Our next technical evening of 2019, was held in the National Library of Technology, looked at research in the area of usage of Artificial Intelligence in the development of autonomous driving, its problems, opportunities and challenges.
Development of autonomous cars is not only a futuristic concept, but it is here already now. Big international companies, as well as universities are joining the race to deliver the technologies. AI is an omnipresent technology associated with almost every area of our lives. Autonomous cars are not an exception.
You listened to the stories of Patrick Perez and Matthieu Cord, representatives of Valeo.ai, a special research group focused on the use of AI technology in automotive and Tomáš Svoboda, chair at the Department of Cybernetics from Czech Technical University.
Date: Wednesday 17. 4. 2019
Národní technická knihovna
160 80, Prague
Entrance number NTK3
[05:00 PM] Doors open
[05:30 PM] Presentations & open discussion
[08:00 PM] Networking & Drinks
Topics & Speakers:
Unsupervised domain adaptation with application to urban scene analysis
Patrick Pérez Director Valeo.ai
Abstract: In numerous real world applications, no matter how much energy is devoted to build real and/or synthetic training datasets, there remains a large distribution gap between these data and those met at run-time. This gap results in severe, possibly catastrophic, performance loss. This problem is especially acute for automated and autonomous driving systems, where generalizing well to diverse testing environments remains a major challenge. One promising tool to mitigate this issue it unsupervised domain adaptation (UDA), which assumes that un-annotated data from the “test domain” are available at training time, along with the annotated data from the “source domain”. We will discuss different ways to approach UDA, with application to semantic segmentation and object detection in urban scenes. We will introduce a new approach, called AdvEnt, that relies on combining adversarial training with minimization of decision entropy (seen as a proxy for uncertainty).
Designing multimodal deep architectures for Visual Question Answering
Matthieu Cord Professor of Computer Science, Sorbonne University and Researcher at Valeo.ai
Multimodal representation learning for text and image has been extensively studied in recent years. Currently, one of the most popular tasks in this field is Visual Question Answering (VQA).
I will introduce this complex multimodal task, which aims at answering a question about an image. To solve this problem, visual and textual deep nets models are required and, high level interactions between these two modalities have to be carefully designed into the model in order to provide the right answer. This projection from the unimodal spaces to a multimodal one is supposed to extract and model the relevant correlations between the two spaces. Besides, the model must have the ability to understand the full scene, focus its attention on the relevant visual regions and discard the useless information regarding the question.
Learning for traversing rough terrain
Tomáš Svoboda Associate Professor at Czech Technical University, chair of the Department of Cybernetics
Autonomous traversal of difficult obstacles and rough terrain in general is one of the essential functionalities required for deploying search and rescue robot. The talk will discuss several learning algorithms for controling robot morphology – configuration of articulated parts and their compliance. Algorithms use both interoceptive and exteroceptive data. Our story begins with a classical Reinforcement learning approach and ends with domain transfer by employing generative adversarial models for bridging the gap between simulated and real world.