📝 Editorial 

Federated learning is often regarded as one of the most important machine learning (ML) techniques for privacy and one of the cornerstones of mobile ML (we covered it in Edge#5). The core idea behind federated learning is that multiple agents (think mobile devices) can collaborate in mastering a specific task without relying on centralized training data. Think about a mobile ML models distributed in a mobile app across millions of devices. Ideally, the model can benefit from the data produced by each instance of the app, but that entails very concerning privacy vulnerabilities. Federated learning enables a way in which only updates on the model are distributed to a centralized location while the training data remains on the device.  

Since Google pioneered the idea of federated learning in 2017, it has become one of the most important methods for secured learning across many agents. However, federated learning implementation remains scarce, primarily due to its technical challenges. Simulating a large number of intelligent agents is not exactly an easy task. Among the labs advancing federated learning research, Microsoft and Google Research seem to be leading the charge. Last week, Microsoft Research open-sourced what can be considered one of the most important contributions to the short history of federated learning.  

Federated Learning Utilities and Tools for Experimentation (FLUTE) is a framework for running large-scale federated learning simulations. FLUTE provides an architecture that allows data scientists to simulate agent interactions in federated learning architectures and find the right balance between training data and privacy boundaries. The framework’s key contribution is to facilitate the experimentation in highly sophisticated federated learning scenarios without requiring large computation resources. Certainly, FLUTE is likely to play a pivotal role in advancing federated learning implementations in real-world scenarios.  

🔺🔻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#193: a recap of the series about distributed training;

Edge#194: a deep dive into Masterful AI, an AutoML training platform for deep learning.

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

🔎 ML Research

Google Assistant’s Contextual Rephrasing 

Google Research published a blog post describing the ML architecture that Google Assistant uses to better understand contextual information →read more on Google Research blog

Microeconomic Behavior in RL Agents 

DeepMind published a paper exploring how populations of RL agents can organically learn microeconomic behaviors →read more in this summary from DeepMind 

Speech Recognition at Meta 

Meta AI published a blog post highlighting some of their recent research in speech recognition →read more on Meta AI blog

Vector-Quantized Image Modeling with Improved VQGAN

Google Research published a paper describing a technique that combines GANs and transformer models to improve image generation →read more on Google Research blog

🤖 Cool AI Tech Releases

Federated Learning Simulations 

Microsoft Research open-sourced FLUTE, a framework for running large-scale federated learning simulations →read more on Microsoft Research blog

PyTorch Training on Macs 

PyTorch announced a training accelerator that leverages Apple silicon GPUs →read more on PyTorch blog

🛠 Real World ML

MLOps Portal at LinkedIn 

LinkedIn provided details about its Pro-ML Workspace, a portal that provides an interactive experience to manage the lifecycle of ML models →read more on LinkedIn Engineering blog

💸 Money in AI

ML&AI&Data

• 🦄 Data intelligence company Near goes public on Nasdaq via SPAC with nearly $1 billion post-transaction pro forma market capitalization. Hiring in California/US.

• 🦄 Real-time database startup Imply raised a $100 million Series D funding round led by Thoma Bravo growth. Hiring in the US and India.

• Heartex, the company behind the open-source data labeling platform Label Studio (covered in Edge#119 and #120), raised $25 million in a Series A funding round led by Redpoint Ventures.

AI-powered

• Construction management platform Buildots raised a $60 million Series C round co-led by Viola Growth and Eyal Ofer's O.G. Tech. Hiring in Tel Aviv/Israel, London/UK, and the US.

• Forma.ai's sales compensation platform Forma.ai raised $45 million in a Series B funding round led by ACME Capital. Hiring remote in Canada.

• Recruiting automation platform Fetcher raised a $27 million Series B investment round led by Tola Capital. Hiring in Bogota/Columbia, Buenos Aires/Argentina and Philipines.

• E-commerce content generation platform ZMO.ai raised an $8 million Series A financing round led by GL Ventures.