19.10.2015 - Seminarium Instytutowe - godz. 13:00, Bogdan M. Wilamowski (Auburn University, USA, członek zarządu IEEE, były redaktor naczelny prestiżowych czasopism IEEE)
The lecture will address, albeit usually very briefly, the following topics:
- New architectures with 10 to 100 times larger capabilities than MLP processing
- Training algorithms for these new neural network architectures
- How to solve the “diminishing gradient problem” when training deep architecture
- Training neural networks without the backpropagation computations
- Adaptive second order algorithm for on-line training
- Over 100 times faster algorithms than traditional EBP
- How to break the problem size limit for second order algorithms,
- RBF networks training resulting with 30 to 100 times smaller network size than obtained with SVM or ELM algorithms