Course Description
The course aims to present basics knowledge of modern approaches which are used for solving computer vision problems: from descriptions of solutions based on deep convolution networks with hacks and practical examples.
Course topics
Theory part (lectures):
- Lecture I: Introduction to Deep learning, Types of Deep Learning commonly used in CV tasks
- Lecture II : Convolutional Neural Network (CNN) for image classification, Recurrent Neural Network (RNN/LSTM) + CNN for action recognition.
Assignments:
- NN, CNN, RNN fundamentals.
- Convolutional Neural Network (CNN) for image classification dataset: cifar10 [may change]; topologies: starting from feed-forward -> LeNet like -> VGG like
- Recurrent Neural Network (RNN) + CNN for action recognition in video dataset: one from UCF family (UCF-50 or UCF-100) [may change] topologies: pre-trained CNN models (VGG-* or ResNet-*) + LSTM model + additional info
Course tools
Python, NN framework: pytorch
Prerequisites
Basic linear algebra, proficiency in Python, machine learning basics (understanding of different types of learning (supervised, unsupervised, reinforcement learning), classification, regression problems, generalization error, overfitting, train/test datasets split). Optional, but desirable: neural network (NN) basics, feed forward NN, different activation functions, backpropagation.
Lecturer
Andrii Lyubonko
Senior Engineer at Samsung R&D Institute Ukraine
I came to Computer Science with natural science background. I got my Master degree in theoretical physics from Taras Shevchenko National University of Kyiv in 2004. Then I continued my research in physics in Max Planck Institute of Complex System in Dresden. Currently, I am working as a senior research engineer in Samsung Research Institute in Kyiv.
Field of interests: Deep Learning, Computer Vision
Contacts: lyubonko@gmail.com