Computer Vision News - February 2017

feedforward models, the importance of stripes are hard coded into the weights. In my algorithm that importance is not addressed at all during learning, it is deferred to the recognition phase. The role of the optimization during recognition is to determine which features are important. It’s figuring out, “Okay, stripes are more important and are present. That’s why I’m going to say zebra instead of horse.” Every feature is the same between the two similar candidates of horse and zebra, except the one different feature. That’s what makes that feature very important. That’s where the idea of uniqueness comes from. How does your model determine neuron activation? It does it through the optimization process. That is the key difference. In feedforward networks, the neuron activation is determined by a multiplication. In our algorithm, the neuron activation is determined by an optimization. This optimization has no other job than to determine the neuron activation. It’s not doing learning. It is finding the activation of the nodes, the neurons. Can you tell us about your organization, Optimizing Mind ? We are a small startup company. We hope to take the field by storm and be the next step in neural networks. We will allow customers to experience the same performance as feedforward networks, but also be able observe reasoning – explainability. They will also have the ability to update the network quickly, more like human brains can. This will give the customers flexibility and understanding of the network, promoting safety and trust. I think most people interested in AI have seen some adversarial examples of how neural networks can be fooled with mundane examples. It is important that this will not happen on the street, when the car is driving. A better understanding of the network will avoid it. If you could discover one thing about the brain that you don’t know, what would it be? Wow, there is so much in the brain that I want to know! I will fill you in more about my approach to understanding the brain. I focus recognition as a gateway to understanding the brain because if you can’t understand how the brain is recognizing, then how are you going to understand how the brain is doing anything else? It’s storing memory, based on what it is recognizing. It’s focusing attention, based on what it is recognizing. It’s doing perception, logic, and everything else based on what it is recognizing. So if you can’t understand how it’s recognizing, you can’t understand the brain. You can superficially record things and see that this area lit up, but for more than that, to really understand the brain, you need to understand how it’s recognizing. Once you understand how it’s recognizing, that will be the key that opens up everything else as well. That’s why I’ve focused on recognition. Computer Vision News Guest 27 Guest “ Once you understand how the brain is recognizing, that will be the key that opens up everything else as well ”