ACM RecSys Winning Solution

You can find me on twitter @bhutanisanyam1

Subscribe to my Newsletter for updates on my new posts and interviews with My Machine Learning heroes and Chai Time Data Science

These are show notes from Episode #136 of Chai Time Data Science Show, where Sanyam Bhutani interviews the ACM RecSys winning team from NVIDIA.

The panel of guests from NVIDIA (in alphabetical order):

{Listen to show in Audio format here}

This episode covers a few topics:

Recommender Systems and what makes them an interesting area of Deep Learning and why are they still underrated.
The team at NVIDIA has won 3 different RecSys competitions in 2021, in just a span of 5 months! In this episode, they discuss the winning solution to the RecSys competition.
As someone who enjoys Kaggle, and is a fan of Kaggle Grandmasters, I also try to extract a few secret tricks from the amazing team. We do get a lot of golden, pun-intended, secrets from them!

TimeStamps to the conversation:

00:00 Intro

2:37 Panel Introduction

3:50 Why are Recommender systems an interesting challenge?

6:31 What is the Merlin Ecosystem and why was it created?

14:30 Was your Kaggle Experience helpful outside of the platform?

20:25 Problem Statement

24:56 What was the fairness requirement?

27:15 How did you approach the competition?

43:27 Feature Engineering Approach

53:35 Inferencing on GPUs, is it useful in production?

59:07 Model Overview

1:05:43 How do you find these ideas?

1:12:52 Stacking Tricks

1:17:39 What is a secret trick that you can share with the audience?

1:21:50 Outro

References:

If you’re interested in knowing more about the solution by the team, please checkout:

You can find me on twitter @bhutanisanyam1

Subscribe to my Newsletter for updates on my new posts and interviews with My Machine Learning heroes and Chai Time Data Science

Interview with the NVIDIA ACM RecSys 2021 Winning Team