Hours:
20 hours (5 credits)

Room:

From remote by using Microsoft Teams. The link will be sent in due time to all students who registered to the seminar.

To register to the course, click here

Short Abstract:
This course aims to provide an introduction to the design and use of deep learning models and reinforcement learning approaches for sensor data processing, machine vision and robotics.  The first part of the course introduces the basic concepts and fundamentals of machine learning and neural networks. The second part presents advanced deep models and their use in monitoring, understanding, control and planning tasks, with focus on robotics and distributed sensing application scenarios. Presentation of the theoretical models and associated algorithms will be complemented by references to popular software frameworks and code. Given the course focus, much of the concepts and models presented will deal with sequential data (e.g. sensor or control timeseries) and visual data (images and video), with insights on relevant problems, including lifelong learning, reinforcement learning, federated learning and learning under resource constraints.

Course Contents in brief:

1. Fundamentals of machine learning: generalization, model-selection, hyperparameters, regularization techniques, error function, maximum likelihood learning, basic concepts of probabilistic learning
2. (Deep) Neural networks introduction: artificial neuron, backpropagation, optimization techniques, multi-layer perceptron, deep autoencoders, pretraining
3. Convolutional neural networks: fundamental building blocks, advanced techniques, notable convolutional architectures, applications to image classification and semantic segmentation
4. Recurrent neural networks: sequential data processing, early recurrent models, gated recurrent architectures, advanced memory models
5. Reservoir computing: efficient recurrent neural models, memory-constrained recurrent model, sensor data processing, echo state networks, deep reservoir computing
6. Generative deep learning: variational approximation, sampling in machine learning, variational autoencoders, generative adversarial networks
7. Reinforcement learning: fundamentals, Markov decision processes, model-free and model-based algorithms, deep reinforcement learning, imitation learning
8. Advanced topics and applications: lifelong and continual learning, federated learning in cloud/distributed environments, relational learning, deep learning for robotics, embedded learning systems.

Schedule:

1. 02/02/2021: 9:00-13:00
2. 04/02/2021: 9:00-13:00
3. 09/02/2021: 9:00-13:00
4. 11/02/2021: 9:00-13:00
5. 15/02/2021: 9:00-13:00