MenTHoL- Machine Learning with Human in the Loop

"Improving the way that machines learn and interact with the world by incorporating human intelligence"


Humans are able to learn from few data, while can easily generalize to unseen data, reason about their decisions and imagine. Machines, have achieved remarkable results in learning the associations between samples, but have limited power as compared to humans in all aforementioned tasks.
Current machine learning approaches (Deep Learning approaches) try to obtain experience by transferring human knowledge via data annotation using enourmous ammounts of data. However, this approach is quite rough and introduces several limitations given that annotation process includes only categorical characterization and scalar values. While the associations between input data and rough annotations can be learnt efficiently, it can be considered that these systems learn a mapping related only to a small portion of the actual description of the data. In this context, this learning approach has an immanent limitation, as the systems are actually never taught neither to reason about data nor to understand higher-level concepts about them, but only a small portion of they description.
The goal of this project is to incorporate humans in both in learning as well as the inference of machine learning algorithms in order to enhance the capabilities of machines to improve the way they learn and reason about the word. In order to achieve this, we propose to transfer the experience of humans into machines using brain signals, human behaviour like gaze, actions, voice and human perceptual annotation. These signals incorporate high-level information which can improve the performance by allowing machines to learn concepts of data. In this context, we study inductive means for transferring human experience, following a synergistic approach between machines and humans.

- Knowledge Transfer between humans and machines
- Boost and improve the learning procedure- learn from few samples, improve generalization- robustness to adversarial examples, learn to reason and imagine
- Build Mechanisms for better collaboration between humans and machines and bidirectional Human Machine Interfaces.
- Medical Applications
- Low-Cost devices for human-machine collaboration
- Attention transfer from humans to machines


University of Patras - UPCV Group :
Prof. Spiros Fotopoulos , Professor at University of Patras
Prof. George Economou , Professor at University of Patras

Dr. Dimitris Kastaniotis , Research Fellow
Dr. Stavros Dimitriadis , Research Fellow

MSc Students:
Mrs. Ioanna Ntinou , MSc student

University of Catania- Perveive Lab Team
Concetto Spampinato , Associate Professor
Simone Palazzo, Post-Doc researcher
Isaak Kavasidis, Post-Doc researcher

Contact us

Prof. Spiros Fotopoulos

Copyright © Electronics Laboratory, Computer Vision Group, University of Patras