Artificial Intelligence SOIREE

Artificial Intelligence SOIREE

https://video.nti.sk/join/index.php?absf1q5o0e9

9.00 am Slovak Time
15 : 00 – China Time
16:00 - Japanese & Korean Time
00:00 am LA , USA Time zone

20.5.2022 09.00 – 13.00 Slovak time zone

HYBRID MODE - join us live in SCIENTIFIC CAFETERIA - COMPUTING CENTER TU KOSICE , Slovakia or from WEBEX LINK

=============================================================

Visiting Speaker: Prof. Daniel W. Carruth – Mississippi State University, USA
https://www.cavs.msstate.edu/directory/information.php?d=21

Title : Automated Self-Driving Vehicles in Off-Road Environments

9.00-9.30 am

Research Interest :

Dr. Carruth's research interests include modeling and simulation of autonomous ground vehicles in military off-road environments, human interaction with autonomous vehicles, use of UGVs by military and law enforcement, as well as physical and cognitive aspects of human task performance in law enforcement, military, and industrial work. He has active projects generating virtual environments and test standards for military off-road autonomous vehicles, building off-road autonomous vehicle platforms for real-world testing, and developing virtual greenhouses for training undergraduate students in electronic control systems for precision agriculture.

Abstract : Significant advances have been made in automated driving (AD) systems in recent decades. However, much of the progress has focused on vehicles operating with high levels of infrastructure, accurate and detailed map data, and on smooth, firm driving surfaces. Many consumer, industry, and military users are interested in operating autonomous ground vehicles in off-road environments at speeds equivalent to human. In this seminar, I will discuss current research activities using both simulation and field testing to address current obstacles for effective operation of autonomous ground vehicles in off-road environments.

Q&A

Pause until 10.00

10.OO-10.30am

Speaker: Lukaš Hruska, MSc. – PhD Student

Title:
Graph-based deictic deep reinforcement learning with a priori information

Abstract:
In this seminar, we will discuss the need to improve the learning speed of neural networks in reinforcement learning. We will gloss over several commonly used approaches and focus on methods aimed at pre-processing the input, especially the so-called deictic representation. Over the course of the meetup, we will introduce graph neural networks and explore their possible use for deictic representation, which promises to result in better generalization and allows for incorporation of expert knowledge.
Q&A

Pause until 11.00

11:00 am

Speaker : Andrinandrasana David Rasamoelina MSc. – PhD student

Title : Title: Incredibly Simple Few-Shot Learner

Abstract: In this seminar, we will briefly talk about Few-Shot Learner. We will look at ways to solve those problem using simple techniques such as data augmentation, normalization and tweaking the optimization process. During the meetup, we will also explore the usage of Few-Shot Learner in real world.

Q&A

Pause until 12.00

12:00 am

Speaker : Dominink Vranay MSc. – PhD Student

Title : Capsule Neural Networks using Segmentation masks

Abstract : In this seminar, we will discuss Capsule Neural Networks and how they could be improved. We will go through a short introduction of Capsules as well as the main advantages and disadvantages. I will also showcase some preliminary results of my experiments using Capsules and some ways, I found, in which the accuracy and usage can be improved. I will talk about the main improvement of using segmentation masks for reconstruction as well as using different reconstruction losses.

Q&A

Closing 13:00 - Event will be recorded