Gradient Flow #30: Pricing Data Products, National AI Strategy, Elastic Computing

Subscribe • Previous Issues

This edition has 560 words which will take you about 3 minutes to read.

“We hoped for the best, but it turned out as usual.” – Viktor Chernomyrdin

Data Exchange podcast

  • Challenges, Opportunities, and Trends in EdTech   Stanford’s Sharon Zhou has been teaching very popular courses on GANs (generative adversarial networks) on Coursera. We discuss the state of online learning platforms, and novel applications of GANs.
  • Pricing Data Products  Surveys show that many data science and machine learning teams cite lack of data as a key challenge. The rise of new data exchanges and markets confirm this point. Jian Pei of Simon Fraser University recently wrote a comprehensive survey paper on how organizations assess the value of data objectively, systematically and quantitatively. 
[Image: Grand Mosque in Abu Dhabi, by Ben Lorica]

Featured Virtual Event

Healthcare NLP Summit is a FREE conference featuring speakers from Carnegie Mellon, Stanford, UC Berkeley, Microsoft Research, Google Brain, MIT-IBM Watson Lab, Curai, and more. You need not be in the Healthcare/Biotech/Pharmaceutical sectors to benefit from the presentations. As the external program chair of this conference, I think anyone interested in building language applications should attend this event.

Register Now

Data & Machine Learning tools and infrastructure

  • Final Report of the U.S. National Security Commission on Artificial Intelligence   Beyond the expected public policy and national defense recommendations, this essential read covers topics such as data infrastructure, information security, hardware, adversarial ML, supply-chains, robustness and resilience, and much more. I found the best way to consume this report is by perusing the easy-to-navigate website.
  • Elastic Deep Learning: Introducing Horovod on Ray  A popular distributed training framework (Horovod) can now be operationalized using the most popular framework for multi-cloud distributed computing (Ray). Key quote: “We believe that Ray will continue to play an increasingly important role in bringing much needed common infrastructure and standardization to the production machine learning ecosystem, both within Uber and the industry at large.”
  • Contextual calibration improves GPT-3 accuracy across many prompt format choices and examples  The goal of few-shot learning tools like GPT-3 is to enable developers to rapidly prototype NLP models and applications. A group of academic researchers are building tools that “improves accuracy, reduces variance, and … makes tools like GPT-3 more effective for end users”. (code)   
  • Meltano   Software engineering tools and processes for data pipelines and ELT.
  • The Netflix Cosmos Platform   Used by the streaming giant for building resource-sensitive applications, Cosmos combines several tools and techniques into this powerful paradigm: “microservices that trigger workflows that orchestrate serverless functions”.

Funding Updates

[Image: Dublin, Ireland, by Lucian Potlog from Pexels]


Closing Short: A classic song in a rustic setting.


If you enjoyed this newsletter please support our work by encouraging your friends and colleagues to subscribe:

%d bloggers like this: