Why is Machine learning not applied statistics?


I’m pretty much self taught software engineer so I probably know about machine learning more than the average person, but haven’t really tried to use it in any way.

Since I skipped uni (most unis would have statistics in the curriculum), I haven’t really touched statistics either and all I know about it is how to make a graph in excel.

My take:

Definition of statistics:

> the practice or science of collecting and analysing numerical data in large
> quantities, especially for the purpose of inferring proportions in a whole from
> those in a representative sample.

And machine learning does exactly that. Collects and analyses some data and infers whatever it was trained to infer.

My understanding is, that it actually is applied statistics, but I’ve read some articles that say it isn’t. I didn’t really understood why.

I understand the data gathering is different, the problems it’s trying to solve are different and even the “how it works behind the scenes” is different. But I don’t see how it’s abstract core differs.

so .. why ML isn’t applied statistics?

In: Technology

I have a masters in cs with a focus in machine learning.

Statistics is used heavily in machine learning but it’s a bit like asking “why is engineering not just applied physics.” There is a lot of application and design that goes into building the right agent for the right

Machine learning has a lot of subfields, one of which is supervised learning. This is closest to what you describe. Take a training set to train a model to identify the trained inputs in a new set of data. There is also another section called unsupervised learning which could be though of as trying to group the data into categories as well as possible (not based on training data).

Even farther away from those examples, there is reinforcement learning. Simply put, you tell the model to work towards something and then it learns to do it over time based on feedback from the environment. This learning is more “on line” and similar to how people learn. This is kind of like learning on the job because the agent is often having to explore what makes a good decision and what makes a bad one.