Emerging Challenges in Machine Learning

Emerging Challenges in Machine Learning: Discovering Latent Representations from Real-World High-Dimensional Data 
Invited Talk: Mihalis A. Nicolaou

Abstract: In the past years, exponential progress has been witnessed in artificial intelligence, and in particular, machine learning. Fueled by the vast amounts of available data, theoretical advances in deep learning, as well as a significant increase in computational resources, unprecedented results were achieved in several problems across disciplines. Nevertheless, the path to achieving the full potential of machine learning is paved with a set of significant challenges that need to be overcome. In the era of abundant data, researchers are faced with massive, complex, multi-relational datasets that pose challenges in terms of scalability, statistical redundancy, and data quality. Often, datasets are of poor quality, perturbed by various types of spatio-temporal noise, contain incomplete entries and are of high-dimensionality. At the same time, emerging challenges include evaluating for fairness and bias, preserving data privacy, as well as providing interpretable solutions with reproducible and generalizable results. In the light of these challenges, I will discuss my research in machine learning and computer vision, with applications in affective computing and human sensing, health, and intelligent game design.

Bio: Mihalis A. Nicolaou is Assistant Professor at the Computation-based Science and Technology Research Center at The Cyprus Institute. Previously, Mihalis was Lecturer at the Department of Computing at Goldsmiths, University of London, and a Research Fellow with the Department of Computing at Imperial College London. He obtained his MSc and PhD from Imperial College, and completed his undergraduate studies at the University of Athens, Greece. His research interests span the areas of machine learning, computer vision, and human sensing. He is most interested in the analysis and interpretation of multi-sensory, high-dimensional data, often conveyed via visual, auditory, social, and biomedical signals.