December 6, 2019

228 words 2 mins read

Rendezvous with AI

Rendezvous with AI

Ted Dunning offers an overview of the rendezvous architecture, which is geared to deal with much of the complexity involved in deploying models to production, thus allowing more time to be spent thinking and doing real data science. Ted covers the ideas behind the architecture, practical scenarios, and advantages and disadvantages of the architecture.
Talk Title Rendezvous with AI
Speakers Ted Dunning (MapR, now part of HPE)
Conference Strata Data Conference
Conf Tag Making Data Work
Location London, United Kingdom
Date May 22-24, 2018
URL Talk Page
Slides Talk Slides
Video

No matter how clever your learning algorithms, two things will still be true: data and deployment logistics will dominate the effort, and you will need more than two versions of your model, even in full production. However, the rendezvous architecture can help mitigate some of the logistical problems in machine learning. Ted Dunning offers an overview of the rendezvous architecture, which is geared to deal with much of the complexity involved in deploying models to production, thus allowing more time to be spent thinking and doing real data science. The architecture specifically addresses how strict SLAs can be met even with novice models that you haven’t characterized. The result is that you can bound the risk of experimentation with new models. Ted covers the ideas behind the architecture, practical scenarios, and advantages and disadvantages of the architecture.

logistics algorithm complexity data science ai machine learning
comments powered by Disqus
Human in the loop: Bayesian rules enabling explainable AI

Human in the loop: Bayesian rules enabling explainable AI

November 25, 2019

Pramit Choudhary explores the usefulness of a generative approach that applies Bayesian inference to generate human-interpretable decision sets in the form of "if. . .and else" statements. These human interpretable decision lists with high posterior probabilities might be the right way to balance between model interpretability, performance, and computation.
Democratizing deep reinforcement learning

Democratizing deep reinforcement learning

December 3, 2019

Danny Lange offers an overview of deep reinforcement learning, an exciting new chapter in AIs history that is changing the way we develop and test learning algorithms that can later be used in real life.
Practical considerations when shifting to using deep learning for your text data

Practical considerations when shifting to using deep learning for your text data

December 3, 2019

Emmanuel Ameisen and Yan Kou share a guide for moving your company toward deep learning using a collection of NLP best practices gathered from conversations with 75+ teams from Google, Facebook, Amazon, Twitter, Salesforce, Airbnb, Capital One, Bloomberg, and others.
Building a Go AI with Kubernetes and TensorFlow

Building a Go AI with Kubernetes and TensorFlow

November 29, 2019

Reinforcement learning approaches can be massively parallelized, so Kubernetes seems like a natural fit, as Kubernetes is all about reducing the overhead for managing applications. However, it can be …
Building career advisory tools for the tech sector using machine learning

Building career advisory tools for the tech sector using machine learning

November 28, 2019

Dice.com recently released several free career advisory tools for technology professionals, including a salary predictor, a tool that recommends the next skills to learn, and a career path explorer. Simon Hughes and Yuri Bykov offer an overview of the machine learning algorithms behind these tools and the technologies used to build, deploy, and monitor these solutions in production.
Machine-learned model quality monitoring in fast data and streaming applications

Machine-learned model quality monitoring in fast data and streaming applications

November 23, 2019

Most machine learning algorithms are designed to work on stationary data, but real-life streaming data is rarely stationary. Models lose prediction accuracy over time if they are not retrained. Without model quality monitoring, retraining decisions are suboptimal and costly. Emre Velipasaoglu evaluates monitoring methods for applicability in modern fast data and streaming applications.