Don Beaver

One Time Tables - "Beaver Triples"

Fast distributed security computations

Secure multiparty protocols are a way for collaborators to calculate a function without revealing inputs and without having any trusted party available. Think: secret ballots. Or: peer-to-peer Tinder.

"Beaver Triples" make the calculations fast and elegant - a computational generalization of one-time pads, and almost as simple.

They're employed in privacy-preserving distributed machine learning software at Facebook and Google to train on input data (or make inferences) without revealing it.

No, I didn't call them "Beaver Triples," but I'll happily brag about 15 minutes of eponymy.  My marketing attempt called these "One-Time Tables."

Apple News Classifiers and Recommenders

NLP, NER, recommendations

Designed and wrote classifiers, entity recognizers and extractors, pipelines, and recommendation systems for Apple News 1.0. Fascinating tricks with wikipedia, graph theory, and ranking algorithms.

Apple iBooks Multitouch

UI, UX and document models

Do you read iBooks on your iPhone? Then you're using my code! Steve Jobs showcased it, too.

Highlighting, bookmarks, navigation and navigation history, in particular, and several interactive textbook illustrations. In addition to implementing beautiful visual design specs from the Apple Design team, we designed interoperable and standardized deep linking, robust to editorial changes in book versions.

Putting textbook creation in the hands of anyone, circumventing school board hegemonies and college textbook profiteering, was SJ's last great vision and mission.

Crypto as a Service

Edge computing for cryptography

In content distribution, in machine learning, you want to enable calculations close to the data, in simplified devices. So, too, for crypto.

In a tradition of unmarketable names, this crypto approach started as "Commodity-Based Cryptography" (FOCS paper), in which cryptographic resources (secure randomness, secure multiparty computation enablers) are disseminated by one or more mutually-distrusting cloud services. Highly streamlined yet weak edge devices consume these high protein products.

It's what you get when you leave academia for the business world and get some new perspectives.

Google Dapper

Observability in service architectures - with tensors [tm]

So you have an enormous collection of microservices making RPC calls to each other but no idea when, where and how much?  Not sure if they are configured properly?

Enter Dapper - a Google infrastructure observability project. I wrote mapreduce aggregation frameworks, UI dashboards and report generators. Got insights into network latency issues that were otherwise inaccessible.

Twitter copied it openly as Zipkin, and Uber took Zipkin forward as Jaeger.

Oh and. In 2009, I was annoying lots of people in my Google software engineering circles by repeatedly explaining the aggregations as cascading a bunch of - yes - tensors.