The Basics of Approximating Functions in Machine Learning

By Daniel Glass

Originally Posted on LinkedIn

In our last post, we considered warehouse task time prediction as a real world example of a function that we want to approximate. In order to approximate it, we need 2 things:

1. lots of examples of tasks (data)

2. starting point for our function approximation (model)

As far as task examples are concerned, let's imagine that we have a list of every task ever performed in our warehouse. This includes the activity, the unit quantity, the tenure of the associate, how far they had to go, the speed of the equipment they are on, etc. etc., and the time it took to complete the task.

Regarding our function approximation, we will call this our model. In machine learning, different types of models are used depending on the problem. For example, large language models look different than image recognition models. Let's not dig too deep into what our model looks like -- just imagine it is similar to our '2x + 5' example. The goal is to be able to pass task characteristics (input) through our model (function approximation), and produce the same elapsed time (output) that we observed in the real world.

So, we have our data and our model starting point... we are done right?

Not just yet -- if we pass input into our model right now, we should see that our task time predictions do not line up with reality. How do we train our model so that it produces outputs that align with reality?

While we wait for the next part of our series, check out how Cellaware has implemented real world applications for AI/ML in the warehousing and distribution industry. Schedule your free demo today!