🛠 Introducing the Real World ML Section

We keep you updated with the most important things that happen in the ML world

📝 Editorial 

Building machine learning (ML) solutions at scale remains an unexplored territory for most companies. Most data science teams have solid ideas of managing the lifecycle of a handful of ML models but how does an ML infrastructure for hundreds of thousands of models look like? Even though the MLOps space has been growing at a rapid pace, the architectures and best practices for applying those stacks at scale are being learned by trial and error. In the current ML market, some of the most advanced ML infrastructures are being built by large technology companies such as Facebook, Google, Uber, LinkedIn, Netflix and others. Analyzing those architectures is one of the most efficient ways to understand the potential challenges and solutions of large-scale ML architectures.  

With this edition of TheSequence Scope, we have added a small section titled Real World ML. The objective of this section is to highlight new, documented best practices adopted in some of the largest ML infrastructures in the world. We think that systematically studying the ML architectures and techniques implemented by large technology companies is one of the best sources of inspirations you can find in the ML world. We hope the Real World ML section will help evangelize some of these ideas. For this week, we’ve included some new details about ML use cases at Uber and LinkedIn.

Happy Reading! 

Share


🔺🔻TheSequence Scope – our Sunday edition with the industry’s development overview – is free. To receive high-quality content about the most relevant developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻

🗓 Next week in TheSequence Edge:

Edge#109: The start of the Transformers series (exciting!)

Edge#110: Overview of Pachyderm, a platform to streamline machine learning experimentation 


Now, let’s review the most important developments in the AI industry this week

🛠 Real World ML

Orders Near You

The Uber engineering team published a detailed blog post about the implementation of the orders near you, a feature in the Uber Eats app ->read more on Uber Engineering blog

Large Scale Data Analytics at LinkedIn 

The LinkedIn engineering team published a blog post detailing the architecture of their big data pipelines to power analytics workloads ->read more on LinkedIn Engineering blog


🔎 ML Research

BlenderBot 2.0 

Facebook AI Research published a paper detailing the second version of its BlenderBot chatbot that incorporates long-term memory and internet knowledge capabilities ->read more on FAIR blog

Feature Learning with Super Wide Neural Networks 

Microsoft Research published a paper proposing a technique capable of feature learning in infinitely scalable deep learning models ->read more on Microsoft Research blog

Vision-Language Contrastive Learning 

Salesforce Research published a paper detailing  ALign BEfore Fuse (ALBEF), a model that uses contrastive learning to achieve state-of-the-art performance in different language-vision tasks ->read more on Salesforce Research blog


🤖 Cool AI Tech Releases

TensoRT8

NVIDIA open-sourced the new release of its popular TensorRT framework designed for high speed, large scale inference jobs ->read more on NVIDIA Developer blog

TonY Goes to the Linux AI Foundation

LinkedIn’s TonY is a framework designed to enable the training of deep learning models in a Hadoop infrastructure, it just joined the Linux AI Foundation as an incubation project ->read more on LinkedIn Engineering blog

Facebook FSDP

Facebook open-sourced Fully Sharded Data Parallel (FSDP), a framework for large scale training with fewer GPU resources ->read more on Facebook Engineering blog


🗯 Useful tweet

NetHack Challenge, which we’ve covered in Edge#100, announced a new track


💸 Money in AI

For devs and engineers:

AI implementation: