The Deep Learning workshop will present modern neural network based techniques that are used in supervised learning. We will cover the basic fundamentals required to understand how neural networks work and then focus on applications of neural networks to problems in computer vision and natural language processing.
The prerequisites are familiarity with basic concepts from linear algebra, such as vectors and matrices, calculus, such as differentiation, and probability theory, such as random variables, probability distributions and expectations. The applied examples will use the popular deep learning python package Tensorflow and require familiarity with the python programming language.
A good introduction to the required background material can be found here.
1.1 Introduction to Neural Networks
- single neuron, activation
- multiple layers
1.2 Fundamentals of Deep Learning
- linear algebra
- vector calculus
- gradient descent
- supervised learning
1.3 Fully-Connected networks
- history/Motivation of deep learning
- fully-connected layers
- activation functions
- loss functions
- training neural networks as computational graphs
- Example) character recognition on MNIST data
2. Computer vision and Natural Language Processing (NLP)
- convolutional layers
- pooling layers
- image classification and segmentation with convolutional networks
- grammars/vocabularies for NLP
- recurrent layers
- enhanced recurrent networks with long-short term memory layers and gated-recurrent units
- analysis/Considerations of deep recurrent networks
3.1 Tensorflow walkthrough of image classification with convolutional networks
- image classification with convolutional networks
3.2 Tensorflow walkthrough of next character prediction with recurrent networks
- next character prediction on the Penn treebank dataset
To view other workshop descriptions, or to get general information about the ICME Summer Workshop Series, click here.
ICME Summer Workshops are open to participants 18 years and older. If you are under the age of 18 and would like to participate, please contact firstname.lastname@example.org.
Questions: please contact the ICME at email@example.com .