Hyperdimensional Computing and Vector Symbolic Architectures for Artificial Intelligence
PhD course, PhD programme in Information and Communication Technologies (ICT), Sapienza University of Rome. 12 hours, 3 credits.
PhD course, PhD programme in Information and Communication Technologies (ICT), Sapienza University of Rome. 12 hours, 3 credits.
The course presents a computing framework that uses high-dimensional vectors to represent and manipulate data structures in machine learning. Known as hyperdimensional computing (HDC) or vector symbolic architectures (VSA), it originated at the intersection of symbolic and connectionist approaches to artificial intelligence and has become a research area in its own right, with applications in perception, analogical reasoning, memory models and language processing. Its connections to neuroscience, computer science, electrical engineering, mathematics and cognitive science also make it useful for studying and interpreting neural network models. The course gives PhD students the tools to understand and manipulate high-dimensional vectors and to use them in problems related to neural networks and artificial intelligence.
Basic linear or abstract algebra, probability theory and elementary logic. Programming experience is welcome but not required.
Introduction to computing with high-dimensional vectors: traditional computing; hyperdimensional computing; vector symbolic architectures.
HDC/VSA models: (Fourier) holographic reduced representations; binary spatter codes; multiply–add–permute; sparse binary distributed representations; other models.
Representation and manipulation of data structures: encoding and decoding; manipulating data structures.
HDC/VSA and artificial intelligence: HD vectors as input and output of (randomized) neural networks; HDC/VSA in AI model design; explainable AI with HDC/VSA.
Applications and implementation: cognitive applications; software and hardware implementation; the TorchHD library; research directions.
Discussion of a scientific paper related to the course, or a practical project connected with the student's research.
Slides, lecture notes and code notebooks are on Google Classroom; the access code is given at the first lecture. The schedule for each academic year is announced on Classroom.