Research themes

Data science for understanding evolutionary processes and how life works in general using data of the global fossil record, present day animal communities, climate, vegetation and environemnt models and sensory observations. Research directions include analysis of macroevolutionary processes, large scale reconstruction of paleoclimate, understanding evolution of faunal communities and contexts of human evolution, and providing quantitative insights for mitigating climate change.

Keywords: Red Queen, ecometrics, macroevolution, biospheric change and conservation palaeobiology, human environments

Key publications

• Reconciling taxon senescence with the Red Queen’s hypothesis by by Žliobaitė et al. 2017 in Nature.
• Herbivore teeth predict climatic limits in Kenyan ecosystems by Žliobaitė et al. 2016 in PNAS.
• Concept drift over geological times: predictive modeling baselines for analyzing the mammalian fossil record by Žliobaitė 2019 in Data Mining and Knowledge Discovery.
• Dental functional morphology predicts the scaling of chewing rate in mammals by Žliobaitė and Fortelius 2018 in Journal of Biomechanics.

Press coverage

• Science News Daily - Reconciling taxon senescence with the Red Queen's hypothesis
• New Scientist - Animal teeth record how species live and die
• Yliopisto-lehti - Hammasfossiili on aikakone
• International Business Times - Animal teeth are being used to reveal Earth's ancient weather

Our long tern focus in computational research has been on handling concept drift in machine learning. Traditional machine learning and data mining methods rely on the assumption that data distribution stays the same during model training and operation. As the world is continuously changing, so evolve data that describe it. We develop machine learning methods that can diagnose themselves and adapt to changing data distribution over time.

Keywords: concept drift, change detection, evolving streaming data

Key publications 

• A Survey on Concept Drift Adaptation by Gama et al. 2014 in ACM Computing Surveys.
• Adaptive Preprocessing for Streaming Data by Žliobaitė and Gabrys 2014 in IEEE Transactions on Knowledge and Data Engineering.
• Active Learning with Drifting Streaming Data by Žliobaitė et el. 2014 in IEEE Transactions on Neural Networks and Learning Systems.
• Evaluation methods and decision theory for classification of streaming data with temporal dependence by Žliobaitė et al. 2015 in Machine Learning.
• Optimizing regression models for data streams with missing values by Žliobaitė and Hollmén 2015 in Machine Learning.
• Towards cost-sensitive adaptation: when is it worth updating your predictive model? by Žliobaitė et al. 2015 in Neurocomputing.

Applied data science is about translating human expertise from various domains to computational proxies and then modeling human expertise in a data-driven way. We have worked in data science in various domains, including: palaeobiology, forestry, biomedicine, chemical engineering, hardware design, traffic and mobility analysis, finance, process mining, sales analysis and customer profiling.

Selected publications: 

• Towards hardware-driven design of low-energy algorithms for data analysis by Žliobaitė et al. 2014 in SIGMOD Record.

• Regression models tolerant to massively missing data: a case study in solar radiation nowcasting by Žliobaitė et al. 2014 in Atmospheric Measurement Techniques.

• Online Detection of Shutdown Periods in Chemical Plants: A Case Study by Martin-Salvador et al. 2014 in Procedia Computer Science.

• Predicting respiratory motion for real-time tumour tracking in radiotherapy by Krilavičius et al. 2016 in IEEE CBMS.

• Environmental control of growth variation in a boreal Scots pine stand – a data-driven approach by Kulmala et al. 2016 in Silva Fennica.

• Herbivore teeth predict climatic limits in Kenyan ecosystems by Žliobaitė et al. 2016 in PNAS.

Fairness-aware machine learning studies in which circumstances algorithms may become biased towards or against groups of people, and how to make predictive models free from such biases, when data, on which they are built, may be biased, incomplete, or even contain past discriminatory decisions. 

Selected publications:

• Handling Conditional Discrimination by Žliobaitė et al. 2011 in IEEE ICDM.
• Why Unbiased Computational Processes Can Lead to Discriminative Decision Procedures by Calders and Žliobaitė 2013, a book chapter.
• On the relation between accuracy and fairness in binary classification by Žliobaitė 2015 in FATML workshop.
• Using sensitive personal data may be necessary for avoiding discrimination in data-driven decision models by Žliobaitė and Custers 2016 in Artificial Intelligence and Law.
• Measuring discrimination in algorithmic decision making by Žliobaitė 2017 in Data Mining and Knowledge Discovery.
• Fairness-aware machine learning: a perspective by Žliobaitė 2017 in Arxiv