Don Beaver
Research scientist and software engineer
Interests
crypto
decentralized protocols | storage | reliability | deniability | proofs | hacks | quantum | coins & chains
machine learning
nlp | autonomous vehicles | vision | prediction | causality | architecture search
compsci
computational complexity | networks | optimization | graph theory | algebra
software
ui | validation and verification | service architectures | observability
assorted
molecular biology | quantum computing
Highlights
Built backends, pipelines, UIs and AIs for Apple News, iBooks, Uber autonomous vehicles, Google infra,
published with a handful of Turing award winners,
invented "Beaver Triples" and the "low degree extension trick"
and correlated-pseudorandomness generators and decentralized random beacons for BFT,
coined "garbled" circuits and the "real vs ideal" security paradigm,
stumbled into an Erdos number of 2.
Software Engineering
Senior Software Engineer, Uber ATG
2015-2019
Full stack engineering for self-driving vehicle user interfaces.
Responsible for calculating, extracting and displaying vehicle intentions (XAI), autonomy states, routes, commercial trip data, trip and route feasibility, dispatch, pickup/dropoff negotiation, mapping and GIS support, QA and triage, security. Operator and Passenger tablet UI, MDM; serverless, observability.
Practicum in computer vision (detection, Tensorflow, CNN, metrics).
Senior Software Engineer, Apple
2010-2015
Apple News 1.0, iBooks Author 1.0, iBooks Multitouch Reader 1.0.
Responsible for multiple features demoed periodically to Steve Jobs, Tim Cook, C-suites.
NLP, named entity recognition and disambiguation, machine learning, topic classification.
Infrastructure pipelines for training, classification, ranking, recommendation engines.
iOS & OS X user interface, multitouch innovation, text annotations and collaboration:
Responsible for novel multitouch book features: interactive image diagrams, embedded full-purpose HTML figures, embedded Keynote presentations.
Co-editor, document annotation standards across ePub and iBooks: ePubCFI.
Co-taught CMU MOOC: Visualization & Insight.
Senior Software Engineer, Google
2005-2009
Dapper co-author (cf Zipkin, Jaeger): network and service analytics for optimization, provisioning and firefighting: petascale data collection, ETL, aggregation pipelines, GFS, MapReduce, BigTable, Sawzall; javascript and GWT dashboards, custom pivot and tabulation libraries, end-to-end network visualizations and interactive breakdowns. Trending, automated critical path analysis and visualization.
Security: implemented security architecture for Google File System; extensive security reviews and consulting.
Research
Chief Cryptographer, Jump Crypto
ZK blockchain bridging, zkSNARK ergonomics.
Staff Research Scientist, Facebook/Novi
Blockchain, ledger, byzantine fault tolerance, decentralized random beacons, zero knowledge STARKs, DeFi, NFT, Move, Solidity.
Senior Research Scientist, Seagate
Self-encrypting storage systems, hardware trusted platform modules. Product: Seagate Momentus.
VP, R&D, CertCo
Cryptographic financial instruments predating modern bitcoin and blockchain, for Deutsche Bank backed startup. Served on public key infrastructure standards bodies.
Member of Technical Staff, IBM/Transarc Corp
Research scientist: web indexing pre-Google, harvesting and organizing secure synopses of web content ("Synopsis File System").
Assistant Professor, Penn State University
Computer science and engineering, tenure track, NSF Career. Funded by NSF, AT&T, IBM. Editorial board for IACR Journal of Cryptology. Board of directors for IACR (International Association for Cryptologic Research).
Member of Technical Staff, Bell Laboratories
Postdoc with adventures in cryptographic protocols. Created "one time tables" / "Beaver Triples." Broke security protocols with a crypto notion of "awareness/transparency"
Education
PhD, Computer Science, Harvard University
Security, Fault-Tolerance and Communication Complexity in Distributed Systems.
Advised by Michael O. Rabin, descended from Church, Poisson, Laplace, Lagrange, Euler.
AB, Biochemical Sciences, Harvard University
Work Products
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.