Event Wrap-up: WTF is… Machine Learning?

What is the best way to spend a Saturday? Learning about Machine Learning is definitely at the top of the list! A week ago, around 40 curious participants joined us at VietStartup London for our third entry in the “WTF is…” series — the WTF is… Machine Learning? event. With the goal of knowledge sharing in mind, the organizers, machine learning PhD and professionals, delivered two sessions — a talk and a practical business case in the morning, and a hands-on coding session with machine learning in Python in the afternoon.

1. WTF is… Machine Learning Talk

Machine Learning, especially Deep Learning, has gathered much hype in the media lately, with many promising projects such as self-driving cars or AlphaGo. So it is important that the first talk put things into context, to give audience a better understanding and debunk a few misconceptions. Our speakers in this part were Tuan Anh Le — Machine Learning engineer at Babylon Health, and Tuan Anh Le — PhD student at Oxford.

The Tuan Anh duo explained what Machine Learning is and why every company is talking about it, touching the history, the whys, the hows, and the problems of ML nowadays. We wanted all the audience, even those who have no background in computer science, to understand the basic intuition of ML, with the help of visual illustrations of Netflix, cat videos, and a classic image recognition quest of classifying hot dog or not hot dog.

Here are a few takeaways from the talks:

• Machine Learning can be phrased as … “how computers learn to solve problems without being explicitly programmed”.
• Foundational ML techniques were invented decades ago, and ML has experienced multiple ups and downs throughout the years. ML, and by extension, AI, is not a shiny new field as people might think. ML and AI becoming hot technology in recent years were partly thanks to the exponential increase in computing power and labelled data.
• The core components behind the recent success of ML are: labelled data, a function to transform the data, cost between the true label and the predicted label, and optimization of said cost. Deep Learning, as Tuan Anh wittily said, is essentially just a longer (deeper) function.
• While having awesome results in a variety of tasks (object recognition, speech recognition, etc.), ML is not a one-size-fits-all tool. It has low generalization capability, is potentially inaccurate due to bias in training data, and must be tweaked intensively to perform well.

This talk stirred up the audience and we had an interesting Q&A round with discussion on comparison between Artificial Intelligence and human brain, and how likely is AI to take over the world (highly unlikely in the near future!).

2. Crack the Love Codes Business Case

The morning session transitioned from theory to a more practical talk, discussing how ML could be applied to a business scenario: predicting high-risk users of a dating website. Phuong Hoa Giang, the presenter for this exercise, is a data scientist at Microsoft, and she “cracked the love codes” by guiding the audience through steps to translate a business problem into a data problem, and translate data results back to business actions.

The audience were put into the shoes of an analyst at the dating website, and brainstormed the approach to utilize the data of male users, including height, income, and age, and their status of being matched after a month, in order to increase retention. The Machine Learning approach is to connect these data, a.k.a the “features”, with their status, the “target”, and use Logistic Regression to predict which users are more likely to not get matched, as well as which features affect the probability of getting matched.

(For those who are interested, the short answer is: For males, height and age don’t matter in finding a match, only income does! (disclaimer: conclusion is drawn from the given dataset))

3. Coding Session with Linear Regression and Logistic Regression

After learning the basics, the afternoon was time for the audience to get their hands dirty by coding themselves Linear Regression and Logistic Regression algorithms in Python, via a series of 3 notebooks prepared by our team.

Linear Regression, which is familiar from any Statistics course, is reintroduced under of the lens of Machine Learning, using gradient descent to minimize the average distance between true target values and predicted target values.

On the other hand. Logistic Regression wraps a sigmoid function around the linear function, making it suitable for classification task: instead of predicting a real number, it predicts a status, which can be only 0 or 1. The cost function is also slightly more complicated, but the idea is very similar.

Delivered with gusto in an impromptu lecture in MS Paint by Dat Nguyen — data scientist at Zopa, followed by a presentation filled with memes and scary maths equations by Manh Dao — data scientist at Pearson, basic concepts such as matrix multiplication, cost function, gradient descent, etc. were made easy. The audience then had a chance to translate these maths equations into Python codes, under the helpful guidance of our organizers. Even those with no coding experience could learn something thanks to the amazingly comprehensive “Introduction To Python” notebook. The afternoon session was the first attempt of VietStartup London at more hands-on knowledge sharing, and it was embraced with enthusiasm.

Many thanks again to Thinh Truong Ha — data scientist at Barclays, Dat Nguyen, for devoting their time to prepare the notebooks alongside me, and to all the organizers for helping with this session!

At the end of the day, the event wasn’t aimed at teaching Machine Learning. It was envisioned by us to give all curious minds, regardless of their level of experience, the intuition and a taste of ML, and more importantly, to inspire them to learn, to read, to discuss, or to try Machine Learning at their work or personal projects. The maths and the coding might be intimidating, but are helpful in demystifying the magic behind Machine Learning. We have also compiled a list of resources for any beginners to start their ML journey here.

Stay tuned for more “WTF is.. “ from VietStartup London. We might do an event on Deep Learning next time. Give us a shout if you really want it to happen, or if you’d like to get involved. Prepare your Python!

VietStartup London is a community of people passionate in tech and entrepreneurship, we connect people, share knowledge, develop projects, and leverage a unique cross-perspective between the UK and Vietnam.

Join us here on Facebook: https://www.facebook.com/groups/284739328332602/