11 March 2019
Graham will talk about an alternative architecture for neural networks which goes by the snappy title of RBF (radial basis function) CCN (cascade correlation networks). They are, so the theory goes, faster to “train” than a back-propagation neural network, less computationally expensive in both training and inference, not prone to adversarial changes in the input, and tell you both the classification they’ve made and a measure of the confidence of that classification. So why don’t more people use them?
Our meeting is at the Oxford Centre for Innovation - please try to arrive shortly before 7:15pm to make door management easier!
The general format is:
• 7:15-7:30 Chat, drinks & nibbles
• 7:30 Bleeding Edge News - summary of the month's AI activity
• Presentation or discussion
• 9:00ish Tidy up!
After the meeting (around 9pm) we normally head to the Lighthouse pub to continue the discussion. That's just across and down the road (towards the railway station) from the Oxford Centre for Innovation.