10.17.2019 Machine Learning

A new buzz word in computer science in the last decade is machine learning. What is machine learning? In the shortest way possible, it is a computer algorithm that is designed to learn something by recognizing patterns in the data, much like a child learns to speak a language by hearing the language over and over again, recognizes patterns, and learns to speak with proper grammar over time.

Ironically the first major use of machine learning was in natural language translation. Algorithms were developed to take large samples of several languages and learn how to put translated words in the correct order. It might seem like the computer is able to think for itself about grammar and languages, but this is not the case at all. It is simply able to copy from examples. 

There are two stages of machine learning. The first stage is the training phase, during which an extremely large dataset is manually labeled by a person so the computer can learn. For example, in the language translation, a person has to label each word in all the training data as a part of speech, i.e., noun, pronoun, verb, etc. This labeling allows the algorithm to recognize the order of the parts of speech in a sentence to correctly order the translated words. This training process is a long and tedious process and any word that is not classified will confuse the algorithm.

The second stage is making predictions. This is the part of machine learning you get exposed to every time you ask Google to translate a phrase, or see a product recommendation while shopping online.  The better you train the model, the smarter the predictions get. Google Translate retrains their translation models using feedback from users of the tool every few days, so it is constantly improving.

To train machine learning models requires large amounts of computing power and memory. The process takes, in the case of Google Translate, years to accurately train the model. There is a need for much faster training processes before machine learning starts to look anything like human learning, which is the goal of artificial intelligence. Next week I will write about some of the new training methods that are forthcoming.