Newsletter #6: Product Management For AI, Reducing Language Bias, Personalisation At Scale And More
Testing FireFox With ML, Language-Agnostic BERT Sentence Embedding, Effective Tech Lead, Personalisation At Scale And More
Hi there,
Welcome to the 6th edition of HubOfML newsletter. Here are the top posts and papers I thought were worth sharing with you this month.
Machine Learning (ML)
What You Need to Know About Product Management for AI
A product manager for AI does everything a traditional PM does, and much more. If you’re already a software product manager (PM), you have a head start on becoming a PM for artificial intelligence (AI) or machine learning (ML), in this post, you will learn what it means to be a product manager for AI or ML. Read more.
A Scalable Approach to Reducing Gender Bias in Google Translate
Societal biases in languages can find their ways into training data for Machine Learning models for language translation. One such example is gender bias. On Google AI blog, Melvin Johnson, a senior engineer at Google, wrote about Google's approach to combating language bias in Google Translate.
Testing Firefox More Efficiently with ML
Browser is an incredibly complex piece of software. With such enormous complexity, the only way to maintain a rapid development pace is through an extensive CI system that can give developers confidence that their changes won't introduce bugs... Here is how Mozilla is testing firefox with ML
Language-Agnostic BERT Sentence Embedding
Google recently shared research on a multilingual BERT embedding model, called LaBSE that produces language-agnostic cross-lingual sentence embeddings for 109 languages. The model was trained on 17 billion monolingual sentences and 6 billion bilingual sentence pairs using MLM and TLM pre-training. A released the pre-trained model is available tfhub, which includes modules that can be used as-is or can be fine-tuned using domain-specific data. Read more.
Personalisation at Fynd
In this post, Vignesh Prajapati wrote about how Fynd builds their new personation engine to improve the shopping experience for customers.
From Pytorch to Pytorch Lightning
This's a well written introduction for Pytorch's users who want to get started with Pytorch Lightning from William Falcon, one of the top Pytorch Lightning contributors. It highlights key distinctions between Pytorch and Pytorch Lightning.
Product Categorization At Scale At Shopify
Shopify houses billions of products from 1M businesses. For Shopify, understanding the types of products companies sell is needed to provide personalized insights to help enterprises to capitalize on valuable business opportunities. In this post, Shopify describes how they built an ML to categorize shopping products to deliver personalized insights.
How TikTok Recommends Videos
The TikTok ForYou feed is powered by a recommendation system that delivers content and creators of interest to a particular user. This post explains how TikTok teams work to counter issues faced by all recommendation services.
Give Me Jeans Not Shoes
Stitch Fix helps people find clothes they love by combining technology with their torch of seasoned style experts. Here is how Stitch Fix is using BERT to deliver personalized styling for users.
Software Engineering
What It Means to Be an Effective Tech Lead
I've read many stories of engineers who got promoted to tech lead positions without prior training. One thing that's common in all the stories I read is, "nobody prepared me for the role." But what does success look like for a tech lead? Edmon Lau especially enjoyed this post, where he provided a general framework for thinking about success as a tech lead.
How to Make Your Engineering Team More Effective
As a leader, how you spend your time is critical. You have an incredible opportunity for impact because your decisions directly affect the output of your entire team. If you're a team lead/manager/senior person on a software team, Edmon Lau provides practical ways to grow your team to become more effective.
High Leverage Code Reviews
Code reviews are a vital process of building a software product. In addition to ensuring quality, it's also an efficient way of spreading knowledge. This post shows you how to approach a pull request to provide meaningful code reviews.
Solving Data Discovery Challenges At Shopify
Every two days, Shopify creates as much data as they created from the beginning of time until 2003. According to Shopify, finding answers to data problems often involved asking team members in person, reaching out on Slack, digging through GitHub code, sifting through various job logs, etc. In this post, Ranko Cupovic wrote about how Shopify solved data discovery challenges by building an internal tool called Artifact.
Deployment at Slack
Slack does about 12 deployments per day. Each deployment requires a careful balance of speed and reliability. This post describes the steps each deployment goes through before hitting the production.
Papers
Intermediate-Task Transfer Learning with Pretrained Models for Natural Language Understanding: When and Why Does It Work
Thanks for reading
Thanks for reading! If you like this newsletter and want to support it, please share with others.
Cheers,
Samuel