Robustness of deep learning models for neural decoding through time
keywords ARTIFICIAL INTELLIGENCE, DEEP LEARNING, NEUROMORPHIC COMPUTING, ROBOTICS, TRANSFER LEARNING
Reference persons STEFANO DI CARLO, ALESSANDRO SAVINO
External reference persons Alessio Carpegna, Paolo Viviani, Alberto Scionti
Research Groups DAUIN - GR-24 - SMILIES - reSilient coMputer archItectures and LIfE Sci
Thesis type RESEARCH / EXPERIMENTAL
Description The work fits in the context of the B-Cratos H2020-FET project (https://www.b-cratos.eu), where a machine learning-based methodology is being developed to translate electric signals recorded by brain-implanted electrodes, into meaningful commands for a robotic hand, while tactile feedback from an electronic skin is sent back to the brain to provide sensory stimulation.
The focus of this thesis falls on the development of a transfer learning methodology to enable the re-use of trained deep learning models for subsequent experimental session during multiple days, being able to tolerate the physiological evolution of the neural signal without the need of significant re-training effort. This would improve the life of patients by reducing the training effort needed each day to calibrate their prothesis.
The duration of the thesis work is expected to be around 9 months, adjustable based on the specific needs and skills.
Required skills MS students in for Computer Engineering; Experience with at least one between Python and C/C++
Valuable skills: knowledge of Deep Learning algorithms and data science frameworks (Keras, Pytorch, Pandas)
Notes Send CV to email@example.com, firstname.lastname@example.org or email@example.com specifying the thesis code and title.
Deadline 26/01/2024 PROPONI LA TUA CANDIDATURA