PhD at KInIT: Focused
on excellence

Machine learning is in the centre of research of artificial intelligence. Many researchers worldwide are dealing with the topics related to machine learning, both in academia and industry. This very dynamic field is characterized with fast transfer of solutions into practical use.

The topics in this domain are defined by premier scientific conferences, where top-class researchers meet, for example ICML (International Conference on Machine Learning), NeurIPS (Advances in Neural Information Processing Systems), IJCAI (International Joint Conference on AI), COLT (Conference on Learning Theory).

This thesis will be advised by an external mentor, who will also define its particular topic.

Interesting research challenges are contained within (but are not limited to) these topics:

• General Machine Learning (e.g., active learning, clustering, online learning, ranking, reinforcement learning, semi-supervised learning, time series analysis, unsupervised learning)
• Deep Learning (e.g., architectures, generative models, deep reinforcement learning)
• Learning Theory (e.g., bandits, game theory, statistical learning theory)
• Optimization (e.g., convex and non-convex optimization, matrix/tensor methods, sparsity)
• Trustworthy Machine Learning (e.g., accountability, causality, fairness, privacy, robustness)

There are many application domains, where advanced machine learning methods can be deployed.

Supervising team

Maria Bielikova

Lead and Researcher

Jana Kosecka

Professor, George Mason University, USA

Branislav Kveton

Principal Scientist, Amazon’s lab, USA

Peter Richtárik

Professor, King Abdullah University of Science and Technology, Saudi Arabia

Martin Takáč

Associate Professor, Mohamed Bin Zayed University of Artificial Intelligence, United Arab Emirates

Peter Tino

Professor, University of Birmingham, UK

Peter Dolog

Associate Professor, Aalborg University, Denmark

The recommender systems are an integral part of almost every modern Web application. Personalized, or at least adaptive, services have become a standard that is expected by the users in almost every domain (e.g., news, chatbots, social media, or search).

Obviously, personalization has a great impact on the everyday life of hundreds of million users across many domains and applications. This results in a major challenge – to propose methods that are not only accurate but also trustworthy and fair. Such a goal offers plenty of research opportunities in many directions:

• Novel machine learning approaches for adaptive and recommender systems
• Off-policy learning
• Explaining recommendations
• Fairness and justice in recommendations
• Biases in the recommendations

There are several application domains where these research problems can be addressed, e.g., search, e-commerce, news, and many others.

Relevant publications:

• V. Bogina, T. Kuflik, D. Jannach, M. Bielikova, M. Kompan, C. Trattner. Considering temporal aspects in recommender systems: a survey. User Modeling and User-Adapted Interaction, 1-39, 2022. https://doi.org/10.1007/s11257-022-09335-w
• M. Kompan, P. Gaspar, J. Macina, M. Cimerman, M. Bielikova. Exploring Customer Price Preference and Product Profit Role in Recommender Systems. In IEEE Intelligent Systems int. Journal, 2021. https://doi.org/10.1109/MIS.2021.3092768

Supervising team

Michal Kompan

Lead and Researcher

Maria Bielikova

Lead and Researcher

Peter Brusilovsky

Professor, University of Pittsburgh, USA

Branislav Kveton

Principal Scientist, Amazon’s lab, USA

Peter Dolog

Associate Professor, Aalborg University, Denmark

Ubiquity of computing is nowadays generating vast datasets potentially usable for training machine learned models. Yet, these data mostly come not annotated, i.e. without labels necessary for their use for supervised (most conventional) training methods (e.g. classification). Conventional acquisition of proper labels using human force is a costly endeavour. The result is that proper labels are either unavailable, few or of poor quality.

To circumvent the issue, many approaches are emerging and are presently researched by many researchers: meta-learning, transfer-learning, weakly supervised learning, zero/one-shot learning, semi-supervised learning. Each of these fields (or combination thereof) presents opportunities for new discoveries. Orthogonal to this, explainability and interpretability of models is an important factor to consider and advances in this regard are welcome (generally in AI and particularly in the mentioned approaches).

There are several domains of applications, where research of methods and models for addressing small labelled data can be applied. These include (but are not limited to) false information detection, auditing of social media algorithms and their tendencies for disinformation spreading, and support of manual/automated fact-checking.

Relevant publications:

• I. Srba, B. Pecher, M. Tomlein, R. Moro, E. Stefancova, J. Simko, M. Bielikova. Monant Medical Misinformation Dataset: Mapping Articles to Fact-Checked Claims. Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval – SIGIR 2022.
• M. Pikuliak, M. Simko, M. Bielikova. Cross-Lingual Learning for Text Processing: A Survey Expert Systems With Applications or its open access mirror. Expert Systems with Applications. Vol. 165, 2021. https://doi.org/10.1016/j.eswa.2020.113765

Supervising team

Maria Bielikova

Lead and Researcher

Ivan Srba

Researcher

The models created in machine learning can only be as good as the data on which they are trained. Researchers and practitioners thus strive to provide their training processes with the best data possible. It is not uncommon to spend much human effort in achieving upfront good general data quality (e.g. through annotation). Yet sometimes, upfront dataset preparation cannot be done properly, sufficiently or at all.

In such cases the solutions, colloquially denoted as human-in-the-loop solutions, employ the human effort in improving the machine learned models through actions taken during the training process and/or during the deployment of the modes (e.g. user feedback on automated translations). They are particularly useful for surgical improvements of training data through identification and resolving of border cases. This is also directly related to explainability and interpretability of models.

Human-in-the-loop approaches draw from a wide palette of techniques, including active and interactive learning, human computation, crowdsourcing (also with motivation schemes of gamification and serious games), and collective intelligence. Each of these fields (or combination thereof) presents opportunities for new discoveries. They border on computer science disciplines such as data visualization, user experience (usability in particular) and software engineering.

The domains of application of machine learning with human-in-the-loop are predominantly those with a lot of heterogeneity and volatility of data. Such domains include online false information detection, online information spreading (including spreading of narratives or memes), auditing of social media algorithms and their tendencies for disinformation spreading, support of manual/automated fact-checking and more.

Relevant publications:

• J. Cegin, J. Simko, & P. Brusilovsky. A Game for Crowdsourcing Adversarial Examples for False Information Detection. 2nd Workshop on Adverse Impacts and Collateral Effects of Artificial Intelligence Technologies – AIofAI 2022
  https://ceur-ws.org/Vol-3275/paper2.pdf
• J. Šimko and M. Bieliková. Semantic Acquisition Games: Harnessing Manpower for Creating Semantics. 1st Edition. Springer Int. Publ. Switzerland. 150 p. https://link.springer.com/book/10.1007/978-3-319-06115-3

Supervising team

Jakub Šimko

Lead and Researcher

Peter Brusilovsky

Professor, University of Pittsburgh, USA

Jana Kosecka

Professor, George Mason University, USA

Peter Dolog

Associate Professor, Aalborg University, Denmark

Thanks to the deployment of new technologies in our daily lives, a huge amount of data of various types is constantly being generated. The created datasets need to be processed efficiently – the information contained in them often supports the correctness and accuracy of decision-making processes. The development of data analysis methods is therefore an important part of IT research. For various reasons, traditional processing methods are not generally applicable, so new approaches need to be sought. They are mostly based on artificial intelligence and machine learning.

Intelligent data analysis has great potential for solving current research tasks in many areas, including the energy domain. This field is undergoing great changes – renewable energy sources, batteries and electromobility convert the nature of the electricity grid and the classic one-way centralized network is becoming a two-way distributed network. This fact raises a number of research questions and problems to solve.

Research topics include:

• optimal management of microgrids (small energy networks with renewable energy sources) enabling energy sharing between customers,
• predicting customer consumption and renewable production needed for efficient microgrid management,
• disaggregation of energy consumption into consumption of individual appliances, which will allow a better understanding of the nature of consumer consumption,
• investigation of anomalies and / or extreme values ​​in the consumption or production of energy from renewable sources.

However, data analysis in a broader sense can be of interest – it is possible to focus on the tasks of prediction, clustering, classification or detection of anomalies in different domains.

Relevant publications: 

• Rozinajova, V., Bou Ezzeddine, A., Grmanova, G., Vrablecova, P., Pomffyova, M.(2020): Intelligent Analysis of Data Streams, Published in: Towards Digital Intelligence Society: A Knowledge-based Approach, Publish date 22 December 2020
• Kloska M., Rozinajova V. (2021): Towards Symbolic Time Series Representation Improved by Kernel Density Estimators. In: Hameurlain A., Tjoa A.M. (eds) Transactions on Large-Scale Data- and Knowledge-Centered Systems L. Lecture Notes in Computer Science, vol 12930. Springer, Berlin, Heidelberg.

Supervising team

Viera Rozinajová

Lead and Researcher

Anna Bou Ezzeddine

Researcher

Gabriela Grmanová

Researcher

Continuous increase of data available world-wide emphasizes the need of its automatic processing and understanding. Particular challenges are posed by the heterogeneous and unstructured nature of text content provided in natural language. Natural language processing (NLP) ranks among the most prospective subfields of artificial intelligence with great potential for innovative applications affecting everyday life.

Recent advances in neural networks and machine learning allowed to push efficiency and scope of natural language understanding and generation forward. Yet, there remain many research challenges related to particular subtasks, application domains and languages. Further research and various resulting phenomena exploration is necessary. Special attention is drawn by the issues of interpretability and transparency of NLP models or by novel paradigms of learning addressing the problem of low-resource languages.

Particularly interesting challenges include, but are not limited to:

• Language models
• Low-resource language processing
• Transfer/multilingual learning
• Fairness, interpretability, transparency, explainability and/or robustness for NLP
• Domain-specific information extraction, text classification
• Visual grounding of natural language, image captioning, multimodal data processing
• Deep learning for NLP

Relevant publications:

• M. Pikuliak, et al. SlovakBERT: Slovak Masked Language Model. Findings of EMNLP 2022. ACL. To appear.
• M. Pikuliak, M. Šimko, M. Bieliková. Cross-lingual learning for text processing: A survey. Expert Systems with Applications, 2020. https://doi.org/10.1016/j.eswa.2020.113765
• P. Korenek, M. Šimko. Sentiment analysis on microblog utilizing appraisal theory. World Wide Web, 2014. https://doi.org/10.1007/s11280-013-0247-z

Supervising team

Marián Šimko

Lead and Researcher

Jana Kosecka

Professor, George Mason University, USA