Short Course No. 1: Brain-Inspired Spiking Neural Network Systems for Explainable and Life-Long Learning
The Ulster University, ISRC - CN3 Autumn School, North Ireland, Oct 2022 Prof N Kasabov ISRC UU
Course organizer: Ulster University
Lecturers: Prof Nikola Kasabov
Reference: N.Kasabov, Time-Space, Spiking Neural Networks and Brain-Inspired Artificial Intelligence Springer, 2019.
1- Why brain-inspired computation?
-
How is knowledge represented in the brain?
-
The challenge is to build AI systems that can learn incrementally and possibly in a life-long learning mode and can be interpreted as knowledge discovery at any phase of their learning.
2- BI-SNN architectures. NeuCube.
-
What is a BI-SNN architecture?
-
The NeuCube Architecture
-
Spike encoding methods
-
Spiking neuron models for BI-SNN
-
Methods for unsupervised learning in SNN: Spike-Time Dependent Plasticity (STDP)
-
Methods for supervised learning in SNN: Rank order (RO) learning rule
-
Dynamic Evolving SNN (deSNN)
-
Deep learning in NeuCube
-
Incremental and transfer learning and knowledge evolution in BI-SNN
-
Life-long Learning (LLL) in the brain and in NeuCube
-
Time-Space Rule (TSR) representation in BI-SNN – “opening the cube”
-
Capturing time-space knowledge as information exchange between clusters
-
Capturing knowledge representation in a BI-SNN through supervised learning with deSNN
-
Extracting Time-Space Rules (TSR) from a trained NeuCube using EEG data for the GAL task
-
NeuCube development environment for SNN system design for TSD
-
Implementation of SNN models: From von Neumann principles and Atanassov’s ABC Machine to Neuromorphic Hardware
-
Quantum (or quantum-inspired) computation
-
Quantum-inspired optimization of eSNN
3- Application specific methods and systems
-
Deep learning and deep knowledge representation of neuroimaging spatio-temporal brain data
-
Understanding human decision-making
-
Understanding brain re-wiring due to mindfulness training using EEG
-
Brain Machine Interfaces using Brain-Inspired SNN
-
NeuCube for BCI with neurofeedback for prosthetic hands
-
BI-SNN for neurorehabilitation
-
Learning and understanding brain-VR/AR interaction in time-space
-
Deep learning and deep knowledge representation of fMRI data
-
Personalized Modelling using both fMRI and DTI data
-
Deep learning and modelling of audio and visual and multimodal audio-visual data in BI-SNN
-
Deep learning of visual information
-
BI-SNN for fast object recognition from video streaming data
-
Deep learning and knowledge representation of moving objects using DVS and retinotopic mapping in NeuCube
-
Personalized predictive modelling of individual risk of stroke
-
Using an ensemble of eSNN for Predicting Hourly Air Pollution in London Area from sensory TSD
-
Using BI-SNN as universal machines for explainable and life-long learning systems for various spatio/spectro -temporal data
4- Discussions and Future Directions
-
Advantages, Problems, and limitations of BI-SNN
-
BI-AI through BI-SNN architectures
-
Computational Neuro-Genetic Modelling (CNGM)
-
Knowledge-based human-machine interaction and symbiosis based on deep learning, knowledge representation and knowledge transfer with BI-SNN architectures
-
Understanding how humans synchronize and transfer knowledge between each other through hyper-scanning and using BI-SNN to model it.