TAMEDWhat's Next in Affect Modeling
12/05/2022
We happy to anounce our "What's Next in Affect Modeling?" workshop that will take place at 10th International Conference on
Affective Computing & Intelligent Interaction (ACII-2021). This workshop puts an emphasis on state of the art methods in machine learning and their suitability for advancing the reliability, validity, and generality of affective models. We will be investigating entirely new methods, untried in Affective Computing, but also methods that can be coupled with traditional and dominant practices in affective modeling. In particular, we encourage submissions that offer visions of particular algorithmic advancements for affect modeling and proof-of-concept case studies showcasing the potential of new sophisticated machine learning methods.
RankNEAT: An Evolutionary Search-based Preference Learner
14/04/2022
Have a look at our new paper accepted for publication to top-tier GECCO conference. It proposes an evolutionary search strategy for training preference learners. Our methodology is able to efficiently explore the deceptive and noisy loss landscapes caused by the inherent bias of subjectively defined labels. We apply our methodology to affect modelling tasks -- where affect labelsare by nature subjective -- and the experimental results suggest that it outperforms gradient-based preference learners.
Modeling Affect in the Wild using Privileged Onformation
01/10/2021
Check out our new paper, "Privileged Information for Modeling Affect In The Wild", presented at the 9th International Conference on Affective Computing and Intelligent Interaction (ACII 2021). In our study, we try to address the question, "How can your emotion models work well in the wild (real-world) when lab measurements (physiology, telemetry, etc) are not there anymore?", and the answer comes from the framework of Learning Using Privileged Information.
Keynote Speaker at NIDS 2021
30/09/2021
I was honoured to be the keynote speaker at the 1st International Conference on Novelties in Intelligent Digital Systems. My keynote talk was on "Reasoning from high-order data: Methods and Applications".
During the talk, I had the opportunity to present my research studies on Computer Vision, Machine Learning and Affective Computing, focusing on applications ranging from human-computer interaction to cultural heritage documentation.
AffRankNet+: A Ranking Model that uses Privileged Information
28/09/2021
Check out my new paper, "AffRankNet+: Ranking Affect Using Privileged Information", presented at the "What's next in affect modeling?" workshop within the 9th International Conference on Affective Computing and Intelligent Interaction (ACII 2021). For the first time, this study proposes a machine learning model for ranking problems that exploits privileged information in the form of preference scores to improve its accuracy. The AffRankNet+ model was evaluated on the public available AFEW-VA dataset and the evaluation results sugest that the exploitation of privileged information can significantly improve the performance of the model.
What's Next in Affect Modeling Workshop
28/09/2021
Today Prof. Georgios Yannakakis, Prof. Bjorn Schuller and I run the first workshop on "What's next in affect modeling?" within the 9th International Conference on Affective Computing and Intelligent Interaction (ACII 2021). Very briefly, this workshop is about exploring state-of-the-art machine learning methods for building more reliable, valid and general models of affect. The workshop included the presentation of nine high-quality papers dealing with different aspects of affect modeling, and a keynote talk by Prof. Michel Valstar on "Affective Computing in medical applications: confidence, risk, and ethical issues".
UM Grants Week
06/07/2021
I was a panel member on the second day (dedicated to MSCA actions) of UM Grants Week event. The Grants Week 2021 was organised by RSSD at the University of Malta, MCST, SEA-EU and supported by EU Commission. The event was designed to provide researchers with key insights into a wide range of Funding Opportunities available at National and EU level.
As a panelist I had the opportunity to share my experience as a widening fellow, as well as the research objectives and research challenges of the TAMED project with the audience and future applicants.
Invited Speaker at University of West Attica
03/06/2021
Today I had the opportunity to give a lecture at the University of West Attica, Athens, Greece on processing and analysis of high-order data using machine learning models. The presentation was a journey to my past and current research. I also had the opportunity to present the research objectives and the ressearch challenges of the TAMED projects as well as the project's outcomes so far.
More than 50 persons attended the lecture, including faculty members and post graduate students from the University of West Attica as well as faculty members and post graduate students from the National Technical University of Athens.
Rank-R FNN: A Tensor-Based Classifier
12/04/2021
An increasing number of emerging applications in data science and engineering are based on multidimensional and structurally rich data. The irregularities, however, of high-dimensional data often compromise the effectiveness of standard machine learning algorithms. In this study, we present the Rank-R Feedforward Neural Network (FNN), a tensor-based nonlinear learning model that overcomes the aforementioned drawback. Moreover, we establish the universal approximation and learnability properties of Rank-RFNN tensor-based model, and we validate its performance on real-world high-order datasets.
Player Modelling Lecture at IDG
12/04/2021
Today I had the opportunity to give a lecture within the frame of postgraduate course "Player Modelling" at the Institute of Digital Games, University of Malta.
That lecture was about supervised machine learning models putting emphasis on tree-based algorithms. I presented and discussed the agorithmic steps and part of the theoretical properties of tree-based classifiers and regressors and how those models can be used for modelling the behavior and/or the state of affect of players.
You can find the lecture's slides here: classification and regression
What's Next in Affect Modeling
18/03/2021
We happy to anounce our "What's Next in Affect Modeling?" workshop that will take place at 9th International Conference on
Affective Computing & Intelligent Interaction (ACII-2021). This workshop puts an emphasis on state of the art methods in machine learning and their suitability for advancing the reliability, validity, and generality of affective models. We will be investigating entirely new methods, untried in Affective Computing, but also methods that can be coupled with traditional and dominant practices in affective modeling. In particular, we encourage submissions that offer visions of particular algorithmic advancements for affect modeling and proof-of-concept case studies showcasing the potential of new sophisticated machine learning methods.
The Pixels and Sounds of Emotion
15/02/2021
The pixels and sounds of emotion! What if we could detect emotions in a general, user-agnostic fashion? Is it possible to capture human emotion solely by looking at the pixels of the screen and hearing the sounds of the interaction? No sensors, no access to biometrics, facial expressions or speech!
We are thrilled to present our new IEEE Transactions on Affective Computing paper (with Georgios Yannakakis and Antonios Liapis from the Institute of Digital Games - University of Malta) in which we transfer and introduce the idea of general-purpose deep representations for gameplaying to affective computing. We show we can predict arousal via audiovisual game footage across 4 very different games with top accuracies (as high as 85%) using the demanding leave-one-video-out validation scheme.
Game AI Lectures at IDG
09/11/2020
I was very happy to give two lectures within the frame of postgraduate course "Game AI" at the Institute of Digital Games, University of Malta.
The first lecture (02/11/2020) focused on developing agents capable of playing games using learning signals from (human) demonstrations. Emphasis was given on neural network supervised learning algorithms. The second lecture (09/11/2020) focused on reinforcement learning methodologies for training agents capable of playing games. During the second lecture, I had the opportunity to discuss and present classical reinfocement learning algorithms, as well as the latest advances in the field.
You can find the lectures' slides here: reinforcement, supervision
Exploring the next frontier of AI and ML at the Institute of Digital Games
29/05/2020
Dr Konstantinos Makantasis, post-doctoral researcher in Artificial Intelligence at the Institute of Digital Games, University of Malta, has been awarded a prestigious Widening Fellowship (Marie Sklodowska-Curie Individual Fellowships) for his TAMED (Tensor-bAsed Machine learning towards genEral moDels of affect) project. TAMED aims to create new methods and algorithms to realise aspects of general emotional intelligence (that is, the ability to understand and manage your own emotions and those of the people around you), one of the core long-term goals of artificial intelligence and artificial psychology.