finneganpierson's blog

By finneganpierson, history, 6 weeks ago, In English,

In a world with ever-evolving smart technology and AI, understanding how machines are programmed to perform tasks becomes more and more complicated all the time. While the actual coding and commands may be confusing (at best) to laypeople, the basic concepts are surprisingly simple. This is because people and machines learn how to find answers in the same way.

Supervised Learning-How We Are Taught ****

Think back to elementary school spelling. When you were being taught how to spell words, you probably had a teacher (or supervisor, if you will) break down the words letter by letter or sound by sound until you learned them by rote. Now, take this literal lesson and think of it as an analogy or how a programmer might teach a machine to recognize a command. The programmer (also a supervisor) will teach the information by feeding data commands that make it possible to determine information about how to recognize a direction or command.

Unsupervised Learning-How We Learn Without Guidance **** To answer the question “what is unsupervised learning?”, we can go back to the same example of the elementary school spelling lesson. After the supervisor/teacher has taught you rudimentary letters and sounds, let’s say they write a word on the board and walk to the back of the classroom and as you to spell it without any help or prompting from them. You will look at the word and use information that you have already been taught to solve the problem of how to spell it. For example, you have been taught to recognize the shape and sound of the letter “A.” If you see the letter “A” in the word, especially see it coupled with another letter you recognize, and your brain will start solving the problem by interpreting familiar items.

Let’s look at a more literal exercise in unsupervised learning that may come up for a machine. If you show it a picture that contains zebras and kittens and ask it to draw a circle around all the zebras, it will need to employ its resources to sort them. It will key in on their similarities and differences and group them accordingly. It will see the zebras have stripes while the cats do not. It will separate them accordingly.

The Two Algorithms **** Unsupervised learning has two over-arching algorithms that machines use to learn, and problem solve. The first is “association.” Associations usually are applied to data to analyze trends. They can create “if, then” hypotheses, like “people who buy mittens also tend to buy scarves. The other algorithm is called “clustering.” In clustering, items are separated into groups based on shared characteristics or habits, like our zebra/kitten exercise.

Real-World Applications **** Visualization is one of the top real-world applications of unsupervised learning. If you have set up (or have access to) a visualization algorithm, you can supply unlabeled, complex data into it, and receive a two- or three-dimensional representation of it. This is great for tracking trends and statistics. Dimensionality reduction (also known as feature extraction) is a means to simplify very complex data with a minimal loss of information. The goal is to speed up processes, and this is accomplished through unsupervised learning algorithms. Lastly, anomaly detection is an extremely useful association-algorithm-based process of unsupervised learning in which unlike items are isolated for further analysis. This can be very useful for fraud prevention departments in banks or credit card companies. If anomalous purchases show up, they can look be alerted and set an investigation into motion.

Ultimately, it does make sense that the way we learn is the same as how machines learn. People created computers, after all. By seeing how supervised and unsupervised learning work, you should have a better understanding of what programmers do, and how systems learn their tasks and functions. You should also have a good grasp of some real-world applications that would be much more difficult, if not impossible, without unsupervised learning.

  • Vote: I like it
  • -26
  • Vote: I do not like it

6 weeks ago, # |
  Vote: I like it 0 Vote: I do not like it

Adversarial examples — change a few pixels in an image to make a machine see it as something completely different — show that machines don't learn like humans at all.