Introduction to Machine Learning: Art of the Impossible

Question 1

What is machine learning?

Answer 1

Machine learning (ML) is the process of training computers, using math and statistical processes, to find and recognize patterns in data

Question 2

What are the steps to create ML applications?

Answer 2

1 - Formulate a problem
2 - Prepare your data
3 - Train the model
4 - Test the model
5 - Deploy your model

Question 3

What are the key terms in machine learning?

Answer 3

Model: The output of an ML algorithm trained on a data set; used for data prediction

Training: The act of creating a model from past data

Testing: Measuring the performance of a model on test data

Deployment: Integrating a model into a production pipeline

Question 4

What is the history of Amazon machine learning?

Answer 4

Amazon has over 20 years of experience with machine learning. In addition, Amazon uses ML to sell more than 4,000 products per minute on Amazon.com, and also in completing the first autonomous Prime Air Delivery in 2016. In July 2020, AWS was recognized as a leader in the Gartner Magic Quadrant for cloud AI developer services. The ML platform Amazon SageMaker received the highest rating among its peer group (84/100) on Gartner’s “Solution Scorecard for Amazon SageMaker.”

Question 5

What is the Amazon approach to machine learning?

Answer 5

The Amazon flywheel was an idea that Amazon founder Jeff Bezos sketched on the back of a napkin. It illustrates how investing in specific key business operations can reinforce other processes and create a positive feedback loop. When Amazon focused on improving the customer experience, more customers joined. Higher customer traffic led to larger vendor pools and broader product selections, which resulted in lower prices and lower cost structures. Amazon reinvested in improving the customer experience, which leads to platform growth, and the flywheel reinforcement continues.

The Amazon ML flywheel uses data collected from parts of a business operation, uses a model to predict future outcomes, and provides ways to continuously improve efficiency and develop new operational capabilities and business practices. With ML, increasing predictions improve growth and efficiency. This leads to more usage and data, completing the feedback loop and reinforcing all parts of the flywheel.

Question 6

How is Amazon using machine learning in products?

Answer 6

1. Amazon uses browsing and purchasing data to provide tailored product recommendations and promotions.
2. Amazon uses ML to facilitate billions of voice interactions per week with Alexa devices using natural language processing (NLP).
3. Amazon uses ML to ship 1.6M packages per day.

Question 7

How is machine learning helping AWS customers?

Answer 7

AWS machine learning services have provided solutions for a variety of customer use cases. AWS ML customers have extracted and analyzed client document data to help speed up critical business decisions and the identification of fraudulent online activities. AWS ML customers forecast their key demand metrics to meet customer demand and reduce waste. These customers have also generated personalized recommendations to maximize customer engagement. Below is a quick overview of various AWS AI, ML, and platform services that customers are using to accelerate business outcomes.

Question 8

What are some examples of machine learning being used today?

Answer 8

Machine learning can be found in about 77 percent of the devices we use (“What Consumers Really Think About AI: A Global Study,” Pegasystems, 2017). For example, ride-sharing apps like Uber and Lyft use data to lower wait times, predict demand, and optimize price setting. Online shopping sites use ML to customize search results and improve product recommendations. Financial institutions use AI to recognize content on mobile check deposits. Credit- or debit-card transaction businesses use ML to scan for fraud. Research suggests that more than 8 billion digital voice assistants will be powered by AI and ML over the next few years (“Digital Voice Assistants in Use to Triple to 8 Billion by 2023, Driven by Smart Home Devices,” Juniper, 2018).

Question 9

What other industries are using machine learning?

Answer 9

Industrial companies use AI and ML services for asset management. This includes using computer vision for equipment monitoring and defect detection, or analyzing operational machine behavior data to enable predictive maintenance. Customer service organizations use ML to transcribe and analyze live and archived calls for sentiment scores. ML can also help prioritize based on categorized customer feedback, and enable software to provide agents with answers to questions as they are being asked.

Question 10

How does the NFL use AWS machine learning?

Answer 10

AWS helps the NFL to leverage the power of its data through sophisticated analytics and ML. The NFL uses training data from traditional box-score statistics. The NFL also uses data collected from the stadium to create new stats, improve player health and safety, and provide a better experience for fans, players, and teams—all in real time. ML models built on the Next Gen Stats (NGS) platform ingest the data. This continually trains and refines the models to help boost accuracy, speed, and insights while reducing the time to get results.

Question 11

How can machine learning help me?

Answer 11

Machine learning can continuously improve results, which means training models can become a part of almost any decision-making process. Machine learning can ingest limitless amounts of data, produce timely analysis and assessment, identify trends and patterns, and generate predictive forecasts.

Question 12

What is artificial intelligence?

Answer 12

Artificial intelligence (AI) is any system that is able to ingest human-level knowledge to automate and accelerate tasks performable by humans through natural intelligence. AI has two categories: narrow, where an AI imitates human intelligence in a single context, and general, where an AI learns and behaves with intelligence across multiple contexts.

Question 13

What is the difference between ML and AI?

Answer 13

Artificial intelligence ingests data, such as human-level knowledge, and imitates natural intelligence. Machine learning is a subset of AI, where data and algorithms continuously improve the training model to help achieve higher-quality output predictions. Deep learning is a subset of machine learning. It is an approach to realizing ML that relies on a layered architecture, mimicking the human brain to identify data patterns and train the model.

Question 14

What are the requirements to implement AI?

Answer 14

The core components of AI are domain knowledge to structure and frame the problem correctly, high-quality input data to train the model, and methods to detect patterns and make predictions.

Question 15

What is the difference between machine learning and classical programming?

Answer 15

Machine learning involves teaching a computer to recognize patterns by example, rather than programming it with specific rules. These patterns can be found in the data. In other words, ML is about creating algorithms (or a set of rules) that learn from complex functions (patterns) from data and make predictions on it (a form of “narrow AI”). ML learns from data and can be reused for unseen, future, or new data without rewriting code. Put another way, with ML, you start with a problem, identify data associated with that problem, use an algorithm to then model that problem, and generate output.

Question 16

How does machine learning work?

Answer 16

Supervised learning is a method where a model learns from a data set containing input values and paired output values that you would like to predict.

Examples:
Classification - Classifying documents / Regression - Forecasting demand for a product

Unsupervised learning is a method in which the training model learns from data without any guidance. The objective is pattern and structure recognition.

Examples:
Clustering - Customer segmentation / Association - Discovering regularities among products

Reinforcement learning is a method in which the training model learns from its environment by being rewarded for correct moves and punished for incorrect moves.

Example: Autonomous driving

Question 17

What kind of solutions can ML provide?

Answer 17

ML can provide predictive solutions (regression and classification), prioritization (rankings and scores), and behavior patterns (recommendations and clustering).

Question 18

What are some potential problems with machine learning models?

Answer 18

Machine learning modeling can be problematic for learning algorithms due to the ingestion of poor quality data. For example, the data may not include enough samples to represent a sufficiently broad scope of relevant variables.

Question 19

What are simple and complex models?

Answer 19

Simple and complex ML models differ when balancing a model’s accuracy (number of correctly predicted data points) and a model’s explainability (how much of the ML system can be explained in “human terms”). The output of a simple ML model may be explainable and produce faster results, but the results may be inaccurate. The output of a complex ML model may be accurate, but the results may be difficult to communicate.

Question 20

What is unexplainabilty?

Answer 20

Unexplainability represents how much of the reasoning behind an ML model’s decision cannot be effectively described in human terms. There are potentially legal, professional, ethical, and regulatory conditions where the tolerance for unexplainability may vary from case to case.

Question 21

When is it (unexplainabilty) not okay?

Answer 21

• When you need to be able to explain to your customer why a loan was declined
• When you need to be able to explain why a transaction was deemed fraudulent

Question 22

When is it (unexplainability) okay?

Answer 22

• When risks of misclassification are low, such as object recognition for catalog search or predicting the probability of completing NFL play
• When humans make the final decisions

Question 23

What is uncertainty?

Answer 23

Uncertainty describes an imperfect outcome. In the context of machine learning, uncertainty arises from using models. These models attempt to fit a training data, which may have imperfect data. The “best” data may also be unknowable.