49:58
Data Pipelines for Real-time Fraud Prevention at Scale
Mikhail Kourjanski discusses the architecture of PayPal’s data service which combines a Big Data approach with providing data in real time for decision making in fraud detection.
37:05
pDB: Scalable Prediction Infrastructure with Precision and Provenance
Balaji Rengarajan describes the platform built on the Celect’s pDB framework, providing multiple use cases such as online personalization, document classification, and geospatial anomaly detection.
35:56
Self-racing Using Deep Neural Networks: Lap 2
Jendrik Joerdening, Anthony Navarro discuss how a team of Udacity students used neural networks to teach in two days a car drive by itself around a track.
50:45
The Black Swan of Perfectly Interpretable Models
Mayukh Bhaowal, Leah McGuire discuss how Salesforce Einstein made ML more transparent and less of a black box, and how they managed to drive wider adoption of ML.
57:37
Java at Speed
Gil Tene talks about getting the most of Java applications and understanding some of the optimizations the latest crop of JVMs are able to apply when running on the latest servers.
39:20
Consensus: Why Can't We All Just Agree?
Heidi Howard takes a journey though the history of consensus, and looks ahead to the future of distributed consensus.
47:03
Fast, Flexible and Functional Programming with OCaml
Gemma Gordon and Anil Madhavapeddy give a brief history of OCaml, and explain how they are unlocking its potential in the “new” world of browsers and IoT.
34:46
C++ for Real-Time Communications in the Cloud
Thiya Ramalingam talks about what Zoom’s platform engineers have learned over the years from running a complete C++ stack in their back-end service.
37:05
From Rocks to Rust: Our C to Rust Paradigm Shift
Esther Momcilovic talks about the reasons why Metaswitch chose Rust, and what it’s been like for the development teams getting to grips with this language.
46:01
Rust 2018: An Epochal Release!
Steve Klabnik talks about where Rust is now, what new features are coming down the pipeline, how it's all being managed, and how this affects Rust's development in the future.
32:58
Counting is Hard: Probabilistic Algorithms for View Counting at Reddit
Krishnan Chandra explains the challenges of building a view counting system at scale, and how Reddit used probabilistic counting algorithms to make scaling easier.
46:19
Developing Data and ML Pipelines at Stitch Fix
Jeff Magnusson discusses thoughts and guidelines on how Stitch Fix develops, schedules, and maintains their data and ML pipelines.
