Fundamental concepts of Deep Learning

Course objectives

To get familiarized with core concepts and methods within the area of deep learning, along with practical applications related to image and sound analysis.

Benefits

Participants of the training will learn:

• what kind of methods are typical for deep learning,
• what distinguishes deep learning methods from other typical AI approaches,
• how to implement basic deep learning models using the Caffe framework,

Discussed methods are accompanied by practical examples.

Training scope

• Introduction to Atrificial Neural Nets
  • Machine Learning,
  • The concept of an Artificial Neural Network (ANN),
  • Building blocks and architecture of an ANN,
  • Mathematical formulation of an ANN,
  • Learning algorithms for an ANN,
  • The concept of overfitting,
  • Theory of Vapnik and Chervonenkis – empirical and structural risk,
  • Implementation of an ANN.
• Introduction to Deep Networks
  • Shallow learning, feature engineering, deep learning,
  • Fundamentals of Deep Learning,
  • Convolutional Neural Networks,
  • Convolutions, filters and the pooling principle,
  • Deep Network’s architecture,
  • Feed-forward and recurrent networks,
  • Region-based convolutional networks for image analysis,
  • A Hidden Markov Model for speech analysis,
  • Long Short-Term Memory Network (LSTM),
  • Long-term Recurrent Convolutional Network (LRCN),
  • Deep Denoising Autoencoders,
  • Deep Belief Networks, Deep Boltzmann Machines,
  • Datasets for image recognition,
  • Datasets for sound and voice analysis.
• Introduction to Caffe with applications
  • Working with Caffe,
  • Implementation of a Deep Network,
  • Training a network,
  • Fine-tuning a network,
  • Using pre-trained models,
  • Parallelization,
  • Case study – image processing,
  • Case study – spund/voice analysis.

Audience

People who want to use deep learning methods or who want to acquire understanding of fundamentals of deep learning methods, with a practical focus on applications of such methods typical for the area of image recognition and sound data analysis.