Data management and analysis

Course info:

Semester: 5

General Foundation

ECTS: 6

Hours per week: 3

Professor: T.B.D.

Teaching style: Face to face, usage of specialized software

Grading: Homework / Projects (60%), Final written exam (40%),

Activity Workload
Lectures 36
Class assignments / projects 42
Independent study 72
Course total 150

Learning Results

Upon successful completion of the course, students will have acquired:

  • Design, train and implement deep learning models for a wide range of applications
  • Possess a critical understanding of the different characteristics, capabilities and limitations of different deep learning techniques (DNNs, CNNs and their variations, RNNs and their variations, Stacked Autoencoders, DBNs, GANs, etc.) and select the appropriate ones for different cases of complex problem solving
  • Background enabling to follow, understand and assess future theoretical and technological developments in the field of Deep Learning
  • Ability to combine and integrate theoretical and technological ideas and mechanisms from other fields to improve and extend the functionality enabled by deep learning

Skills acquired

  • Data and information retrieval, analysis and synthesis
  • Decision making
  • Individual work
  • Team work
  • New research ideas generation
  • Promotion of free, creative and inductive thinking

Recalls on Neural Networks, Multi Layer Perceptrons, Backpropagation

Loss functions, Hyperparameter tuning, Regularization, Model selection, weight decay, dropout, Optimization (SGD, Rprop, adam, rmsprop)

Deep Neural Networks

Convolutional Neural Networks (CNN), LeNet/AlexNet, Deep Residual Networks (ResNet). Application sof CNNs (Single-Image Super-Resolution, Object detection)

CNN variations and other solutions for object detection: R-CNN, Fast R-CNN, Faster R-CNN, Mask R-CNN, SSD, YOLO

Recurrent Neural Networks, Long Short-Term Memory Networks, Gated Recurrent Units, Bidirectional LSTM

Transformers, sequence-to-sequence (seq2seq) learning, attention

Generative Models. Restricted Boltzman Machines, Deep Boltzman Machines, Deep Belief Networks). Autoencoders, Stacked (Denoising AutoEncoders), Variational Autoencoders. Generative Adversarial Networks.

  1. Deep Learning, I. Goodfellow, Y. Bengio and A. Courville, MIT Press, 2016, http://www.deeplearningbook.org.
  2. Deep Learning with Python, F. Chollert, 2nd Edition, Manning
  3. Neural Networks and Deep Learning: A Textbook, C. Aggarwal, 2018, Springer

Related scientific journals:

  • IEEE Transactions on Pattern Analysis and Machine Intelligence
  • Journal of Machine Learning Research
  • IEEE Transactions on Neural Networks and Learning Systems
  • NeurIPS conference proceedings
Learning Results - Skills acquired

Learning Results

Upon successful completion of the course, students will have acquired:

  • Design, train and implement deep learning models for a wide range of applications
  • Possess a critical understanding of the different characteristics, capabilities and limitations of different deep learning techniques (DNNs, CNNs and their variations, RNNs and their variations, Stacked Autoencoders, DBNs, GANs, etc.) and select the appropriate ones for different cases of complex problem solving
  • Background enabling to follow, understand and assess future theoretical and technological developments in the field of Deep Learning
  • Ability to combine and integrate theoretical and technological ideas and mechanisms from other fields to improve and extend the functionality enabled by deep learning

Skills acquired

  • Data and information retrieval, analysis and synthesis
  • Decision making
  • Individual work
  • Team work
  • New research ideas generation
  • Promotion of free, creative and inductive thinking
Course content

Recalls on Neural Networks, Multi Layer Perceptrons, Backpropagation

Loss functions, Hyperparameter tuning, Regularization, Model selection, weight decay, dropout, Optimization (SGD, Rprop, adam, rmsprop)

Deep Neural Networks

Convolutional Neural Networks (CNN), LeNet/AlexNet, Deep Residual Networks (ResNet). Application sof CNNs (Single-Image Super-Resolution, Object detection)

CNN variations and other solutions for object detection: R-CNN, Fast R-CNN, Faster R-CNN, Mask R-CNN, SSD, YOLO

Recurrent Neural Networks, Long Short-Term Memory Networks, Gated Recurrent Units, Bidirectional LSTM

Transformers, sequence-to-sequence (seq2seq) learning, attention

Generative Models. Restricted Boltzman Machines, Deep Boltzman Machines, Deep Belief Networks). Autoencoders, Stacked (Denoising AutoEncoders), Variational Autoencoders. Generative Adversarial Networks.

Recommended bibliography
  1. Deep Learning, I. Goodfellow, Y. Bengio and A. Courville, MIT Press, 2016, http://www.deeplearningbook.org.
  2. Deep Learning with Python, F. Chollert, 2nd Edition, Manning
  3. Neural Networks and Deep Learning: A Textbook, C. Aggarwal, 2018, Springer

Related scientific journals:

  • IEEE Transactions on Pattern Analysis and Machine Intelligence
  • Journal of Machine Learning Research
  • IEEE Transactions on Neural Networks and Learning Systems
  • NeurIPS conference proceedings