Interesting. So what do you use for your AGI experiments as an alternative to deep learning?
Issam Laradji

I use a combination of various types of unsupervised spiking neural nets. My nets only learn small elementary patterns that can be instantly combined to form any complex object.

The most important ingredient in intelligence is timing. It is essential to everything from perception to reasoning, planning and goal-oriented motor behavior. Most important of all, it is the key to unsupervised learning of both patterns and sequences.