Debunking Common Machine Learning Myths

There is a lot of talk about machine learning these days and talks often propagate myths. In this article, we would like to bust some machine learning myths that are far from reality.

Artificial Intelligence refers to the ability of machines to solve problems and learn. These systems mimic the human cognitive functions, allowing decision-making. Machine learning is the subfield of artificial intelligence. It aims to develop methodologies that allow machines to learn. Machine Learning is the ability to identify complex patterns in data sets. It also builds models using algorithms and predicts future behaviors.

This is a popular topic of discussion these days. But the concern that ML and AI-based systems/robots will replace humans is not ambiguous. Robots are automating manual work at many places like factories, production processes, and medicine. But, their implementation is as assistive technology and not as employee replacement. ML has simplified many redundant processes and also improved process management and productivity.

Read more – Myth Busting – Artificial Intelligence Will Take Your Job

Myth #4: You can “just implement machine learning once and forget about it”

Let us take an example of information security here. There is a huge difference between detecting any system issue and detecting faces. Faces are faces today and they remain so tomorrow, nothing changes, except for aging. But new security issues can come up any day. If we talk about the application of ML in cybersecurity, you can’t just install and forget. Everything is changing. Criminals are people with a specific motivation (money, espionage, terrorism, etc.) and their intelligence is not artificial. They can change malware so that smart systems don’t catch them.

Machine learning models are usually constant. But, in information security, you may have to train your ML models from scratch. A security solution that does not update its antivirus base will be useless in long term. Criminals can be dangerously creative when necessary.

While working on an ML project then it might tempt you to add more data points to the training data. Yet, it is not always a very good idea to dump in more and more data.

If you feed too much data to your model, it will memorize it and there can be the issue of model overfitting. This further leads to high error rates for unseen data. Keep in mind – ‘Garbage In Garbage Out’ principle. You don’t just need to add data but high-quality data that can help in creating better quality ML models.