So you would like to try to use generative deep learning models for a small side project? Or maybe you only want to have a better understanding of the latent spaces inside a deep neural net, gain an insight on random sampling inside an NN or learn about the mysterious reparametrization trick?
In any case, join us to explore the family of generative models based on the autoencoder architecture. In this talk you will get an intro to Autoencoders (AE), Variational Autoencoders (CAE), Categorical VAE (CVAE), unsupervised VAE and semi supervised VAE.
This presentation will give a brief introduction to contextflow, a leading AI in radiology company based in Vienna, before moving on to discussing self-supervised learning for computer vision. The concept behind self-supervised learning will be explained and it will be taken a look at the setup of current state-of-the-art methods, e.g. SimCLR, MoCo, and SwAV. Finally, possible future directions will be illustrated, followed by a QnA on discord.
Pie & AI is a series of deeplearning.ai meetups independently hosted by community groups. This event is hosted by Deep Learning Graz.
Please complete your registration on the signup form here.
After the event we will provide a limited course promo code for attendees who sign up through the form and complete a post-event survey sent by deeplearning.ai after the event. The code is for 50% off first-month subscription to any of deeplearning.ai’s courses on Coursera.
Abstract: Contrastive loss terms allows someone to learn good representations in DeepLearning without labels based on the data alone. This can be used in a later stage f.e. to train classifiers with a greatly (about 100 times) reduced need in labels. F.e. the xent loss learns by data augmentation and negative sampling good representations in an unsupervised manner. We show how to use this loss and train a network, present choices for data augmentation, discuss where it’s useful or not, and present some use cases the presenters worked on with the term.
Many real-life problems, including Automated Driving, can be formulated as sequential decisionmaking and control tasks. Solving these tasks is a major topic of Control Theory, Artificial Intelligence and Robotics, with complementary methods roughly grouped as Control, Planning and Learning. Planning relies on deliberative reasoning about the current state and sequence of future reachable states to solve the problem. Learning, on the other hand, is focused on improving system performance based on experience or available data. Combining these methods by learning to improve the performance of planning, based on experience in similar, previously solved problems, is ongoing research.
This talk provides a concise introduction to basic Planning and Learning methods, specifics of Automated Driving problems and a state-of-the-art combined Planning and Learning approach.
About the speaker:
Zlatan Ajanovic is Senior researcher at VIRTUAL VEHICLE Research GmbH, Graz (Austria). He started his career in automotive in 2011 and held several positions since then, in Prevent Group (Bosnia and Herzegovina) and AVL List (Austria). He received Bachelor and Master degree from the Univesity of Sarajevo and a Ph.D. degree from the Graz University of Technology, all focused on Automation and Control. Through ITEAM project, as a Marie Curie Fellow, he was a visiting researcher at TU Delft, University of Sarajevo, AVL List and Volvo Cars.
Currently, he serves as a member of the IFAC Technical Committee for Intelligent Autonomous Vehicles. His current research interests include Planning, Learning and Control methods applied to Autonomous Vehicles. He publishes regularly on major Robotics, Artificial Intelligence and Control events, and he is the recipient of the IFAC Young Author Award and Hans List Scholarship.
In this talk, Halil Beglerovic will talk about the applications of convolutional neural networks in the domains of – Scenario extraction from recorded data – Object detection – Semantic segmentation and Behavioral cloning.
About Halil Beglerovic:
Halil Beglerovic finished his bachelor studies at the University of Sarajevo, Faculty of Electrical Engineering, Department for Automatic Control and Electronics. He received the Master’s degree in Robotics from Tohoku University in Sendai, Japan. During the master studies his research focus was on passive mobile robots with an emphasis on formation control. Afterwards he worked in the Royal Military Academy in Brussels, Belgium on the ICARUS FP7 project focusing on unmanned ground vehicles used in Search and Rescue operations, map building and fusion of 3D environment data. Currently he is a Maria Curie fellow at AVL, Austria and PhD candidate at the Technical University Graz through the iTEAM project. His work focuses on test methods for verification and validation of autonomous vehicles with a focus on scenario detection using deep learning.
So you want to control a car or a robot using deep learning?
Gregor Gregorcic shows in this introductory session what the foundations of neural networks are and provides codesamples. At the end of this session, you will have a clear understanding about the inner workings of a neural network.
This is a preview to our upcoming course in AI which will start in autumn 2019. The lecture will be held in english.
About Gregor Gregorcic: Gregor holds a Ph.D. in Control engineering from the University College Cork, Ireland, which he did in 2004.
2018 stand die künstlerische Tätigkeit von mur.at ganz im Zeichen von Machine Learning. Die Ergebnisse daraus wurden in der Akademie Graz präsentiert. Das Team rund um Reni Hofmüller und Jogi Hofmüller gibt beim Meetup einen Einblick in die Arbeit und die Ergebnisse von sensu lato.