You are in taxonomy page
Self-awareness for autonomous systems
Self-awareness is a broad concept borrowed from cognitive science and psychology that describes the property of a system, which has knowledge of “itself,” based on its own senses and internal models. This knowledge may take different forms, is based on perceptions of both internal and external phenomena, and is essential for being able to anticipate and adapt to unknown situations. Computational self-awareness methods comprise a new promising field that enables an autonomous agent to detect non-stationary conditions, to learn internal models of its environment, and to autonomously adapt its behaviour and structure to the contextual tasks. In this talk I will introduce the concept of computational self-awareness, explain its key capabilities and discuss the current state of research and open challenges.
Symbolic, Statistical, and Causal Representations
In machine learning, we use data to automatically find dependencies in the world, with the goal of predicting future observations. Most machine learning methods build on statistics, but one can also try to go beyond this, assaying causal structures underlying statistical dependencies. It turns out that causality can play a central role in addressing some of the hard open problems of machine learning, due to the fact that causal models are more robust to changes that occur in real world datasets. The talk will argue that causality has some shortcomings that are complementary to those of current machine learning, and the study of causal representation learning may help unify the advantages. It will also introduce some algorithms and applications in this field.
Digital Pathology: On the intersection of Computer Vision and Data Science
Due to the proliferation of whole-slide-imaging (WSI) digital scanners it is now possible to leverage computer vision, image analysis, and machine learning techniques, such as deep learning to process the digital pathology images in hopes to derive, diagnosis and prognosis markers. The convergence of digital imaging, data science and pathology gave rise to a new research area known as digital pathology. Digital pathology greatly enhances diagnostic accuracy and allows a variety of pathology tasks to be completed with greater efficiency. This presentation will offer a general introduction to the topic, it will outline and discuss imaging tasks needed for the successful implementation of a digital pathology pipeline, and it will offer overview and insights on how data-driven solutions such as deep neural networks can be used to derive markers from digital pathology slides. It will also be shown that a diagnostic system that combines deep learning and prior histological knowledge can provide useful diagnostic/prognostic markers. Lastly, open research issues and implementation challenges will be briefly discussed.
Hybrid AI for knowledge representation and model-based medical image understanding
Image understanding benefits from the modeling of knowledge about both the scene observed and the objects it contains as well as their relationships. We show in this context the contribution of hybrid artificial intelligence, combining different types of formalisms and methods, and combining knowledge with data. Knowledge representation may rely on symbolic and qualitative approaches, as well as semi-qualitative ones to account for their imprecision or vagueness. Structural information can be modeled in several formalisms, such as graphs, ontologies, logical knowledge bases, or neural networks, on which reasoning will be based. The problem of image understanding is then expressed as a problem of spatial reasoning. These approaches will be illustrated with examples in medical imaging, illustrating the usefulness of combining several approaches.
Geometric deep learning
Unlock the world of Geometric Learning and Graph Convolutional Networks (GCNs) in this comprehensive course designed to empower you with cutting-edge knowledge and practical skills. Geometric learning is a fascinating field at the intersection of computer vision, machine learning, and data analysis, with applications ranging from image processing to 3D modeling and beyond.
Domain Adaptation and Generalization
There is an issue of domain shift in machine learning models, which occurs when models trained on one dataset perform poorly when tested on data from a different source. This problem is particularly relevant for visual models operating in the real world or when no labelled data is available for the target scenario. Domain adaptation (DA) algorithms aim to bridge the gap between these different input distributions. They work by using labelled data from a source domain to create a model for a target domain with limited or no labelled data. The text explains that various DA techniques have been developed to address this problem, with a focus on recent trends in deep neural networks. Additionally, it mentions domain generalisation, which is an even more challenging task because it assumes that target data is unavailable, requiring models to generalise effectively without specific adaptation to classify previously unseen samples.
