One of the most enduring drives of man is to overcome itself: to break out of self-imposed limits which are used paradoxically to define our world. Transhumanism is one of these drives. It takes our use of emerging technologies to be an unnatural, albeit necessary, extension of our collective will through which we can move beyond being merely “human”. The creation of neural networks is us sharing our fruit of knowledge with machine in order to expand beyond our limitations without losing what makes us human. Using our central nervous system as inspiration, we’ve designed computational models that can take stimuli from our chaotic environment and distill it into pure, accessible data. Their benefit to computer science is in their impact on diverse fields such as data mining, medical research and linguistics.
One of the most fascinating uses of these networks is the idea of computer vision which seems to be the opposite of computer graphics: taking images and processing them back into useable data. “The idea is that instead of having teams of researchers trying to find out how to find edges, you instead throw a ton of data at the algorithm and you let the data speak and have the software automatically learn from the data,” said Dr. Andrew Ng, the computer scientist leading Google’s research team. Google’s X laboratory was able to create a “brain” consisting of 16,000 processors that taught itself to recognize pictures of cats by flipping through 10 million 200x200 pixel images. “We trained our network to obtain 15.8% accuracy in recognizing 22,000 object categories from ImageNet, a leap of 70% relative improvement over the previous state-of-the-art.”
What’s amazing is that these photos were unlabeled and found similarities among the pictures similar to the way we are conditioned to recognize faces in random stimuli such as the man on the moon or Jesus in grilled cheese. This is important considering the fact that objects like cat faces are considered to be “high level” compared to the detection of lower level objects such as high-contrast blobs and edges. For comparison, the use of random guessing instead of the neural network only met .005% accuracy using the same data. Being able to recognize the features of a cat may seem a bit whimsical, but this same capacity to recognize abnormalities in vision is now being researched to save lives.
The medical uses of the neural networks allow for the detection of early signs of cancer with oncologists assigning values to artificial neurons. The programmer starts by conditioning the network to diagnosis cases whose results are already known as opposed to Google’s unlabeled image experiment. Since patients can be immune to, or just have a lower risk of having, various cancers based on criteria such as their sex, if they are a smoker and their age, the researches program the artificial nodes to have values reflecting such criteria. For example, the algorithm won’t have to account for finding an...