🔷 End-to-End vs. Best-Of-Breed Machine Learning Platforms~
Weekly newsletter that discusses impactful ML research papers, cool tech releases, the money in AI, and real-life implementations
📝 Editorial
Which platform to choose for machine learning development is one of the questions that torments organizations embarking in the space. The rapid growth of the machine learning market has translated into an explosion of startups dedicated to machine learning development. At the same token, we have cloud platform giants such as Microsoft, AWS and Google providing very complete platforms that cover almost every aspect of a machine learning pipeline. As a result, it’s becoming increasingly hard to determine which platforms to use for a given machine learning problem. Should you go with end-to-end platforms like Azure ML or AWS SageMaker, or bank on innovative startups that are focusing on specific capabilities of a machine learning pipeline?
The lifecycle of machine learning solutions is very complex and, consequently, very difficult for a single platform to deliver quality value on all of its stages. At the same time, the current state of the machine learning market makes it really hard to determine which categories will remain as standalone submarkets. Data labeling, continuous deployment, model optimization are some of the areas that have the potential of creating a new generation of standalone platforms. However, companies like Microsoft, Amazon and Google are becoming highly acquisitive and rapidly building some of those capabilities into their platforms. When it comes to machine learning, deciding between end-to-end or best-of-breed platforms is far from trivial.
We would love to hear from you about this debate. When the time comes to select a machine learning platform, do you favor the consistency of end-to-end machine learning platforms like Azure ML or SageMaker? Or the fast innovation of best-of-breed startups? What do you base your choice on?
🔺🔻TheSequence Scope – our Sunday edition with the industry’s development overview – is free. To receive high-quality educational content every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻
🗓 Next week in TheSequence Edge:
Edge#19: the concept of graph neural networks; DeepMind research about the usage of graph neural networks to enable combinatorial reasoning; the review of the deep graph library open-source framework.
Edge#20: the concept of MLOps; the review of Google’s TFX paper; the overview of some of the top technologies in the MLOps space.
Now, to the most important developments in the AI industry this week
🔎 ML Research
Reasoning About Abstract Concepts
Researchers from MIT published a paper proposing a model that can identify abstract concepts in videos ->read more on MIT News
Quantum Chemistry Simulations
Researchers from the Google AI Quantum published a paper exploring ideas to conduct large chemical simulations in quantum computers ->read more on Google Research blog
Traffic Predictions with Graph Neural Networks
DeepMind and Google unveiled research that improves the accuracy of Google Maps real-time ETAs using graph neural networks ->read more on DeepMind blog
DeepFakes’ heartbeats
Binghamton University and Intel researchers proposed a deepfake source detector that predicts the source generative model for any given video by analyzing unique biological and generative noise signals ->read more in the original paper
🤖 Cool AI Tech Releases
Opacus
Facebook open-sourced Opacus, a framework for differential privacy in PyTorch models ->read more on Facebook blog
PSI
Microsoft open-sourced the platform for situated intelligence (PSI), a framework for research and implementation of AI models that use heterogeneous input data streams ->read more on Microsoft Research GitHub
AllenAct
Researchers from the Allen Institute for AI open-sourced AllenAct, a research framework for embodied AI ->read more on AllenAct GitHub
💬 Useful Tweet
The method that seamlessly removes objects, watermarks, or expands field-of-view from casually captured videos. This looks amazing!
💸 Money in AI
Mustard, an app that analyzes an athlete’s mechanics and offers corrective tips to help them improve their techniques, has raised $1.7 million to improve its tool. From every sport possible, they started from baseball.
Defense Innovation Unit (DIU), a Pentagon organization that brings consumer technology into the military, approved Google Cloud for the contract to supply Veterans Affairs hospitals and Defense Health Agency treatment facilities with AI for predictive cancer and disease diagnosis. The financial part of the contract is not disclosed.
Process automation startup Hypatos raised almost $12 million in seed funding. It uses deep learning, NLP and computer vision for a wider range of back-office automation.
AI health IT company Biofourmis raised $100 million in a Series C funding round. They use data from patient health histories and wearables to create biomarkers that reflect overall health as well as show signs of specific diseases much earlier than traditional screenings and tests. The team describes its end-to-end Biovitals platform as a “physiology-based data analytics engine”.
Analytic software startup Legion just raised $22 million to develop retail scheduling and demand forecasting. It learns behavioral patterns and projects them to the future, considering things like local events, weather, seasonality, and customer-specific promotions.
Location data analytics provider Cosmose AI today raised $15 million in a funding round. The team works on creating a “holistic view” of in-store shoppers. They claim to be able to understand, predict and influence how one billion people shop offline, pinpointing customers’ locations down to store aisles (about two meters) with 73% accuracy.
Exoskeleton company Sarcos Robotics raised $40 million in Series C. The team produces robots that augment humans to enhance productivity and safety. One of Sarcos’ ideas is to have a subscription-based, multi-year “robot-as-a-service” model, which would include maintenance, support, and upgrades of its robots. Another plan is to launch an AI platform through which the wearers will be able to “augment” their suits.
TheSequence is a summary of groundbreaking ML research papers, engaging explanations of ML concepts, exploration of new ML frameworks, and platforms. It also keeps you up to date with the news, trends, and technology developments in the AI field.
5 minutes of your time, 3 times a week– with TheSequence you will steadily become knowledgeable about everything happening in the AI space.
I'd suggest giving this a look at: https://docs.neuro-ai.co.uk it's a pay for compute time API. It wraps AI frameworks instead of hosting. Pretty new so a lot of support needs to be added, but good for the basics.
If you are into ML research on-premise you the best-of-bread approach is better to sell internally. That's basically happens in my team rn.
Personally I just started a new web app pet project. There is no way I will go down the rabbithole of hacking a backend API together myself. So went with Firebase + GCP APIs (I didn't choose Amplify bcoz AWS UI sucks and I don't care about IT salesman arguments)