🌊 Google’s New Wave of Machine Learning Capabilities

📝 Editorial 

Microsoft, Amazon and Google are embarked on a frantic race for artificial intelligence (AI) cloud dominance. While in other areas of the cloud space, these technology giants match each other literally feature by feature, machine learning might be the area where they all exhibit highly differentiated capabilities. AWS and Azure are certainly ahead in terms of general customer adoption but machine learning remains one of the strongholds of the Google Cloud offering. This week, at its annual I/O conference, Google unveiled a new series of capabilities that continue to enhance its already robust machine learning offering.  

The AI announcements at Google I/O were highly diverse. One of the most interesting was the unveiling of Vertex AI, a new managed cloud service to accelerate the deployment and maintenance of ML models. Vertex AI is a particularly interesting release and it shows significant overlap with Google Cloud ML. Another announcement that captured the headlines was LaMDA, a new conversational model that can produce more natural dialogs by better understanding its context. Google also announced a new generation of Tensor Processing Units (TPUs) chips to power AI workloads as well as new AI-native personalization capabilities to its Firebase mobile backed platform.  

The new additions to the Google Cloud continue to push the boundaries of AI/ML innovation in areas such as hardware, mobile, and language, in which Google has traditionally been particularly strong. Just like in previous years, Google I/O 2021 didn’t disappoint from the ML perspective.  


🔺🔻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#91: we discuss what is model-free Reinforcement Learning; we explore Agent57, a model-free reinforcement learning (MFRL) agent that outperformed the standard human benchmark on all 57 Atari games; we explain DeepMind’s OpenSpiel – an open-source reinforcement learning framework for games

Edge#92: deep dive into Cogito, a workforce solution provider for data annotation, content moderation and other data processing services

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

🔎 ML Research

Efficient Transformers for Video Intelligence 

Microsoft Research and Nvidia collaborated on a new research paper that proposes an efficient transformer architecture for video representation learning ->read more on Microsoft Research blog

Integrating Knowledge Graphs with Language Models 

Google Research published an intriguing paper proposing a model that can convert structured information in a knowledge graph into language representations that can be used to train more efficient transformer models ->read more on Google Research blog


Facebook AI Research(FAIR) published a paper unveiling Expire-Span, a technique to teach deep learning memory systems to “forget” at scale in order to improve their memory ->read more on FAIR blog

❗️Call for writers ❗️

TheSequence is looking for freelance writers who cover machine learning and artificial intelligence space in-depth. Send links of your blog or published articles to ks@invectoriq.com.


🤖 Cool AI Tech Releases

Vertex AI 

During its I/O conference, Google announced the release of Vertex AI, a new managed machine learning platform for both data scientists and business users ->read more in this press release from the Google Cloud team


Google also demoed LaMDA, a new language model for better understanding of conversations. Like many recent language models, including BERT and GPT-3, it’s built on Transformer. We’ve covered Transformers in Edge#3 and will be doing a series on them soon ->read more about LaMDA on Google blog


Uber open sourced Orbit, a new framework for statistical time series inference and forecasting ->read more in Uber engineering blog

TensorFlow MoveNext 

TensorFlow just added MoveNext to its pose detection libraries. The new model is able to detect over 17 key points in the human body in a fast and accurate way ->read more on TensorFlow blog

💬 Useful tweet

Follow us on Twitter

💸 Money in AI

  • Data science platform Explorium raised a $75 million funding round led by Insight Partners. The startup develops an automated data and feature discovery platform that automatically discovers thousands of relevant data signals and uses them to improve analytics and machine learning.

  • Cloud-native data catalog startup Data.World raised an $11.05 million growth round led by Vopak Ventures. The company helps its clients validate, manage and organize data to improve model deployment for various processes.

  • Deployment-as-a-service library and AIOps platform Coiled raised $21 million in Series A funding led by Bessemer Venture Partners. Coiled helps the data science community build scalable, distributed, and intelligent applications powered by Python.

  • No-code data automation platform Syncari raised $17.3 million in a Series A led by Crosslink Capital. The startup empowers operations professionals to unify, clean, manage, and distribute trusted customer data across the enterprise.