Neuroengineering

Synthetic program:

- ARTIFICIAL NEURAL NETWORK PRINCIPLES: this part will focus upon the formulation of the artificial neural networks (ANN) as a computational tool mediated from human brain structure and functions for classification, feature extraction and function approximation tasks. The students will be provided with tools to understand the link between reinforced learning and neural plasticity with mathematical models of the formal working neuron, the emerging properties of neural populations and the training mechanisms based on error backpropagation. Feed-forward and feed-back neural architectures will be described along with supervised and unsupervised learning approaches. The students will learn how to select the specific neural network typology for the particular problem to address, making practice about the use of ANN for complex function reconstruction and pattern classification in biomedical applications.

- DEEP LEARNING TECHNIQUES: this part will capitalize on ANN methodologies to extend the modeling ability of the feed-forward networks in the representation of information by means of multiple layers of progressive encoding as the human brain does when processing and storing for example the sensorial information. Examples about the human visual system will help to understand how to deploy ANN-based computation models able to represent information in depth. Convolutional and autoencoder networks will be discussed as models for implementing deep learning strategies with specific exemplifications and real-life applications as gesture and emotion recognition from video images. The students will make practice using advanced SW tools and devoted toolboxes.

- NEURAL MEMORIES: this part will describe the main features of the human memory from both physiological and modeling points of view by focusing on episodic storage. Reinforcement learning will be revised under the paradigm of synaptic connection potentiation in the long-term memory. Recurrent networks will be described as computational tools to implement neural memories able to store and retrieve information. Procedures for pattern encoding and decoding will be presented in the paradigm of the Hebbian learning. The implementation of auto-associative memories will be described by means of binary network evolution, describing both parallel and sequential dynamics. The student will learn the criteria for optimal information storage, accuracy of pattern recall and network stability. The students will make practice about pattern modeling, encoding and memory implementation using advanced SW tools.

- COMPUTATIONAL NEUROSCIENCE: the design of bioinspired computational models of neurons and of neuronal networks how they are designed and validated against physiopathological data. Brain microcircuit simulations to achieve a better understanding of the physiology and physiopathology of the brain and simulate its information processing functionalities. The example of computational models of the cerebellum will be discussed in details.

- REHABILITATION ROBOTICS: understanding the reason why robots are increasing their importance in the rehabilitation field. Learn the requirements and specification of robot design tailored to rehabilitation applications and study their control strategies. Given some characteristics of the patients (pathology, level of impairment) and the treatment (e.g. home or clinics), the student should be able to design the main requirements of the robot to be used, select the most suitable control strategy, the sensors to be embedded and the type of actuation.

- NEUROPROSTHESES: learn what neuroprostheses are, the rationale of using them for patients affected by neurological diseases. Learn the current technologies and their limits. Given some characteristics of the patients and the treatment, the student should be able to design the main requirements of the NP to be used, select the most suitable control strategy and configuration.

- NEUROENGINEERING FOR BIOLOGY AND PHARMACOLOGY: the students learn the most used optical and electrical solutions for interfacing in vitro neuronal cultures, both aiming at reading electrical activity and stimulating it. Signal processing methods to detect neuronal activity from extracellular electrical recording. Given a neuronal mechanism to be studied, the students should be able to select the best method/technology to be used, justifying the selection.