🥗 Will Machine Learning Data Infrastructures Become Commoditized?

The Scope covers the most relevant ML papers, real-world ML use cases, cool tech releases, and $ in AI. Weekly

📝 Editorial 

The value capture in software technology trends fluctuates between infrastructure and applications. Some market cycles are dominated by momentum in infrastructure companies. Then the pendulum swings gradually towards applications. This cycle repeats constantly. However, after long market cycles, many infrastructure building blocks become commoditized. Think about the trajectory of storage and compute infrastructure of platforms like AWS and Azure; or database infrastructure of platforms like Oracle. Following that thesis, we should expect many of the infrastructure building blocks of machine learning (ML) to become increasingly commoditized. But will they?  

ML has proven to challenge many traditional conventions in traditional software markets. Infrastructure commoditization might not be an exception. The current ML market cycle is, without a doubt, dominated by infrastructure companies, but the value capture seems to be increasing rather than decreasing. Nothing should be more a commodity in ML than storage and compute infrastructure. And yet, this week, Databricks announced a monster funding round that values the company at $38 billion, which seems inconceivable for a pure data-compute infrastructure platform. That valuation is about nine times the market capitalization of big data pioneer Cloudera. Given ML’s tight dependency on data and compute, it seems likely that the value capture dynamics of this new market are different from other technology trends. From that perspective, the idea that ML infrastructure will get commoditized is not trivial at all. Certainly, the best days of ML infrastructure seem to be ahead of us.  

Share


🔺🔻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#121: we discuss transformers and time-series; we explore Google Research’s paper about Temporal Fusion Transformers; we overview GluonTS.

Edge#122: in this ‘What’s New in AI’ edition, we deep dive into Unified VLP, a transformer model for visual question answering.


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

🔎 ML Research

Self-Supervised Learning for Anomaly Detection 

Google Research published a paper proposing a self-supervised learning method for anomaly detection in classification problems ->read more on Google Research blog

Dataset for 3D Object Reconstruction 

Facebook AI Research published a research paper and open-source version of Common Objects in 3D (CO3D), a dataset to train models in 3D object reconstruction problems ->read more on FAIR blog

Fast Reinforcement Learning 

Salesforce Research published a research paper and open-source version of WarpDrive, a framework for fast performance multi-agent reinforcement learning models ->read more on Salesforce Research blog

Computational Graph in PyTorch  

The PyTorch team published a very insightful blog post explaining the internal mechanisms used to build computational graphs in the deep learning framework ->read more on PyTorch blog


🛠 Real World ML

Detecting Abusive Activity at LinkedIn 

The LinkedIn engineering team detailed the deep learning techniques used to detect abusive user activity ->read more on LinkedIn blog

Real-Time Streaming Analytics at Uber 

The Uber engineering team explained the architecture used to power real-time streaming analytics at the transportation giant ->read more on Uber Engineering blog


🤖 Cool AI Tech Releases

Apache Drill v1.19 

Apache Drill just reached a new version with enhanced SQL querying capabilities ->read more in their press release

Run:AI’s ResearcherUI and MLOps Support 

ML compute management platform Run:AI announced a new release that includes integrations with several MLOps stacks such as KubeFlow, Airflow, and MLFlow ->read more on Run:AI blog


🗯 Follow us on Twitter

and win one of three Premium subscriptions to TheSequence. Check the terms in the post ->


💸 Money in AI

ML&AI

AI-powered:

  • Virtual insurance agent Insurify raised $100 million in a Series B funding round led by Motive Partners. Hiring in Sofia, Bulgaria/Cambridge, US.

  • Decision intelligence platform Peak raised $75 million series C led by SoftBank Vision Fund 2. Hiring in New York/UK/India.

  • SEO intelligence tool Botify raised $55 million in a Series C funding round led by InfraVia Growth. Hiring in Paris/New York/London/Tokyo/Sydney.

  • SaaS voice automation company Skit raised a $23 million Series B round led by WestBridge Capital. Hiring in India.

  • Ad fraud protection, privacy, and compliance analytics platform Pixalate raised $18.1 million in growth capital led by Western Technology Investment and Javelin Venture Partners.

  • Sales experience platform Walnut raised $15 million in Series A funding, led by Eight Roads Ventures. Hiring in Tel Aviv/US/Remote.

Acquisitions:

  • Conversational analytics startup ScopeAI was acquired by the contact center software Observe.AI. The companies didn’t disclose the purchase price.