Machine Intelligence from Cortical Networks (MICrONS)
1. Incorporating feedback
In computer neural networks the signals move forward - most of the time. Yes, there is back-propagation and also "recurrent" neural networks where connections does go backward, which helps them deal with inputs that change with time. The real brain however, has a lot of feedback: there are greater number of fibers coming back than going forward. In one well-studied part of the visual cortex, says Tai Sing Lee (Professor at Carnegie Mellon), "only 5-10 percent of synapses are listening to input from the eyes. The rest are listening to feedback from higher levels in the brain".
2. Constant forecasting
The brain is constantly trying to predict its own inputs. While the sensory cortex is processing current frame of what we see, the higher levels of the brain are trying to anticipate the next frame, and passing their best guesses back down through the feedback fibers.
It is this feedback and forecast which enables the human brain to do "one-shot-learn" and makes it resilient to small changes - unlike the highly sophisticated Neural Networks like Inception-V3 and ResNet which take millions of images to train and are still very fragile to aberrations.
Emergence of Tools to provide fire power to MICrON
Beyond brain activity monitors and MRI machines which could not monitor individual neurons: the spatial resolution was about one millimeter at best. Development of techniques for making neurons light up when they fire in a living brain is a game changer. Scientists typically seed neurons with fluorescent proteins that glow in the presence of calcium ions. When a cell fires, the calcium ions become active.