🤜🤛 AI/ML startups align to build a canonical stack and compete with the incumbents

📝 Editorial 

In December 2020 we wrote about the frenzy of AI acquisitions by the large tech firms that made it a bit difficult for AI startups to achieve maturity. That situation remains a long-term challenge for the evolution of the AI/ML market. But not only that. The current AI/ML startup landscape is really fragmented, looking more like a set of scattered puzzles that are not necessarily combinable. Recently, TheSequence joined the AI Infrastructure Alliance (AIIA) whose mission is to align AI/ML startups and community members to make it more like Lego blocks that can be stacked together. The Lego metaphor is the same analogy used by Dan Jeffries, the managing director of the AIIA and Chief Technical Evangelist at Pachyderm. One of the missions of the AIIA is to define and frame the key components of the AI/ML canonical stack (like the LAMP stack for software development if you know what we mean ;).

There are almost 50 members in the AIIA now. Some of them are direct competitors. Wait, isn’t it controversial to unite with the competitors? Despite the innovation delivered by startups, the AI/ML market remains dominated by the tech giants, such as Amazon, Google, Facebook etc. They have access to enormously big data, human resources, money, built tools and infrastructures – all that makes it brutally hard to compete with them. Equalizing the competitive landscape is another goal of the AIIA. As H.O. Maycotte, Molecula’s CEO puts it: “AI and ML have held so much promise for the last several decades, but successful AI has remained exclusive to the tech giants and therefore has remained impractical.” Joining the AIIA, Mr. Maycotte hopes to work with other members to create practical, operational end-to-end AI/ML stacks that can bring the promise of AI to life and create step-function improvements for the world.  

For Mariya Davydova, Head of Product Neu.ro, the AIIA also means building a canonical ML stack, and through “collaboration and competition, creating a clear path to interoperability for the universe of open-source and proprietary ML tools.” Diego Oppenheimer, Algorithmia’s CEO also mentions that it is important for them to “partner with other vendors to ensure an interoperable AI stack.” For David Aronchick, co-founder of Kubeflow and the SAME Project, the formalization of the stacks of ML tools “enables both reproducibility and avoidance of vendor lock-in.” Alex Shee, BD at Sama, pointed to the importance of a shared vision for an ethical AI infrastructure, “ensuring that we are building fair and trustworthy algorithmic tools that represent society at large and empower people through AI.” Frans van Dunné, Chief Data Officer at ixpantia, was the most poetical: “Being part of the AIIA allows us to be close to the fire where MLOps best practices are forged.” 

Forging the best practices together, the members create the important standards, help democratize the AI/ML industry and move it forward.

*All mentioned companies are members of the AIIA. This is just a small part of the quotes we’ve received. We put the rest of them in the comment section. Please go and check these quotes, these are the voices of the MLOps community united by AIIA.

🔺🔻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#87: the concept of model-based reinforcement learning; how Google Dreamer uses model-based reinforcement learning to learn long-horizon tasksUber Fiber, a distributed computing framework optimized for RL agents.  

Edge#88: deep dive into IBM’s use case with Snorkel Flow.

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

🔎 ML Research

Deep RLSP – Better Reinforcement Learning Policies by Simulating the Past 

Berkeley AI Research (BAIR) published the evaluation of Deep RLSP, a reinforcement learning model that can learn a policy without human supervision, by simulating past states in a given environment ->read more on BAIR blog

SCoRe – task-oriented conversational system    

Built upon Microsoft Research’s previous work in RAT-SQL and StruG, SCoRe’s pretraining methodology learns language representations to achieve state-of-the-art performance in different task-oriented language benchmarks ->read the overview of RAT, StruG and SCorE on Microsoft blog

FELIX – Flexible Text Editing Through Tagging and Insertion) 

In their new paper, Google AI introduces a text-editing system that is both fast and flexible compared to seq2seq approaches (a 90x speed-up) and previous text-editing systems ->read more on Google AI blog

🤖 Cool AI Tech Releases

IBM Quantum Experience

A fascinating reminiscence by IBM about how, five years ago, they put their first 5-qubit quantum system on the cloud, so that anyone could run their own quantum computing experiments, and how it changed the whole field ->read more on IBM blog

Google Chat Assistant  

Google Cloud released for public preview an improvement to their Contact Center AI (CCAI). The new Agent Assist for Chat provides two key features to make the conversations better: Smart Reply provides response suggestions, and Knowledge Assist makes suggestions about useful articles and FAQs from your knowledge base ->read the details on Google blog


Microsoft open-sourced Counterfeit, an automation tool to conduct security risk assessments of AI systems and help prevent hacking ->read more on Microsoft blog

💬 Useful Tweet

It means more options to infuse core-level innovations to improve capabilities for leading-edge workloads like AI and cloud computing.

💸 Money in AI

  • Cyber protection startup Acronis raised a humongous $250 million funding round led by CVC Capital Partners VII. With AI-based antimalware and blockchain-based data authentication technologies, Acronis unifies data protection and cybersecurity to deliver integrated support, solving the safety, accessibility, privacy, authenticity, and security challenges of the modern digital world.

  • AI-based insurance SaaS provider Shift Technology raised $220 million in a Series D investment round led by Advent International. The company creates a wide range of products that apply AI and advanced data science to key insurance processes.

  • Photonic computing startup Lightmatter raised $80 million in a Series B round led by Viking Global Investors. The company created ultra-fast photonic chips specialized for AI work. It promises higher performance with lower environmental impact of AI compute solutions.

  • Time-series database maker Timescale raised a $40 million Series B led by Redpoint Ventures. Its core product is TimescaleDB, an open-source and free relational database for time‑series, optimized for advanced analytics and built for production use cases.

  • Predictive analytics startup Pecan.ai raised $35 million in a Series B round led by GGV Capital. First, their AutoML platform automates data preparation, feature engineering and selection, and then uses AI-based predictive analytics algorithms to transform the results into actionable predictions.

  • Cybersecurity management automation startup JupiterOne raised a $30 million Series B round led by Sapphire Ventures. JupiterOne makes security teams more efficient by centralizing the data from dozens of cloud services into a single hub for management, analysis, and alerts.

  • AI-powered market intelligence company Crayon raised $22 million in a Series B round led by Baird Capital. It enables businesses to capture, analyze, and act on market movements from their competitors.