Nikolaos (Nikos) Nikolaou

Research Interests

On the machine learning theory side, my interests revolve around the following:

• Causal Inference -- going beyond statistical associations to identifying cause-effect relationships and quantifying effects of interventions
• Learning Theory & Generalization -- characterizing predictive models whose predictions are good beyond the sample they were trained on; guiding learning algorithms towards producing such models
• Ensemble Learning -- combining several weaker predictive models to construct stronger ones
• Information Theory -- quantifying & leveraging the information content of random variables regarding one another
• Feature Selection -- identifying (ideally minimal subsets of) input variables useful for predicting target variables
• Model Selection -- comparing predictive models
• Uncertainty Quantification -- quantifying the uncertainty of predictive models in their predictions
• Multi-modal Data Fusion -- methods for leveraging information from data coming from multiple sources
• Model Interpretability -- obtaining explanations for predictive models' predictions and/or inner workings
• Resource-efficient Learning -- methods for making machine learning algorithms -especially deep neural networks- more data-efficient and/or computationally-efficient

The application areas I worked or I am currently working in, include:

• Emotion recognition from music
• Photovoltaic power generation (Predicting effects of partial shading, modelling solar irradiance variability)
• Adaptive computer memory controller design
• Earth observation
• Astronomy & Planetary Science (exoplanet detection, exoplanet characterization, galaxy classification, inferring galactic redshift)
• Nuclear Fusion (predicting tritium breeding, accelerating fusion reactor plasma simulations)
• Pharmaceutics & Healthcare (medical imaging, bioinformatics focusing on survival modelling for oncology)