This seminar will cover theory and applications of modern artificial neural networks, in particular, deep neural networks.
We will discuss basics of machine learning, mathematics behind artificial neural networks, their applications to computer vision (with convolutional neural networks), sequence data analysis (time series prediction, sentiment analysis with recurrent neural networks), give a look on bleeding edge research areas.
This seminar is called practical, because we will not only look on slides and derive formulas, but launch code and test results to give better intuition of how algorithms work.
During the course students will get understanding of how deep learning works, what areas of applications it has, how to implement, train and deploy own neural network in Python and have a plan of further development and getting a job in machine learning.
Due to limited availability of room space, computer science students interested to attend the course should send an e-mail to their representative Erik Pillon for organizational purposes
- Linear algebra
- Probability theory and statistics
- Practical skills in any programming language (is a plus)
- Numerical optimization (is a plus)
- Day 1: Introduction
-- Introduction to course
-- Applications of deep learning
-- Machine learning basics (data-driven modeling, linear and non-linear classifiers, overfitting, feature engineering)
-- General ideas about deep learning, "blocks" paradigm
- Day 2: Neural Networks
-- Artificial neuron
-- Artificial Neural Networks
-- Backpropagation algorithm
-- Practical tricks for learning neural networks (a lot!)
-- Coding session: simple computer vision example
- Day 3: Convolutional Neural Networks (CNN)
-- Human vision systen review
-- Convolutions for feature extraction
-- CNNs and why it's better for computer vision problems
-- Practical use of CNNs, knowledge transfer, 100-layer neural networks
-- Object detection, segmentation, image style transfer and other applications
-- Coding session: advanced computer vision example
- Day 4: Recurrent Neural Networks (RNN)
-- Modelling sequential data
-- Recurrent neural networks
-- Backpropagation through time
-- Coding session 1: time series prediction
-- Coding session 2: sentiment analysis
- Day 5: Modern research and career path
-- Bleeding edge results of deep learning
-- Can we solve general AI problem?
-- Job positions in machine learning
-- How to get your first job in machine learning: step-by-step guide
-- Course project discussion