✴️ Amazon’s Big ML Week
Weekly news digest curated by the industry insiders
📝 Editorial
This week Amazon hosted another edition of its prestigious re:Invent conference in which they announced the newest advancement in AWS technologies. Similar to recent years, machine learning (ML) was front and center of the announcements this week. Amazon has embarked on a frantic race with Google and Microsoft to become the dominant end-to-end ML platform in the market. The re:Invent ML announcements ranged from low-level ML-first hardware to no-code ML solutions for business analysts.
Amazon has continued to make SageMaker the core of its ML platform strategy, and they back that up this week with plenty of new capabilities. Now the capabilities of SageMaker extend to non-developers with the release SageMaker Canvas, a no-code platform for ML solutions. Similarly, SageMaker Studio Lab is a new interface that allows anyone to experiment with ML models without committing to an AWS account. For more hardcore ML engineers, SageMaker Inference Recommender is able to optimize ML models across many instances while SageMaker Training Compiler is able to take a Python ML model and generate GPU optimized code. Finally, SageMaker is entering the crowdsourced data labeling space with the addition of Ground Truth Plus, a new service that uses an expert workforce to accelerate the labeling of datasets. Other announcements included new chips optimized for ML workloads and even educational programs.
The number and depth of the ML announcements at re:Invent were overwhelming. Amazon keeps its position as one of the biggest and most complete platforms in the ML market. We will cover many of these new capabilities in future editions of TheSequence.
🔺🔻 TheSequence Scope is our Sunday free digest. To receive high-quality educational content about the most relevant concepts, research papers and developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻
🗓 Next week in TheSequence Edge:
Edge#147: we explain model serving; we explore the TensorFlow serving paper; we cover TorchServe, a super simple serving framework for PyTorch.
Edge#148: we deep dive into the OpenAI model that can solve text math problems.
Now, let’s review the most important developments in the AI industry this week
🔎 ML Research
AI and Pure Math
DeepMind published a new paper about using deep learning to accelerate discoveries in pure mathematics →read more on DeepMind blog
Microsoft’s Latest Massive Model
Microsoft Research published an insightful blog post detailing T-NLRv5, the latest model of the Turing family that outperformed many of its predecessors in several benchmarks →read more Microsoft Research blog
NeuralProphet
Meta (formerly Facebook) AI published a paper detailing NeuralProphet, the evolution of its Prophet model for time series forecasting →read more on Meta AI blog
MURAL
Google Research published a paper detailing MURAL, a multi-task learning model applied to image-text pairs across hundreds of languages →read more Google Research blog
🤖 Cool AI Tech Releases
Amazon ML Announcements
During its re:Invent conference, Amazon unveiled plenty of new ML technologies. Here are some of the most relevant additions to the AWS ML stack:
SageMaker Inference Recommender: A new addition to SageMaker to optimize model performance across ML instances.
SageMaker Training Compiler: Another new component of SageMaker that can generate GPU kernels optimized for a given model.
SageMaker Ground Truth Plus: A new service that uses an expert workforce to build high-quality training datasets.
SageMaker Canvas: A low-code tool for building ML models.
SageMaker Studio Lab: A new service that allows anyone to experiment with ML models without needing an instance.
Graviton3: A new version of Amazon’s chip for inference workloads.
Trn1: A new set of Ec2 instances optimized for ML model training.
🛠 Real World ML
Graph Neural Networks at LinkedIn
LinkedIn’s engineering team published a blog post detailing how they use graph neural networks to complete member profiles →read more on LinkedIn blog
Adapting to Changes in Data
The TensorFlow team published a thoughtful blog post detailing an ML architecture optimized for continuous data changes →read more on TensorFlow blog
🗯 Useful Tweet
We find the best free courses and share them with you
💸 Money in AI
AI-powered
Cloud-based translation technology and language services Smartling raised $160 million in a venture capital round led by Battery Ventures. Hiring globally.
Attack surface management startup CyCognito raised $100 million in a funding round led by The Westly Group. Hiring in Tel Aviv.
Mobility as a service startup Kakao Mobility raised an additional $55 million from strategic investor GS Retail.
Ecommerce platform Convious raised $12 million. Hiring in Amsterdam/Netherlands.
Ecommerce platform Particular Audience raised $7.5 million in Series A funding in a round led by Equity Venture Partners.