About me

I do DevOps and Kubernetes platform development work at Statistics Canada; my main project is as a lead on the Advanced Analytics Workspace, a data science platform for internal and external researchers, part of the Data Analytics as a Service project. I co-presented the platform at the CAMDEA Digital Forum.

My home base is in the Data Science Division; we're a machine learning shop who does consulting for clients inside of Statcan and the federal government. I do platform enablement & DevOps, I co-founded the Data Engineering Centre of Expertise and served as its first chair, and I work on operationalizing machine learning processes using cloud native methods and technologies.

I did my master's in ergodic theory and random graphs, but before math I also spent a few years studying linguistics and computer science. I've also been a long time linux user/evangelist.

Birnam, Scotland

Operationalizing Machine Learning

I'm in the operationalization group in the division, and I joke that I study Data Scientists, not Data Science --- I'm interested in the design process in machine learning and the lifecycle of ML applications.

I study the requirements of ML projects, and I work on the infrastructure side to develop the needed tooling, working with Seldon, Workflow engines like Argo Workflows and Pachyderm, Analysis environments like Kubeflow, etc. I also help data science projects adapt to a more MLOps/GitOps model where possible, and I run ahead of the project to ensure that it has a path to production, ensuring that CI/CD, security, architecture, are built into the project early (but not too early), so that projects go smoothly.

My first project at Statcan was a satellite imaging PoC which originally took +128gb of RAM and several hours to process a single image. I took it down to less than six gigabytes of RAM and thirty minutes, turning the project around from prohibitively expensive to almost free, and I built the whole ingestion and preprocessing pipeline on Pachyderm. The end result will be a pipeline that automatically figures out what crops are growing in agricultural regions of Canada from space.

This website lives on Kubernetes, on Digital Ocean, served through a CDN. Is that necessary? No. Is it fun? Yes. A diagram of how it works is here.

Math

My mathematical interests are in ergodic theory of discrete systems and symbolic dynamics. This is basically the analyst's version of discrete-time discrete-space stochastic processes, but in either the measure category or topological category. My research was in random graph theory, I'm also very interested in learning more about random walks, complex systems theory, and social network analysis.

Thesis

My master's thesis was on the Benjamini-Schramm limit (a distributional limit for graphs) of Rauzy graphs (subgraphs of de Bruijn graphs) of low-complexity singly-infinite words (= words of linear factor growth).

The third Rauzy graph of the Thue-Morse word, 0110100110010110...

I showed that when the word has uniform frequencies, the Benjamini-Schramm limit exists, because there is a single shift-invariant measure (the system is uniquely ergodic in this case). I also provided partial results and showed that in the non-uniquely ergodic case that the limit may or may not exist, with examples and counter-examples.

I was supervised by Dr. Vadim Kaimanovich, and I successfully defended my thesis on August 7, 2019.

Problem Frameworks

Aside from my technical work, I'm interested in more interdisciplinary approaches to problem solving and sense-making. I like learning about how mistakes get made (Kahneman, Taleb), why reasonable people disagree and how problems get identified and placed in a framework. My intuition is that this philisophical work is probably much more important in the long run than any technical skill, but it's also likely much harder! As a smart man once said:

If people do not believe that mathematics is simple, it is only because they do not realize how complicated life is.
- John von Neumann

Even if math isn't the be-all end-all, I think that math is good at providing a wealth of metaphors that many people don't have access to.

St. Augustine, convergence and divergence

An excerpt I enjoyed from Hadamard's Essay "The Psychology of Invention in the Mathematical Field"

Indeed, it is obvious that invention or discovery, be it in mathematics or anywhere else, takes place by combining ideas.1

  1. Max Müller observes that the Latin verb "cogito," for "to think," etymologically means "to shake together." St. Augustine had already noticed that and also observed that "intelligo" means to "select among," in a curious connection with what we say in the text.

Organizational Behavior and Management

I am also very interested in organizational behavior. Conway's Law says

Any organization that designs a system (defined broadly) will produce a design whose structure is a copy of the organization's communication structure. - Melvin E. Conway

As a person who cares about software and design, this makes organizational behavior a part of my business. It's always been understood to be a key part of the DevOps pattern, but I think in the context of the federal government, we know software development is harder than it should be, which makes it particularly important to work towards fixing these organizational issues (or at least being aware of them). In this area I've found The Innovator's Dilemma really interesting, in the context of navigating change and adoption of Cloud Native technology and Machine Learning in the federal government. In the Data Science Division, where I play a role in capacity building of engineering expertise, I also found An Elegant Puzzle: Systems of Engineering Management to be a great read on creating engineering teams, fostering talent, and managing scale-up.

Some related things

Publications

[1] Blair R. Drummond, Christian J.G. Tessier, Mathieu F. Dextraze, and Corrie J.B. daCosta. scbursts: An R package for analysis and sorting of single-channel bursts. SoftwareX, 10:100285, 2019. [ DOI | http ]
[2] Blair Drummond. Limits of Rauzy graphs of low-complexity words. Master's thesis, Université d'Ottawa/University of Ottawa, 2019. [ DOI | http ]
[3] Joey Beauvais-Feisthauer, Richard Blute, Ian Dewan, Blair Drummond, and Pierre-Alain Jacqmin. Finiteness spaces, étale groupoids and their convolution algebras. Semigroup Forum, 2020. [ DOI | http ]
[*] Computational poetry research poster from my time as an NLP research assistant in second year of university (2015). Project link.

Hobbies / Fun

Open Source Software

I've also been a hobbyist programmer (and later a professional programmer) who likes linux, Open Source Software, and functional programming (particularly Haskell). I've been using linux for about 10 years. I find Golang and Python (and bash) most useful for me at the moment, but I really enjoy Julia, and I would like to play with Rust sometime. I also want to spend more time learning about gRPC for data apis and eBPF for performance analysis and observability.

Math

Language

I was once an NLP research assistant in the digital humanities, and I still think these projects are very cool.

SC-Bursts

I was the author of an R package written for the dacosta lab for assisting with the analysis of single-ion channel experiments. The library is available on CRAN. I am also the principal author of an open-access paper published on it.

Fun

I also enjoy funk+jazz music (Vulfpeck, Jamiroquai, Azymuth, and Snarky Puppy, Bill Evans, Ahmad Jamal), and cycling.

My music collection as a datasette!

Contact

You can reach me at blairdrummond@protonmail.com

My resume is available here

I also have gitlab, twitter, github, and linkedin.