patterns or making smart and individualized suggestions for energy footprint reductions. However, such approaches are often black boxes from the perspective of users, being inaccessible to scrutiny and problematic in terms of privacy implications or hidden biases in recommendations and assessments. The project will focus on developing and evaluating XAI for CHT with the aim to strengthen data autonomy and sovereignty of users. For doing so, the PhD project will capitalise on results of information science, while looking more explicitly at implications to CHT design and technology appropriation based on the following activities: (1) Structured literature review of XAI (2) Specification and operationalization to measure the explainability of XAI in domestic environments. (3) Baseline measure: Study of dedicated XAI approaches for the smart home. (4) Co-design of XAI concepts for smart homes and iterative, lab-based development of prototypes for XAI systems. (5) Lab user study to measure the XAI of the prototypes compared to the baseline measure. (6) Field trails (in the living labs) to evaluate the appropriation of XAI in the daily life and its effect to domestic energy practices.
(1) Survey about XAI and their opportunities for smart home technologies. (2) Reliable, practical and theoretical grounded measure to evaluate domestic XAI empirically. (3) Report about the level / goodness of explainability in current XAI approaches (3) Working prototypes of XAI systems with a higher level of explainability that allow different forms of interactive exploration and reflection of the used algorithms and data in the context of smart homes. (5) Empirical validation of the goodness of explainability compared to the baseline measure. (6) Demonstration that a better XAI supports user acceptance and reflecting domestic energy practices.
Link to the application:
EURAXESS, Fraunhofer FIT advert platform
Hosting Institution: Fraunhofer Institute für Applied Information Technology FIT
Supervision: Prof. Dr. Alexander Boden, B.Essing
(1) KI based personal assistants are a recent trend within informatics and have been spreading into both professional domains (e.g., as chat-based service bots) as well as private domains such as households (e.g., as Amazons Alexa or Google Assistant). Due to their embedding in everyday life practices and CHT, such assistants arguably have a potential to provide feedback on energy related routines of users or affect their behaviour in positive ways, e.g., by means of nudging. The aim of this PhD is to explore the possibilities of IPAs for energy-related routines based on the following activities: (1) Empirical analysis of energy-related practices and opportunities for providing awareness about energy use in different contexts (at home, at work, in transport). (2) Participatory development of concepts for intelligent personal assistants that provide integrated feedback on energy consumption and lead to energy reduction. (3) Prototypical implementation of an IPA for interactive energy feedback. (4) Qualitative analysis of the appropriation of the IPA in living labs, covering contexts at home, at work, and mobile. (transport) (5) Systematic concept refinement based on the practice-based studies as well as co-design results.
(1) Empirical report about useful practices and design opportunities for consumption feedback technologies. (2) Design concepts for energy awareness based on IPAs. (3) Working prototype of an adaptable IPA for providing energy consumption feedback in different contexts. (4) Qualitative, comparative evaluation and validation report about the IPAs and appropriation effects. (5) Systematic collection of practice-based design concepts for IPAs for energy awareness.
EURAXESS, Fraunhofer FIT advert platform
(1) Identify how smart metering and producing own electricity (micro-generation) influence domestic routines and habits related to energy consumption on basis of detailed qualitative studies of 15-20 households with prosumption (jointly with AU); (2) Explore changes in timing and performance of energy consuming practices related to appliances and how this relates to different modes of learning (e.g., situated or experiential learning). (3) On basis of these insights, contribute to energy efficiency design strategies targeting prosumers’ management of own demand and generation and validate these through discussions in focus groups with prosumers.
(1) Understanding better how prosumption influences energy consumption in households; (2) New insights into relationships between learning and practice theories; (3) Theoretical and empirical-based contributions to the design of feedback solutions related to appliance use and demand/generation management that better integrate learning dynamics.
Link to the application:
Aalborg University job portal
Hosting Institution: University of Aalborg
Supervision: Dr. Toke Haunstrup Bach Christensen, Prof. Dr. Kirsten Gram-Hansen
The real energy consumption in smart buildings is often significantly higher than estimated consumption. This project will close this ‘energy performance gap’ through a combination of social science research on how people use technology and provide input to engineering, Machine Learning-driven Connected Home Technology (CHT) design approaches to quantify energy usage and understand technological constraints. In particular, the objectives are: (1) Energy reductions within existing building stock are important and challenging; occupants’ interaction with smart automation solutions for heating control is studied in detail in order to contribute to filling the lack of knowledge on interaction between occupant practices, building technologies and feedback mechanisms. (2) Studying the role of CHT and other smart solutions for practices and energy performance. (3) Exploring design implications of the results and wider implications related rebound effect and sufficiency.
(1) A comprehensive literature review on CHT solutions, energy consumption and performance gap; (2) Quantitative study of energy performance gap patterns in new-built (energy efficient) Danish homes on basis smart meter (register) data; (3) Qualitative studies of practices related to CHT, indoor climate control and energy consumption of smart homes (app. 10 dwellings) combined with interviews with developers and designers in order to uncover how the relationships between users, indoor climate control and energy consumption are conceptualized by developers and designers; (3) Relate findings to energy performance gap discussion; engage critically in analysing how the CHT design can improve the interaction between occupants and technical solutions to reduce energy consumption.
Supervision: Prof. Dr. Kirsten Gram-Hansen, Dr. Toke Haunstrup Bach Christensen
This project aims to develop and test the acceptability of smart devices and control systems that ‘script’ energy-demand reductions. ‘Scripts’ describe how technologies are used. Scripts are written through user-technology interactions which define functionality, usage frequency and patterns, and the extent of automation or machine control. The scripts are strongly influential on the energy outcomes of these user-technology interactions. This project is particularly interested in the role of machine learning-supported scripting in different domestic environments and with different user types.
As an example, the project could make use of disaggregated smart meter data to make inferences about when different domestic activities are taking place, and then explore with households in the living labs whether low-energy scripting could intervene in domestic activities in order to reduce energy demand.
This project will involve strong academic interactions between UEA and the University of Strathclyde (with input from Aalborg University). as well as living lab and co-design workshops to explore user perspectives on scripting.
(1) Analysis of disaggregated household energy and activity data;
(2) Trials of activity-based interventions to establish the effectiveness of low-energy scripting;
(3) Confirmatory review of key insights by relevant partners through a multi-country survey.
Link to the application:
The application process closed on 24 Jan 2021
Hosting Institution: University of East Anglia
Supervision: Prof. Dr Charlie Wilson, Dr. Tom Hargreaves
This project will explore the relationships between ML-enabled CHTs and the lived experience of energy poverty. Methodology will be based on qualitative social science approaches, co-design techniques and any relevant insights from ML-enabled CHT. In particular, the objectives are: (1) Undertaking literature review exploring social justice concerns and implications of CHT and how they are envisaged to impact on energy poor households. (2) Undertaking qualitative research with expert stakeholders and involving energy poor households to explore how smart energy technologies can impact upon social justice concerns and the lived experience of energy poverty. (3) Holding multi-stakeholder workshop(s) to validate and reflect on findings, and to co-design principles/guidelines to ensure social justice concerns are addressed in potential smart energy futures. This will include principles that seek to ensure concerns relating to social justice and the lived experience of energy poverty are addressed by and embedded within future ART ML-enabled CHT applications.
(1) Literature review report and conceptual framework on the implications of CHTs for social justice and the energy poor; (2) In-depth analysis of how CHTs impact on the lived experience of energy poverty; (3) Set of principles or guidelines to improve the design, development or impact of CHT in relation to concerns about social justice in smart energy futures.
Link to the application:
on EURAXESS, on FindAPhd
Supervision: Dr. Tom Hargreaves, Prof. Dr Charlie Wilson
Since the energy consumption signals contain aggregate information from multiple sources, it is quite challenging to extract “semantic knowledge” from the total stream. Thus, the research objectives will be: (1) Developing new adaptable deep learning structures, the parameters of which, i.e., the network weights, are automatically updated with respect to the environmental, housing and living conditions (since static deep learning structures such as Convolutional Neural Networks (CNNs) or Deep Belief Networks (DBNs) are not suitable for modelling the highly dynamic energy consumption signals). A critical aspect is the development of a recurrent and adaptable network through for instance the use of NTUA’s novel approach of Adaptable Time Delay Convolutional Neural Networks (TDL-CNN) which model recurrent CNNs to extract longrange dependent features. The idea of TDL-CNN recently published in 68 on fusing different types of tensor-signals for water security purposes. (ii) Semi-supervised deep learning since the overwhelming majority of the incoming data are not labelled (no ground truth) while only a small portion of them is labelled. Semi-supervision intends to transfer knowledge from existing small labelled data to the vast amount of the unlabelled ones so as to improve classification accuracy. (iii) Exploitation of tensor-based algebraic concepts for improving the classification performance. This proves to get reliable results especially when a small number of training samples are available30. (iv) Semantic knowledge extraction and identification of key energy consumption patterns. (v) The identification of new physical phenomena from the energy data by examining the aggregate signals. The goal is to identify a set of different physical phenomena that will enable towards a semantic description of the energy signals. (vi) Experimentation of the research on real-life energy consumption patterns coming from different sources.
(1) Adaptable recurrent deep learning models for energy signal disaggregation. (2) Semi-supervised deep learning for energy signals analysis. (3) Tensorbased algebra for classifying energy signals. (4) Accuracy results on real-life large datasets of different sources and types of buildings. (5) Identification of salient entities and key properties
Link to the application:
Hosting Institution: National Technical University of Athens
Supervision: Prof. Dr. Anastasios Doulamis, Prof. Dr. Nikolaos Doulamis