So, machine learning is part art and part computer science. Failure is frequent, requires a great deal of human intervention, and as we'll see in data analytics, preparing the data is currently very labor intensive and lot of human intervention involved. You got to ask the right questions, you got to choose the right data, you got to choose the right algorithms, you got to tune the algorithms and those hyper parameters, you have to test your results. I wanted to just spend just a moment talking about. So we've been talking about machine learning, this is IoT class, so what could machine-learning do in the five application areas that we looked at earlier in the semester? These are my opinions, I just steps back and thought about each of these application areas and came up with some ideas. So automotive and transportation, autonomous navigation and safety. Then if I wrote and weather conditions perhaps. Everyone see the article recently with a Uber auto car killed pedestrian crossing the road? It's sad, sad for the family and the people involved. This will evolve over time and get better, and I think we talked about it at the beginning of the semester, auto autonomous driving vehicles, human beings walking and animals just like deer crossing the road and raccoons, and all those unexpected stuff right? The algorithm has been trained to track the side of the road, track the vehicle in front of it and beside it and behind it, okay. Yet, when unanticipated things happen, it's very hard for the algorithms where they are today, to respond in a way a human driver would, but I think we'll get there. Industrial, we spend some time talking about that. So, predicting equipment failures, identifying opportunities for improved operational efficiency. You own a factory or a business, you want to keep all of your machines up and running as close to 100% of the time, and be moving raw material and producing finished goods as efficiently as possible. In building automation, that might be bioelectric based, physical access might be visual recognition, retina, fingerprints, who knows? So SparkFun had a fingerprint reader, that might be fun to get one of those and play around with that, I don't know what I'll do with it, but that would be cool. Energy use optimization would be the big thing there, they want to make a smart building and it's about pulling down the cost of operating that office, electricity heating, cooling. Oil and gas could be deployed in predicting where hydrocarbon locations are. They do, I can't remember the name of it, they put probes in the ground and then they'll thump the ground [inaudible] , create a small detonation, and it creates waves that get reflected back up, and can give them ideas about where hydrocarbons are located. And it's possible that a machine-learning algorithm could, instead of human beings looking at that data, could be given to a machine learning algorithm to say "We think there's hydrocarbons gas or oil or something is located over here or over there" based on the sensor data. And also predicting equipment failures, again, it's similar to the industrial space, they want to keep their equipment up and running as much as possible. Maybe other tests that they do to determine where hydrocarbons are, deep underground, besides just the vibration based one. But again, it could be an opportunity there for machine learning. In agriculture, we saw the guy with the avocado farm. He's lowered his water use, these systems could be used to predict that the consumables, seed and herbicides and fertilizer and so forth. Applying the algorithm. So, this is part of the embedded system course. So how can I apply these algorithms in my embedded system? Help. Way I see it, you have two basic approaches. You train offline in the lab and then deploy the trained algorithm in the field, much like the hearing aid example. I foresee a day when the training would happen in the device, would be happening autonomously or maybe an ongoing, might be lieutenant commander data on Star Trek continuously training his positronic network. Certainly, could happen.