Artificial Neural Networks and Machine Learning – ICANN 2018

Everyday subjective experience of the stream of consciousness suggests continuous cognitive processing in time and smooth underlying brain dynamics. Brain monitoring techniques with markedly improved spatiotemporal resolution, however, show that relatively smooth periods in brain dynamics are frequently interrupted by sudden changes and intermittent discontinuities, evidencing singularities. There are frequent transitions between periods of large-scale synchronization and intermittent desynchronization at alpha-theta rates. These observations support the hypothesis about the cinematic model of cognitive processing, according to which higher cognition can be viewed as multiple movies superimposed in time and space. The metastable spatial patterns of field potentials manifest the frames, and the rapid transitions provide the shutter from each pattern to the next. Recent experimental evidence indicates that the observed discontinuities are not merely important aspects of cognition; they are key attributes of intelligent behavior representing the cognitive “Aha” moment of sudden insight and deep understanding in humans and animals. The discontinuities can be characterized as phase transitions in graphs and networks. We introduce computational models to implement these insights in a new generation of devices with robust artificial intelligence, including oscillatory neuromorphic memories, and self-developing autonomous robots. On the Deep Learning Revolution in Computer Vision