Hi. I'm Kirsten Dupart with AWS Training and Certification. Welcome to this introductory course on Amazon Machine Learning. I've been with Amazon for a year and a half, and I'm currently responsible for curriculum development within the AWS Training and Certification organization. I'll begin this course with an overview of machine learning, and we'll talk about how data plays an important role. After the overview, we'll discuss an innovative way to build smart applications and walk through a few use cases, and then we'll wrap up with a discussion of the AWS frameworks and services you can use for machine learning applications. Machine learning is a subset of artificial intelligence. It helps you use historical data to make better business decisions. Machine learning is also a process where machines take data, analyze it to generate predictions, and use those predictions to make decisions. Those predictions generate results, and those results are used to improve future predictions. Machine learning can make predictions from huge datasets. It can also optimize utility functions and extract hidden patterns and structures from those datasets by classifying data. This enables a software program to learn and make predictions in the future. Machine learning enables you to establish a cycle of improvement using the data you collect from things like clickstreams, purchases, and likes. Machine learning is used in a number of ways across a number of industries. For example, it can be used to detect fraudulent transactions, filter spam emails, flag suspicious reviews, and so on. You begin by mining large amounts of data to identify patterns among the card transactions. With these patterns, you can train the machine learning model to flag fraudulent transactions. It can also be used to personalize content for users by recommending content and predictive content loading. Machine learning can be used for targeted marketing, matching customers with offers they might like, choosing marketing campaigns, and cross-selling or upselling items. Machine learning can also be used to automate categorisation of documents such as matching hiring managers and resumes by learning to understand written content. It can be used in customer service to provide predictive routing of customer emails based on the content and the sender, as well as social media listening capabilities. Machine Learning systems discover hidden patterns in data, and use these patterns to predict patterns in the future. For example, if you're analyzing retail data and a product name contains words like jeans or jacket, then this product category likely belongs to a parallel. Machine learning systems learn from examples in the same way that children learn language or patterns. It can group data into a summary, and it can also define data in a more granular concise way. Think of machine learning as a combination of methods and systems. These methods in systems predict new data based on observed data, extract hidden structure from the data, summarize data into concise descriptions, optimize an action given a cost function and observed data, and adapt based on observed data. The field of machine learning is often classified into the following broad categories: supervised learning, unsupervised learning, and reinforcement learning. In supervised learning, the inputs to the model including the example outputs also known as labels, are known and the model learns to generalize the outputs from these examples. In unsupervised learning, the labels aren't known. The model finds patterns in structure from the data without any help. In reinforcement learning, the model learns by interacting with its environment and learns to take action to maximize the total reward. In supervised learning, the inputs to the model in the example outputs are provided, and the model learns to generalize the outputs from these examples. The human teachers experience is used to tell the model which outputs are correct and which are not. This doesn't mean that the teacher has to be physically present, only that the teachers' classifications must be present. With the help of a large training dataset, the model learns from its error and changes its weight to reduce its prediction error. In classification, the output variable is a category like color, which could be red, blue, or green, and it results in true or false for a particular question. In regression, the output variable is a number or a value like weight, dollars, or temperature. In unsupervised learning which is also called self-organization, there's no teacher. It's based solely on local information. Here, the model uses only the data presented to the network without any labels, and it detects the emerging properties of the whole dataset. The model then constructs patterns from the available information without any pre-trained data. In clustering, the model discovers groupings in the data like grouping customers based on their purchasing behavior. In association, the model discovers rules that govern large chunks of data, for example customers who buy product A, also tend to buy product B. In reinforcement learning, a software agent determines the ideal behavior within a specific context for a particular problem. The agent takes the input and decides the best action for the problem, and then based on the results of the action, the agent than receives simple reward feedback to allow it to learn from its behavior. The agent is encouraged to select an action that maximizes the reward in the long-term. This type of machine learning algorithm is inspired by behavioral psychology. Part of getting useful information out of your Machine Learning system is having a smart application. Your smart application will use machine learning to analyze your data and predict future outcomes, which are necessary to make business decisions. This can include using machine learning for business questions such as predicting customer trends like whether customers will use a particular product of yours, or to determine if an order is fraudulent. Based on the customer data you already have, you can find patterns in the data and then generate predictions to drive your product features and improvements. While machine learning is a rapidly growing field with an enormous upside for companies to use, there are some challenges to take into consideration when building your machine learning based smart application. For instance, some machine learning technology can be complex to use and implement appropriately, requiring high levels of expertise that can take time to hire or develop. Another challenge is finding the right technology that scales to the needs of customers. Finally, being able to tie machine learning to a business application can take time. In other words, refining your models so that your product app can use that model productively can require a lot of time. These are the three primary considerations you should take into account when building your machine learning application. One way to help address these challenges could be to use Amazon Machine Learning. We have offerings in Amazon Machine Learning and Spark on Amazon Elastic MapReduce or Amazon EMR for customers who want to fully manage platform for building models using their own data. For developers and data scientists who want to focus on building models, the platform services remove the undifferentiated overhead associated with deploying and managing infrastructure for training and hosting. Amazon Machine Learning support supervised machine learning approaches. These enable you to predict specific machine learning tasks such as binary classification, multiclass classification, and regression. Binary classification predicts the answer to a yes or no question. For example, is this email spam or not spam? Is this product a book or a toy, or is this review written by a customer or a robot? Multiclass classification predicts the correct category from a list. For example, is this product a movie or clothing? Is this movie a romantic, comedy, documentary, or thriller, or which category of products is most interesting to this customer? Regression predicts the value of a numeric variable. For example, what will the temperature be in Seattle tomorrow? Or for this product, how many units will sell? Lastly, how many days before the customer stops using the application? At a broad level, these are the steps involved in building a smart application using Amazon Machine Learning. To train a model, you need to create a data source object pointing to your data, explore and understand your data, and transform data and train your model. To then evaluate and optimize the model, you need to understand model quality and adjust model interpretation. After that, you can retrieve batch and real-time predictions. Let's take a quick look at a case study for using Amazon Machine Learning. Zillow is a company that provides home valuations online in the United States. When the company needed to provide more timely home valuations for customers, they decided to run their home valuation tool using Amazon Kinesis for data ingestion, and Apache Spark on Amazon EMR for data processing and analysis. Now, Zillow runs its machine learning tasks in hours instead of days, and it provides more accurate valuation data too. I hope you learned a little something and we'll continue to explore our other courses. Again, I'm Kirsten Dupart with AWS Training and Certification, thanks for watching.
