 Hello everyone, this is Alice Gao. This video is the beginning of a brand new unit where I'm going to talk about probabilistic reasoning and inference using Bayesian networks. After doing quite a bit of review on probability theory, we're finally ready to talk about Bayesian networks. Let me give you a few examples of the kind of problems where we might want to use Bayesian networks as a model. You might not have seen a lot of examples of Bayesian networks in your daily life, but hopefully once you see these examples, you will realize that they are pretty natural kind of models for certain scenarios. This first example is sort of genetics, modeling the inheritance of handedness, whether someone is left-handed or right-handed. We have notes here, modeling the handedness in the parents. And given the handedness in the parents, those genes will decide on the genes in the child. But whether these genes will be expressed and how they're expressed in the individual might be different. So we have additional nodes, modeling sort of how these genes are expressed in that individual. So we have how the genes are expressed in the mother and the father, and also how the gene is expressed in the child. So we have notes for the genes, and then we have notes for the expressions. And we have edges representing the relationships. This is an example of a car diagnostic network. You will see that a lot of the Bayesian network examples are about diagnostic or diagnosis of problems in certain scenarios. So in this scenario, we may have problems with our car in a particular part of the car, and that may be caused by many different reasons. So the battery might cause the lights to have different problems, or it may cause the engine to have problems, and that may in turn cause the engine to be able to start or not. Obviously I'm not an expert on cars, but maybe some of you know better than me to understand everything in this network. This is a network on diagnosing the problems in nuclear power plant. Someone much more serious problem than when we have trouble with our car. So this network has some interesting structure. We have three tiers in this network. The top tier represents the root causes, the underlying causes of some problems. And then the middle tier represents, well, given the underlying causes, what kind of events we might have because of these underlying causes. And finally, the third layer, well, the bottom layer says that if we have certain events, then we will realize these events are happening because of the reports or outputs from some sensors. So a lot of real-world scenarios are often like this. There are a lot of root causes, the top things that we never get to really observe ourselves. And then there are events happening and often these middle events, we also don't observe them directly. Instead, we only observe the things in the bottom layer, which are outputs from sensors or output from some alarms, indicating that there might be something happening underneath. This is a much simpler scenario where we have a fire alarm situation for a particular in the building, and then it's saying that an actual fire can cause either smoke or alarm, and then it could cause people leaving the building, and that could cause people to report to authorities that something might be happening. Another possible cause for the alarm could be somebody tempering with the alarm. Our home scenario is quite similar to this, where we have an alarm and then we have two different causes for the alarm. It could be caused by burglary or it could be caused by earthquake. But we do not directly observe the alarm. Instead, we only observe calls from Dr. Watson or Mrs. Gibbon, notifying Mr. Holmes, maybe the alarm is going off and then maybe there's a burglary happening. Another prominent application of Bayesian Network is a medical diagnosis scenario. If you remember when I explained the base rule, I talked about the fact that often we know that certain diseases will cause some symptoms, but we never directly observe whether somebody has a disease. We only really observe the symptoms and we want to infer whether they have a particular disease. A lot of medical diagnosis scenarios are like that, where we have underlying causes, where we have some disease, and these diseases could again have some underlying causes. Like, it depends on your genes, it depends on your age, weight and things like that. It could cause you to have or does not have diabetes and then diabetes could cause additional symptoms, like so. And as doctors, we often only observe the results of diagnostic tests and also symptoms and we're using all of these to try to infer whether somebody has a disease. In this scenario, it's interesting that we also observe things from the top layer. We could observe things about patient information, their age, their weight, their parents' health information, so on and so forth. We sort of observe things in the top tier and in the bottom tier, and we're trying to infer things in the middle. That's everything for this video. I hope this video gave you a sense of what kind of applications there are for patient networks. Thank you for watching. I will see you in the next video. Bye for now.