PhD Studentship in DeustoTech - University of Deusto (Bilbao, Spain)

The University of Deusto invites applications for several PhD projects to be performed in DeustoTech. Deusto Institute of Technology -DeustoTech- (http://deustotech.deusto.es/) located in Bilbao (Spain), is a Research Institution of the Faculty of Engineering at the University of Deusto, and was created with the mission of promoting research and postgraduate training in Information Technology and Communications (ICT) through the participation in research projects of interest to society and industry. DeustoTech is looking for promising young researchers in the areas such as Artificial Intelligence, Pervasive Computing and Computer Systems. The positions are directed to master graduates and they are intended to offer three years fellowships.

Grant summary

The grants will have duration of 36 months, with annual renewals. Each 12 months the performance of the doctoral student will be evaluated to check if he/she achieves the PhD research goals established for the period. The exact amounts awarded will be established by the University of Deusto. In the last year call, the annual gross salary was 16,450€ for the first two years and 17,625€ for the third year. The application is open to worldwide research applicants.

A total of 12 PhD grants are offered to be performed in some of the research topics belonging to the ARTIFICIAL INTELLIGENCE area (see annex 1), and a total of 5 PhD grants are offered to be performed in some of the research topics belonging to the PERVASIVE COMPUTING area (see annex 2).

Requirements

Candidates should have a first class or good 2.1 honours degree in Software Engineering, Computer Engineering or Telecommunications Engineering (other equivalent disciplines will be also considered). An appropriate degree at Masters Level will be mandatory in order to access to the PhD program (applicants finishing a Master’s degree along this academic year will also be considered). Proficiency in spoken and written English is desired; knowledge of Spanish is not a requirement. To be eligible, candidates must become a full-time worker at DeustoTech facilities. All qualified candidates will be considered.

Closing date

April 27th, 2020 (applications will be evaluated in order of arrival). The University of Deusto call closes by May 8th, therefore, after April 27th selected candidates will be guided in the application process.

Further details

Applicants should forward their CV and a summary of research interests to the vacancy contact an copy to Alfonso Bahillo at alfonso.bahillo@deusto.es. The specific vacancy of interest for the candidate should be denoted.

Annex 1: Open research topics on artificial intelligence

Topic #IA1: Large-scale optimisation of metaheuristics using TensorFlow or PyTorch

This thesis proposes the development of new parallelisation techniques for metaheuristics based on graph-based computing platforms such as TensorFlow or PyTorch. The aim is to take better advantage of the computing power of new devices (e.g. GPUs, TPUs) and their large-scale availability. Contact: Antonio D. Masegosa (ad.masegosa@deusto.es)

Topic # IA2: Automated Design of Digital Twins for Advanced Manufacturing Systems

The research proposed here aims at developing Artificial Intelligence techniques for the automatic design of off-the-self Digital Models of industrial systems that can be then integrated into Digital Twins. To this end, we will make use of raw data acquired from the system and expert knowledge in the form of modelling guidelines. The selected researcher is expected to use Automatic Algorithm Design techniques that will combine white- and black-box models of the system’s components to build the most appropriate and accurate digital representation. Contact: Antonio D. Masegosa (ad.masegosa@deusto.es)

Topic # IA3: Collaborative and dynamic optimisation of last-mile logistics

The objective of this thesis is to study the use of new models of last-mile logistics optimisation that integrate interrelated problems (e.g. routes, staff planning, infrastructure booking, etc.), that consider dynamism in the objective function and multiple objectives. If possible, the use of the robust optimisation paradigm will be studied. The results of the research to be carried out in this thesis will lead to more realistic and robust last-mile logistics optimisation models and improve the current optimisation methods found in this area. Contact: Antonio D. Masegosa (ad.masegosa@deusto.es)

Topic # IA4: Deep Learning and Explainable AI for a better understanding of human mobility

The research proposed will be focused on the study and design of Explainable Artificial Intelligence methodologies for applications demanding heterogeneous data, considering different spatial, temporal and contextual information in order to provide textual explanations. The research will include the use of recent methodologies such as attention layers, SHAP and model agnostics explanations and they will be applied over different data structures, including graphs and time series. Contact: Enrique Onieva (enrique.onieva@deusto.es)

Topic # IA5: Heterogeneous deep learning architectures for smart city application

The research proposed will be focused on the study and design of ensembles of heterogeneous deep learning architectures for smart city application. This includes the study of the combination of different types of deep learning architectures for dealing with heterogeneous data (temporal, graph connected, image) coming from smart city applications in order to derive high-quality predictions. Contact: Enrique Onieva (enrique.onieva@deusto.es)

Topic # IA6: Optimization methods for collaborative transportation problems

The research proposed will be focused on the study and design of hybridization and parallelization of methods for optimization methods for large scale heterogeneous and collaborative transport scenarios. This includes the study of optimization in novel scenarios derived through new delivery and transport paradigms such as collaborative logistics, car-sharing and dial-a-ride. The developments will consider parallel and distributed computing, in order to provide scalable solutions to high dimensional problems. Contact: Enrique Onieva (enrique.onieva@deusto.es)

Topic # IA7: Machine learning algorithms for object recognition in low-quality images

The research proposed will be focused on the study and design of capsule networks architectures for small object detection in low-quality images. This includes the study of algorithms to improve low-quality and low-resolution images with capsule networks. Then with these improved images different techniques will be studied in order to detect small objects. These methods could be applied in a wide range of fields like surveillance video images. Contact: Contact: Iker Pastor (iker.pastor@deusto.es)

Topic #IA8: Deep learning techniques for embedded industrial applications

The research proposed will be focused on the research and adoption of novel machine learning and deep learning algorithms, to adapt them to devices with limited computation capabilities. The candidate will focus his research on the applicability of such algorithms in embedded systems, with the final objective of solving complex industrial problems under the paradigm of local edge computing in machines, with limited resources). Examples of this type of applications can be: advanced robotics applications using reinforcement learning, collaborative robotics applications in changing environments, or advanced machine vision applications that are not resolvable by existing classical image processing techniques. Contact: Alberto Tellaeche (alberto.tellaeche@deusto.es)

Topic #IA9: Causal modeling for the understanding of the energy demand on the residential and service sectors

Energy efficiency and demand response are among the most important topics that the EU Commision are pushing forwards to accomplish the new regulations established in climate agreements of Paris. Nevertheless, in both cases the role that human beings play a great role that it has not been sufficiently studied. The aim of this open position is to find a researcher with a strong background in behaviour change theories and causal modeling to formally define which are the drivers and barriers that people have in relation to the use of energy. Specifically, the objectives are: to create user models according to socioeconomics and cultural profiles, to define and establish which are the most prominent behaviour change models in the context of energy expenditure (e.g. TPB, VBM, Transtheoretical model, etc.), to take into account the most important determinants as well as external variables (such us weather, costs, source of energy, etc.) to create an holistic model of behaviour, to design and simulate according to standards such as Theoretical domains or Behaviour Change Wheel in order to understand how these interventions may impact over drivers and barriers defined previously (including policy, gamification, nudging and/or persuasion), and to specify which are valid models related to confidence in technology, perceived usefulness and adoption of artificial intelligence that ally with human beings to improve the energy efficiency or perform demand response actions. Contact: Cruz E. Borges (cruz.borges@deusto.es)

Topic #IA10: Data Democratization through Knowledge Graphs for streamlining Decision Making

This thesis aims to tackle the challenge of Data Democratization, i.e. making data sustainable in time and more widely exploitable by both people and machines. This thesis will focus on the “V” of “Variety” corresponding to one of the 4 or more “V” usually defined to qualify Big Data. This thesis will work on bringing together, quality assure, interlink and explore structured and non-structured data from heterogeneous sources, namely personal sensing devices, IoT sensor networks, Open Government Data, user-generated data through apps or social networks and even private data. The thesis should contribute with new approaches and techniques to make sense out of data, progressing from data into knowledge, e.g. to apply Knowledge Graph continuous refinement by applying Hybrid Intelligence techniques. A core area that wants to be tackled is how to interpret human generated data and how to portray information to people in a more understandable manner. The final aim will be to progress on data understandability both by/for machines and people, transiting from Open Data towards Open Knowledge, and, thus, aid in decision making to a wider range of societal sectors. Contact: Diego López-de-Ipiña (dipina@deusto.es)

Topic #IA11: The transition to zero-waste-ecosystems and the rebound effect in a circular economy

According to the World Health Organization the urban population accounted for 54% of the overall population in the world. Cities are becoming as the main consumers of resources and, consequently, where great amounts of waste are generated. Hence, when talking about urban waste, it can represent an opportunity if we are able to convert it into a new resource and face the challenges this brings to policy makers, companies and citizens. In this context, the concept of circular economy is born as a proposal to face this challenge from a global and systemic approach. In fact, it is increasingly common to find manufacturing and service solutions based on it to optimize the use of resources and minimize their environmental impact. However, improvements in resource use efficiencies often result in an increased, rather than a decreased (rebound effect). Consequently, the transition to zero waste ecosystems, fostered by circular economy industrial business models could lead to a rebound effect, potentially compromising the goals of sustainability of these solutions. In this thesis the rebound effect of a collaborative economy solution will be addressed. Contact: Ainhoa Alonso (ainhoa.alonso@deusto.es)

Topic #IA12: SOFC-model: Modelling and optimization of SOFC cells in the agroindustry sector

Some crops as tomatoes have a strong demand on energy as they have to keep the temperature around 20-23 ºC. This mean that in winter they need to heat (normally using oil boilers) and in summer to cool especially in the south of Spain (normally using an industrial heat pump) . Moreover, they also have some relevant electric loads to keep the optimal points in radiation and humidity, lumps and water pumps, respectively. In overall the energy demand supposes over 50% of the total costs of the greenhouse. So, the optimization of the energy demand will have a huge impact in the food costs as well as to improve the food security of some parts of Europe. On the other hand, small producers have a lack of competitiveness due to high cost of this technologies. So, alternative more efficient and low costs distributed technologies will contribute to the rentability of this sector. In this context, the aim of this thesis proposal is to analyse the benefits and possibilities of using SOFC cells as a more efficient and sustainable decentralised and integral solution for the agroindustry sector. To this end, artificial intelligence techniques will be used to model and optimize the SOFC cells according to the needs of the greenhouse considering aspects such as: costs of the energy and crops, degradation of the cell or the environmental impact. Contact: Ainhoa Alonso (ainhoa.alonso@deusto.es)

Annex 2: Open research topics on pervasive computing

Topic #PC1: Passive and wirelessly connected sensors based on Computational RFID tags for enabling IoT solutions

The research proposed will face several challenges to deploy novel CRFID applications such as those related to the transmission of data, the management of the radio signal, or the energy harvesting system. One of the main challenges of CRFID is the need to optimize communications with sensors, arbitrate their responses when there are more than one, and assure the quality of the streaming data. Therefore, the main objective of this thesis will be to develop strategies for STREAMING DATA from passive sensors, optimizing the communication in dynamic scenarios (with arriving, staying and leaving sensor tags). Contact: Hugo Landaluce (hlandaluce@deusto.es)

Topic #PC2: Interpretation of passive RFID sensing values using machine learning techniques

The research proposed will be focused on the design, development, simulation, and implementation of a multidisciplinary system composed of a passive RFID sensing network that provides streaming data from passive sensors. Additionally, these data will be collected by a device, extracted and processed using ad-hoc pattern recognition techniques to provide different services, e.g. moving a robot with the hand gestures using several passive wireless accelerometers. Contact: Alberto Tellaeche (alberto.tellaeche@deusto.es)

Topic #PC3: Inclusive Location-based Services through Advanced Localization Techniques on Smart Devices

Research, development and evaluation of advanced localization techniques making use of the latest smart devices available in the market (i.e. smartphones, wearable devices) with the aim of developing innovative and enhanced location-based services that foster the inclusion of vulnerable people. Technical environment: Signal processing, data fusion, machine learning and artificial intelligence techniques on data coming from satellite navigation systems, inertial sensors and/or RF interfaces, among others. Contact: Luis Enrique Díez (luis.enrique.diez@deusto.es)

Topic #PC4: Sonouroflowmetry in recording of urinary flow

Listening to the patient while urinating without being intrusive but automatically (from the smartwatch) and more comfortable for the patient (in his own home and as many times as he urinates without the need of any additional action) would help to do preventive health. On the one side the actual tests (for those who are at risk) at the hospital are not good enough because the patient does not feel comfortable so they do not urinate as usual. On the other, we want to do a preventive test so we need a non-intrusive solution. The initial proposal is to start with a smartwatch as a device that records the urination sound (what better place for it than to do it from the wrist). That recording can be sent to an app on the mobile via Bluetooth. Subsequently, either from the mobile or from a cloud service (depending on how complex the models and algorithms are) we can do a contrast job with a set of previously established patterns (using Machine Learning algorithms) or trying to model the sound (using statistics) to be able to diagnose the patient based on that urination. The information would go directly to the doctor PC for review/check/validation. In this way we will achieve non-forced uroflowmetry (every user will be comfortable at home while doing his own diagnosis). Contact: Alfonso Bahillo (alfonso.bahillo@deusto.es)

Topic #PC5: Hybrid Intelligence powered analytics for IoP mediated Environments

This thesis proposes the analysis of the potential of smart solutions and environments grounded on the extensive usage of data and analytics through the collaboration of machines, i.e. IoT devices, and humans, approaching the new paradigm of the Internet of People (IoP). It aims to face the challenge of how to make sense out of data through the collaboration of people & machines. For that, Hybrid Intelligence (HI) is proposed which is an approach for combining human and machine intelligence for decision making and data interpretation. HI benefits from the human ability to express and deal with complexities and the automation, availability and accuracy that machines provide. The interception of the following areas to give place to Smart Environments will be explored:

• Machine Learning: ML methods can be used for activity recognition performed to detect what a user is doing, e.g. walking or resting. ML can also be used to correlate user actions into activities that configure behaviours.

• Human Computation: is about involving citizens to annotate the world or correct/validate inferences performed by machines, gamification or persuasive technology can be used to improve and validate the results driven by machines. Active learning methods with Human in the Loop underpin this second concept.

• Knowledge modelling & exploitation: human and machine-driven knowledge can be effectively merged and represented in a Knowledge Graph (KG), combining data gathered from machine learning and human contributions. KGs are exploited to infer new knowledge or build recommenders through transparent and interpretable algorithms

Contact: Diego Casado (dcasado@deusto.es)