Machine Learning
When most people hear “Machine Learning” they might picture a robot doing something or kind of fictional characters from movies like terminator depending on whom you ask. But Machine Learning is not a futuristic fantasy, it’s already here. In fact, it is around us from several decades in some specialized applications, such as Spam Filters in mail. In this article we will further explore about machine learning from what to why, you will also try out yourself using a practical dataset.
What is Machine Learning?
First, we will see some formal definitions
Machine learning (ML) is a field of inquiry devoted to understanding and building methods that ‘learn’, that is, methods that leverage data to improve performance on some set of tasks. — Wikipedia.
Machine Learning (ML) is the field of study that gives computers he ability to learn without being explicitly programmed. — Author Samuel, 1959
A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E. — Tom Mitchell, 1997
These definitions may feel overwhelming when you are hearing about them first time but don’t worry think ML as some kind of intelligent systems that understand underlying data patterns and act accordingly. Your initial concern right now is probably how ML systems conquered conventional programming?
Let’s just explore how ML used where traditional programming can’t be the solution. Consider that you are creating a filter for the spam mail protection of your organization.
- Traditional way of Programming
- Study the problem.
- Write some rules which detect spam mails.
- Evaluating the results which may lead to two scenarios:
1. If it passes all test conditions launch the algorithm.
2. Else re-write the rules which have missed previously.
Does this approach actually aid in the detection of all potential spam emails? The apparent answer to this question is a resounding NO, and we must alter and adapt the new rules each time a new spam message appears, which is tiresome, time-consuming, and increases the program’s line count. ML can truly assist us in this situation.
- ML way of Programming
- Study the problem.
- Gather relevant historical data.
- Adapt a machine learning method to the nature and kind of the data.
- Fit the data and evaluate the algorithm performance.
1. If it performs well publish model.
2. Else re-tune the model using updated data, or some missed use-cases.
Now you might think that even we are tuning here then what might differ? The answer to this question is instead of formulating rules every-time we see a new example in ML we can automate model to adapt the changes and learn as it cannot identify some new cases. The major benefit of ML is that it may find previously unknown data by past experience or learning that it has done through data fitting.
Note ML is best when:
- Issues for which the current solution calls for extensive manual tweaking or a big number of rules: ML may improve performance and make code simpler.
- Environments that are always changing: A machine learning system can adjust to new data.
- Gaining knowledge regarding difficult issues and massive volumes of data.
Do we need to bucket our problem while using ML?
As every challenge has a unique solution, ML also provided us with several buckets to categorise our problem depending on the kind of data we utilise.
Conclusion:
On closing note, we are living in the era of intelligent systems roaming with us in our pockets, hands and houses helping us doing things simpler and easier with simple commands and clicks like google photos bucketing photos for us, mail spam filter, Product and movie recommendations from amazon, Netflix etc. Hope this article has introduced ML in understandable format and I would like to hear back from you.
