About this video
Machine learning is only as good as its training data. The video shows where training data comes from, such as streaming history, website image tests and medical scans, and explains how gaps in the data create bias, like X-ray data collected only from men. It ends by saying that choosing data is a form of programming: the data is the code.
You'll be able to
- describe ways training data is collected.
- explain what bias in training data is and how it causes biased predictions.
- use two questions to check whether training data is good enough.
Key words
- Training data
- The examples a machine learning system learns from.
- Bias
- A blind spot in data that favours some things and de-prioritises or leaves out others.
- Prediction
- The computer's guess or decision about new data, based on what it learned.
- Algorithm
- The set of steps a computer follows; in machine learning it is shaped by the training data.
Transcript
Machine learning is only as good as the training data you put into it. So, it's super important to use high quality data, and lots of it. But if data is important, it's worth asking where does training data come from? Often, computers are collecting training data from people like you and me, without any effort on our part.
A video streaming service might keep track of what you watch, then it can recognize patterns in that data to recommend what you might want to watch next. Other times, you're directly asked to help, like when a website asks you to spot street signs and photos, You're providing training data to help a machine learn to see, and maybe even one day drive.
Medical researchers can use medical images as training data to teach computers how to recognize and diagnose diseases. Machine Learning needs hundreds and thousands of images, and training direction from a doctor who knows what to look for, before it can correctly identify disease.
Even with thousands of examples, there can be problems with the computer's predictions. If X-ray data is only collected from men, then the computer's predictions may only work for men. It may not recognize diseases when asked to diagnose the X-ray of a woman. This blind spot in the training data creates something called bias.
Biased data favors some things, and de-prioritizes or excludes others. Depending on how training data is collected, who is doing the collecting, and how the data is fed, there is a chance that human bias is included in the data. By learning from bias data, the computer may make biased predictions, whether the people training the computer are aware of it or not.
When you are looking at training data, ask yourself two questions: Is this enough data to accurately train a computer? And, does this data represent all possible scenarios and users without bias? This is where you, as the human training, play a crucial role. It's up to you to give your machine unbiased data.
That means collecting tons of examples, from lots of sources. Remember, when you pick and choose data for machine learning, you're actually programming the algorithm, using training data instead of code. The data IS the code. The better the data you provide, the better the computer will learn.
Video © Khan Academy, used under CC BY-NC-SA. Watch on Khan Academy. Puzzicle is not affiliated with Khan Academy.