• Grey Twitter Icon

ML 101: for beginners

A Guide to Machine Learning*

*and may be even life. Hint: it’s all about asking good questions, being honest about your goals, and evaluating accommodations and tradeoffs.

Introduction

There’s a lot of talk about data, machine learning, deep learning, artificial intelligence (AI). And these talks range from calling AI the most extraordinary idea of our time (one that would make us superhuman or destroy us) to calling AI a boring hyped idea. It doesn’t really matter which one you believe. Because current AI systems are both far more boring and far more extraordinary than you actually think – in the sense that they are quite boring compared to our sci-fi stories, but are extraordinary because of the impact they are already having.

 

Most recent advances in AI have been achieved in the area of Machine Learning (ML) (and Deep Learning (DL)). Almost everything you think of when you think of existing AI is in the area of ML.

 

Machine learning is not a simple process. And it is not an easy process.

 

The hardest challenge is not a math challenge or a technological challenge (even though they are hard). The hardest challenge is often being uncomfortably honest and explicitly clear about the problem, goals, constraints, and tradeoffs. And we’re usually not used to this type of honesty and clarity. But once those conversations are done, they lead to better conversations about data, technologies, tools, algorithms, accountability, and actions.

Why should you learn about any of this?

The field of AI is only a few decades old, and applications of machine learning and deep learning are much much younger. In some sense, they’re infants. Yet, these applications are (already) both inspiring and intimidating us. They are both saving our lives as well as running us over in a car. 

 

Many important decisions now involve machine learning systems and they are affecting us at a deeply personal level. We are no longer mere “users” of these ML systems, we are intricately intertwined with these systems. And failures of these systems could eventually lead to failures of different parts of our lives.

 

All of us must learn to ask, explore, and understand potential limitations and implications of these systems, before we get carried away with using them. We must be as deliberate and responsible as possible while designing such systems, but also while buying and interacting with such systems.

Why another ML workshop?

Because this crucial element is missing from most workshops: ML systems are all about tradeoffs and accommodations. Sort of like life. You can’t have it all. 

 

It turns out that at every step of building any ML system, important decisions will have to be made, all of which will require you to understand data, models, evaluations, accommodations, and tradeoffs.

 

Tradeoffs involve balancing somewhat incompatible components of your situation and accommodations involve some sort of a compromise, a convenient arrangement of some type.

 

In life, you need to understand your situation, understand your objectives, have a robust process and trustworthy models (of people, of environments, of society, etc.) to make good decisions. Even when you do these well, you are still not always right. But when you make poor decisions, you’ve usually gotten one or more of the above wrong.

 

The same is true for ML systems: in order for you to trust your system, you need good data, you need good models, and you need good tests. Understanding and trusting data, models and their results is a hallmark of good process.

 

I believe literacy about tradeoffs and accommodations in the context of ML systems is crucial, so more people can understand ML and AI systems they interact with, and most importantly can participate in discussions about building future ML and AI systems.

About this workshop

This workshop is organized for absolute beginners...well, pretty much anyone who’s curious and doesn’t know where to start. I will give an overview of ML, for all.

 

The goal is to provide a view of machine learning that focuses on ideas, concepts, models, and most importantly tradeoffs (not yet on math or programming). Honestly, it is neither possible nor advisable to avoid math for too long, since math will usually aid understanding, not hinder it. But timing matters. So, we’ll delay it here - until next time or the time after that.