Udacity’s AI Product Manager — A review

Original article was published on Deep Learning on Medium

Udacity’s AI Product Manager — A review

I just finished Udacity’s AI Product Manager nanodegree. In this post, I share why I took the course and if I can recommend it.

My Profile

I always find it hard to read through reviews of online courses without knowing the profile of the person writing the review. Here we go:

I am a Product Manager and have worked in a few startups in different industries. I have studied Management but taught myself how to do (basic) coding. I have a large interest in machine learning and did Udacity’s Intro to Machine Learning two years ago, some Dataquest data science courses a while ago and generally followed e.g. Lex Friedman’s great AI podcast (highly recommended).

Motivation

Working as a product manager, I wanted to know when & how to leverage machine learning for building better products. Therefore I was more interested in the why, rather than the how. I did not want to learn all the theoretical details of a specific machine learning algorithm, but rather wanted to learn how to decide whether machine learning might help me at to solve a business problem. Hence, I don’t want to become a data scientist myself since I have one in my team who will do a much better job on the how.

Analysing the offer of online courses about machine learning, there are so many courses and it really took me a while to find out which ones might fit my interest.

I clustered the courses by the profile they target: There are courses that target people working in Data Science, Programming and Business / Product / Other. While the first two categories have a larger overlap, courses of the third category clearly differ in terms of content.

Here a list of the courses I looked at — some of these courses seem to have a more hands-on, top-down approach(e.g. Fast AI) while others are more theoretical and bottom-up (e.g. deeplearning.ai).

Data Scientists

Programmers

Business / Product Managers / Other

I decided to start with a course of the last category and then potentially do a more hands-on course afterwards. While deeplearning.ai’s AI for everyone course looked too basic for me, the MIT Sloan course was simply too expensive. Remaining, there were two Udacity Degrees, one targeting business leaders one targeting product managers. You can find a comparison of both degrees here:

It took me a while to decide which of the two to take. Since both programmes are very short (I get to that), I wish there was a programme covering both, since the product manager is the owner of the business case anyway.

Course Review: AI Product Manager (Udacity)

Let’s finally talk about the course: The AI Product Manager nanodegree from Udacity.

It targets Product Managers, more specifically Product Owners in an agile team who want to know how to leverage machine learning for their problem space & product landscape.

The degree has four parts. While the first part is only theory, the other parts have some theory and also a challenge where you have to hand in an assignment.

First part: Introduction to AI in business

Theory: It runs through the elementary concepts of machine learning, such as un/supervised learning, reinforcement learning, neural networks. Each topic is only about 5min, so don’t expect to get to the details. However, it gives a nice overview of the concepts and shows in which industries and for this user cases machine learning typcially is applied.

Review: I knew most of the content already but still really liked the introduction.

Part 1: The concepts and applications of machine learning. The additional links were quite useful.

Second part: Creating a dataset

Theory: This part explains how to gather and train your data. While other courses will already provide you labelled datasets to run your models on, here the challenge is actually to label the data points correctly. The videos explain how to use the Appen platform (partner of the programme) which will be used in the assignment.

Assignment: You are asked to design a data annotation job which humans will execute. Hence, you need to make sure your instructions are clear so that you won’t have any bias in your dataset. The goal is to train xray images taken of children who might show symptoms of pneumonia.

Review: The theory videos were quite helpful but mainly explained how the platform works that you will use in the assignment. The assignment itself was very interesting since it is more tricky than you might think to make sure your instructions result in good training data.

Third part: Build a model

Theory: The theory part explains how to prepare and analyse the data. The theory part is quite short and the main learning for me were the concepts of precision and recall.

Assignment: You use the dataset of the previous assignment (x-ray images) and run four different models on the Google ML platform. I really liked this exercise since you see, for example, how an unbalanced dataset affects your model’s performance. Further, it was great to use the Google ML platform and I will use for future projects.

Review: The theory was very short and I didn’t learn very much. The assignment was great, I really liked it.

Fourth part: Measuring impact and updating models

Theory: In this part you learn how to measure the impact of your machine learning product on the impact and how to roll out such a product. I honestly felt this part was quite weak and all the slides could apply to any product. This part might be only interesting to people who are very new to product management.

Assignment: You need to write a proposal for a product using machine learning. You need to describe the industry, the problem and its business value, the dataset and how to mitigate risks of bias etc. I used one use case of a company I had worked for. It was pretty easy and I am not sure how useful this exercise actually was. Further, I wrote 5 pages and got a reply within a few hours that I had passed my assignment. I think my report was pretty good, but I would have liked that I get challenged way more by Udacity’s reviewer.

Time investment:

I had bought a 2 months access to the programme since this was cheaper than paying per month. A mistake. I invested around 5–10 hours per week and finished the course in a bit more than three weeks (!).

Conclusion:

I have mixed feelings about the course. I really like the concept and idea of the course and definitely enjoyed the first two exercises — the data annotation project and the model project are really great. The theory videos are well produced and the shown content is useful in most of the cases but I can’t say that I learned a hell of a lot. You can definitely learn more for way less money — however then you also need to combine the material of different blogs/courses.

Should you take it?

If you are new to Product Management and just started looking into Data Science & Machine learning, this course might be interesting for you. However, it is also quite expensive for what you actually get. You can get the same or even better content by combining other courses and blog posts.

My next steps

Next, I want to get my hands dirty and run models and participate in Kaggle Contests. I like hands-on courses and recently watched the amazing podcast episode of the AI Podcast with Jeremy Howard, founder of the fast.ai platform:

I really enjoy his view on getting into machine learning and will take Fast.Ai’s Practical Deep Learning for Coders course.

Are you a Product Manager and also did some machine learning course? What are your thoughts?