AI ethics in action: the case of the EU’s ethics review procedure
Uploaded by AUTH
The talk/presentation will provide a comprehensive overview of the main ethical risks and issues raised by AI but also of the ethical principles/declarations put forward by various actors at the national and international levels. The talk will focus on the European Commission’s efforts to operationalize high-level ethical principles and requirements in the domain of research and its efforts to create a first set of specialized guidelines in the domains on fairness, algorithmic auditing and the assessment of the impact upon fundamental rights. In addition, the challenges and the opportunities associated with the gradual implementation of EU’s AI Act will be highlighted along with the particularly important work of other international organizations in the ethical governance of new and emerging technologies. The presentation will shed light on the importance of developing a specific, distinct framework for ethical assessment and value alignment in the field of artificial intelligence. Such an approach could enhance EU’s efforts to materialize the development of a human-centered and trustworthy model of AI research and development which will be aligned to its efforts to promote responsible innovation.
Designing AI tools to enhance teaching and learning: A design thinking approach
Uploaded by AUTH
Artificial Intelligence (AI) has the potential to revolutionise education by enhancing learning experiences for students and teaching processes for educators. This interactive training explores how AI tools can be designed to address key educational challenges, such as collaboration, boosting learning motivation and reducing cognitive load, through the lens of psychological theories and the design thinking process. Drawing from foundational concepts like Self-Determination Theory (for fostering intrinsic motivation) and Cognitive Load Theory (for optimising mental processing), participants will develop a deeper understanding of how these theories can inform the creation of AI-powered educational tools. Using design thinking, a human-centred problem-solving approach, attendees will work collaboratively to empathise with learners and educators, define educational challenges, ideate potential AI solutions, and develop prototypes tailored to real-world needs. Through a blend of theoretical presentations, group discussions, and hands-on activities, this session emphasises active learning and creative problem-solving. Participants will engage in exploring psychological frameworks to guide the development of AI tools, practicing design thinking to create user-focused solutions and building prototypes that address practical challenges in education. By the end of this training, participants will have generated actionable ideas for AI tools that enhance motivation, support cognitive efficiency, and foster meaningful learning experiences. They will leave equipped with a practical understanding of how AI can be designed to empower learners and teachers alike.
AI bias: overview, measurement, mitigation and application to computer vision
Uploaded by AUTH
AI bias is an emerging concern in the field of AI, due to the widespread deployment of AI-based services and applications with ubiquitous effects on our daily lives. AI bias is particularly important in high-stakes decision making scenarios such as CV ranking and hiring, credit scoring and recidivism prediction, but it is also becoming a growing concern in the context of generative AI systems, where harmful stereotypes are perpetuated and amplified through modern foundational This tutorial will introduce the emerging field of AI bias and fairness, starting from the general AI setting, and then will proceed with a more in-depth study of the problem in the context of computer vision models and applications. It will also include two hands- on interactive sessions, where participants will have the opportunity to experiment with methods for assessing and mitigating bias, first in general tabular datasets and then in computer vision datasets and models.
This session consists of four talks:
1.”Introduction to AI bias and fairness”, E. Ntoutsi (40 min)
2. “Bias in Computer Vision”, S. Papadopoulos (40 min)
3. “Overview of visual bias mitigation approaches”, C. Diou (40 min)
4. “Bias assessment and mitigation using FairBench: case study on visual data (hands-on)”, E. Krasanakis, G. Sarridis (2 hours)
https://icarus.csd.auth.gr/tropical-algebra-and-geometry-for-machine-learning-and-optimization-2/
Uploaded by AUTH
Tropical (a.k.a. max-plus) algebra has been developed in the 1980’s and has been applied successfully in nonlinear image processing, control and optimization. Tropical geometry is a relatively recent field in mathematics and computer science combining elements of algebraic geometry and polyhedral geometry. The scalar arithmetic of its analytic part pre-existed in the form of max/min-plus semiring arithmetic used in finite automata, nonlinear image processing, convex analysis, nonlinear control, and idempotent mathematics. Tropical algebra and geometry have recently emerged successfully in the analysis and extension of several classes of problems and systems in both classical machine learning and deep learning. Such areas include (1) Deep Neural Networks (DNNs) with piecewise-linear (PWL) activation functions, (2) Morphological Neural Networks, (3) Neural Network Minimization, (4) Optimization, and (5) Nonlinear regression with PWL functions. This tutorial will cover the following topics:
- Elements from Tropical Geometry and Max-Plus Algebra (Brief synopsis).
- Neural Networks with Piecewise-linear Activations, including DNNs with ReLU activations and max-out units. We will study their representation power under the lens of tropical geometry.
- Morphological Neural Networks: Recently there has been a resurgence of networks whose layers operate with max-plus arithmetic. Such networks enjoy faster training and capability of being pruned to a large degree without severe degradation of their performance.
- Neural Network Minimization. We will present several methods, based on approximation of the NN tropical polynomials and their Newton polytopes, to minimize networks trained for multiclass classification problems together with experimental evaluations on known datasets.
- Approximation Using Tropical Mappings. Tropical Mappings, defined as vectors of tropical polynomials, can express several interesting approximation problems in ML, including: (a) tropical inversion; (b) tropical regression; and (c) tropical compression. Potential applications include data compression, data visualization, recommendation systems, and reinforcement learning. We will unify these problems via tropical matrix factorization and present solution algorithms.
- Piecewise-linear (PWL) Regression. Fitting PWL functions to data is a fundamental regression problem in multidimensional signal modeling and machine learning. It has been proven analytically and computationally very useful in many fields of science and engineering. We focus on functions that admit a convex representation as the max of affine functions (e.g. lines, planes). This yields polygonal or polyhedral shape approximations. For this convex PWL regression problem we present optimal solutions and efficient algorithms.
More information and related papers can be found in http://robotics.ntua.gr.
Geometric Learning: Foundations and Applications
Uploaded by AUTH
Geometric Learning developed as a specific field of research that aims to learn from non-Euclidean domains, like graphs, manifolds, etc. In this tutorial, we first introduce the basic theory and challenges related to learning from these data, presenting basic architectural solutions for graphs, point clouds, and meshes. Then, we will present some concrete applications that are becoming more and more common in this domain.
Generative AI for Animating 3D Human Face and Body Behaviors
Uploaded by AUTH
Human behavior, both for face (e.g., expressions) and body (e.g., actions) has been studied in detail (for example for expression / action classification and prediction), but there have been few works exploring generation of novel behaviors. Generating novel sequences of human facial expression, talking heads, or body to form a natural and plausible action with continuous and smooth temporal dynamics is a challenging problem. These motions can either simulate full body movement, like for gait, or part specific movement, like in playing the guitar or phone call, or involve facial expressions, action units or even mouth movement when a person speaks or reads a text. With the advent of powerful generative models such as GANs or Diffusion models, novel data generation paradigms have become possible, and these networks have shown to be powerful in many image generation tasks. However, many issues remain to be solved especially when passing from the static to the dynamic case and new research problems emerge.
We expect generating synthetic and realistic static and dynamic data of humans can have a big impact in several different contexts. A straightforward outcome that developing such techniques could have, is that of generating an abundance and variety of new data that could be otherwise difficult, very expensive and time consuming to obtain from reality. Such data can be essential in simulation, virtual and augmented reality, in training more robust learning tools, to cite a few. For example, we could expect new applications in the game and movie industry, where fully synthetic actors could be used in the near future, without the need of explicit modeling.
In this talk, we will address some recent works in this domain for generating facial expressions, talking heads and body animation of 3D human avatars.