🦾 Serverless ML Execution
Weekly news digest curated by the industry insiders
📝 Editorial
One of the most challenging aspects of modern ML solutions is to match the right infrastructure for executing a given ML model. Some are executed continuously, while others intermittently. Some models experience periods of high demand and traffic followed by idle times. The point is that accommodating a single server infrastructure to a variety of ML models is nothing short of a nightmare. The serverless computing paradigm has evolved over the last few years under the premise of executing code functions without the need to pre-provisioning a server infrastructure. Recently, we have seen several attempts to adapt serverless computing to ML models. Just this week, we saw one of the biggest announcements in this new ML trend.
Amazon SageMaker Serverless Inference was initially announced at the end of 2021 with the premise of deploying ML models for inference without requiring the provisioning of server infrastructure. A few days ago, Amazon announced this platform's general availability, making it one of the first large-scale attempts to integrate serverless computing in the lifecycle of ML models. It is not surprising that Amazon decided to optimize for inference models, given that they account for a large percentage of ML scenarios. Just like traditional serverless computing scenarios, SageMaker Serverless Inference dynamically launches and scales the infrastructure required to execute ML inference models based on their traffic. Serverless Inference is another addition to the robust serving and execution capabilities of the SageMarker platform.
🔺🔻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#185: we overview Centralized vs. Decentralized Distributed Training Architectures; +GPipe, an Architecture for Training Large Scale Neural Networks; +TorchElastic, a Distributed Training Framework for PyTorch
Edge#186: a deep dive into the Evolution of Feature Stores
Now, let’s review the most important developments in the AI industry this week
🔎 ML Research
Language for Object Detection
Google Research published a paper detailing Pix2Seq, a model that can tackle object detection in images as a language problem →read more on Google Research blog
Data Drift on Edge ML Models
Microsoft Research published a paper detailing a continuous learning technique to minimize the impact of data drift in edge ML models →read more on Microsoft Research blog
Learning to Prompt
Google Research published a paper detailing Learning to Prompt, a continual learning technique that addresses the catastrophic forgetting in ML models →read more on Google Research blog
Selecting and Optimizing Objectives
OpenAI published an insightful blog post about the mathematics used to select, evaluate and optimize objectives in complex ML models →read more on OpenAI blog
🛠 Real World ML
SageMaker Server Inference
AWS announced the general availability of SageMaker Serverless Inference which enables the serving of ML inference models as serverless functions →read more on AWS blog
51-Language Dataset
Amazon Research released a massive dataset containing labeled data in 51-languages targeted to advance research in multilanguage models →read more on Amazon Research blog
✏️ A Survey: Data Labeling for ML, part 4
Please take a simple survey to help us work on our articles about data labeling. It will take about 2-3 minutes.
As a thank you, we will send you a cheat sheet with 40+ free ML & data science books and courses! We appreciate your help.
🤖 Cool AI Tech Releases
Meta’s Looper
Meta (Facebook) AI Research (FAIR) unveiled some details about Looper, an API for the optimization and personalization of internal ML models →read more in this blog post from the FAIR team
💸 Money in AI
Automation&ML&AI:
Global intelligent automation company Laiye raised a $160 million Series C funding round led by HOPU Magnolia.
No-code AI platform Mutiny raised a $50 million Series B funding round co-led by Tiger Global and Insight Partners. Hiring in the US.
Enterprise ML platform Jarvis ML raised $16 million in a seed funding round led by Dell Technologies Capital. Hiring in the US and India.
AI-powered:
“Digital twin” service for clinical trials Unlearn.ai raised a $50 million Series B funding round led by Insight Partners. Hiring in San Francisco/US or remote.
AI-powered presentation productivity platform prezent.ai raised a $20 million Series A funding round led by Greycroft. Hiring in the US and India.
Domain and IP intelligence platform alphaMountain.ai raised $2.7 million in seed funding led by Mercato Partners' Prelude Fund. Hiring remote.