It’s 6:30pm on a Wednesday night, and I’m sitting in the Anvil (an abandoned church turned hacker space) about to hear a talk about “Deep Machine Learning,” by Eugenio Culurciello. I’m hoping to get insight for the Hackathon since he’s an expert. Here’s just where I’m taking notes…
What is Deep Learning?
-Deep learning consists of 3 branches: Electrical Engineering, Computer Science, Neuroscience
-An example of deep learning in today’s technology is improved image recognition from about 70% to 94% in terms of percent accuracy. Continuing on that trend machines are now able to form sentences of say a landscape that it “sees.”
-Another example is like google translate, where the machine translates live
-Deep Learning-is the de-facto standard of learning patterns and even creating images
-Deep learning is also used in enabling technology such as low-cost autonomous driving and real-time hardware tagging videos
-A recent modern example is Pokemon go that was widely enjoyed by a population, even though it didn’t necessarily recognize objects, people could play it live as they were moving and their location was shown.
-In the recent news through the google deep mind challenge, the robot was able to watch or sparse through data of many matches in Chinese checkers. Then two AIs were built to play against each other and finally after that training the robot was able to beat the human champion in the game. The same was possible for poker
-The end goal for Deep learning is for robots to be able to perceive the world and make it better
-In the same way when a human expects a ball to fall down if he or she throws it up because of experiences, robots are being “taught” experiences. An example of robots learning is AI being able to reproduce similar frames of a driving experience after seeing footage on it
-Neural Networks: Have an input, hidden, and output network respectively and they are used to make decisions with big data