A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.
I don’t know about you but concrete objects and real stories make abstract definitions like these much easier to understand and remember.
Stated another way, a computer program is said to learn if it improves as it gains experience.
For example, say you have a computer program that is used to estimate the current value of an apartment. The performance of the program can be calculated by comparing the actual buy/sell price and the estimate given by the computer program. The program is said to learn if the accuracy of the estimates improve as it gains more experiences estimating apartments.
Based on the initial definition:
- T = the task of estimating an apartment
- P = the improve accuracy of the estimation compared with the actual buy/sell value
- E = the experience of estimating apartments
As such, the true power of Machine Learning comes from the incremental improvements that can be made over time as it gains experience and learns.
For me this is why applications like Google Translate, while at times it can give strange and completely wrong translations, will over the long term completely overtake and dominate markets currently powered by people.