Welcome back. How did you do? Hopefully you were able to find the answers. Just in case though, let's go over some of the key points. The first video you watched is one of my favorites. Katarina and Frederick identified an important problem. Bees along with other insects around the world are dying and we don't know why. They wanted to use machine learning to uncover probable causes. One other objectives was to determine whether the hive was inhabitable. They built hive monitors that used cameras to track bee population. This is referring to the vision capability of ML. If the number of bees returning to the hive is lower than the number that left, they'd know that the hive ecosystem is the issue. Why is this beneficial? They collected data that didn't exist before or was impossible to collect. Possible reasons for why bees were dying and supporting data can now be shared with various groups who can then identify the necessary solutions, from beekeepers to landscape and facilities maintenance to government and policymakers. Let's review the answers for the second video. The second video covers a specific use case for Google Assistant to enable those with disabilities, in this case, Parkinson's disease, to gain independence with the use of smart devices like Google Home and Nest. The ML capability called out was voice recognition for controlling multiple devices. Doctor Fred Wilson is able to use Google Assistant, which is connected to Nest and his cell phone to adjust the temperature or to make a phone call. The benefits here are easy to spot. For everyday use, this is convenient, creates efficiency and helps to improve human lives. That's not all. Did you notice that even though these examples don't directly tie to a business value, they were both designed to improve life, bees and humans? This is an important factor about using ML and AI responsibly which I'll cover in more detail later. Next, you may have noticed that there was more than one ML capability used in a solution. This is possible. Finally, what I hope you took away from these videos is that you don't have to be an ML practitioner to use ML. What you need is to combine your domain expertise with what you know about machine learning to start using it. You might be wondering if ML is so simple and has been around for so long, why has it only become popular recently? Much of the excitement around ML now is because the barriers to entry have fallen. You no longer need to be an astrophysicist to do machine learning. I'll give you a few key points that have eliminated the barriers to ML. Computers have become more powerful and more accessible. With more powerful computers, experts are now able to build more mature and sophisticated ML algorithms. Similarly, the availability of powerful computers at lower cost has increased the availability of data. Think about any digital device that generates data such as your smart phone or your smart watch. What about the Cloud? All of these are also part of the benefits of the cloud. You no longer need to invest heavily to setup and maintain your own IT infrastructure to use ML. Google design new types of hardware specifically for ML such as TPUs. The Tensor Processing Unit or TPU is specifically optimized for ML and it has more memory and a faster processor. Google has been working on the TPU for several years and has now made it available to other businesses like yours for really big and challenging ML problems. Open source machine learning platforms like TensorFlow are one of the resources that give businesses access to the means to build, train, and serve ML models. Cloud technology makes it easier to adopt ML for businesses at many levels of ML expertise. For example, with predefined ML models related to sight, language, conversation for processing structured data. Earlier, I mentioned that when you have more examples generally, your ML model is better. Take a look at this graph. It shows that our loss or error decreases as we have more data. But take an even closer look. Check out the X axis. The ML error rates decrease linearly as the amount of data doubles. That's an exponential increase in the amount of data needed. That's it for understanding the fundamentals of how ML works. Remember, there are many types of ML problems. We're focusing on supervised learning. We talked about the four key components in the ML definition, standard algorithm, data, predictive insights, and repeat decisions. We looked at several real life examples and closed with why ML is gaining popularity now. Check out the next module to learn how to build and evaluate ML models.
