Open-source Vespa Enterprise Search

Yahoo open-sourced Vespa.ai in September. We've been working with it since, 01 Dec 2017

Vespa is Yahoo’s ‘big data serving engine’ and it was open sourced it in September 2017. We’ve been working on getting it into production deployment for a great start-up called Metsitaba.

Data is the new Oil?

If data is oil then at best it is expensive shale oil and at worst worthless snake oil., 21 Sep 2017

Hype is part of most marketing plans, creating a need for products and services - AI hype likes comparing data with oil. Making comparisons with other valuable resources implies great value in drilling for insights in the data. We would agree data is a valuable resource, but the resource it has most in common with isn’t oil.

Database Architecture for ML

Adapted from a Robert Blumen interview with Dr Michael Stonebreaker, 17 Sep 2017

Many database solutions exist, which one is the ‘best’? Which database vendor/philosophy should you choose? What are the architecture choices we need to understand? This article attempts to explain the underlying mechanisms used in both traditional and alternative technologies.

Building Robust ML Systems

Are Modularity and Hierarchies right for ML design?, 16 Sep 2017

The human cognitive engine is very robust, meanwhile even insect cognition exceeds that of many ML systems. Some ML solutions can now exceed human ability for specific tasks, but are these algorithms are very fragile. How do we bridge this gap?

Gaussian Processes

Part 1 of an introductory series, 16 Sep 2017

Demonstrating a basic Gaussian process fitting of an unknown function, this will be expanded to demonstrate the hyper-parameter search difficulties and issues with cross validation in a temporal setting.