Friday Hacks #207: Bayesian Networks and Packet-Processing Systems
Posted on by Zhang Ziqing
Date/Time: Friday, September 03 at 7:00pm
Venue: Online on Zoom
Zoom Link: https://nus-sg.zoom.us/j/83651946480?pwd=dG84TWFOQUltWjE4bUlWaWF1bEcvdz09
Bayesian Networks - Thinking about causality in ML
Consider a client who wants to make a decision, using historical data. In this type of analysis the commonly known fact that “correlation does not imply causation” comes to life and it is crucial to distinguish between events that cause existing inefficiencies or problems and those that merely correlate.
Causal inference aims to determine which available controls drive specific outcomes. This is a distinctly more demanding condition than learning the correlation & is fundamental to many of the projects we run at McKinsey. Many machine learning approaches disregard causal inference, despite a wide range of approaches to causal inference have been proposed in the literature.
This tutorial will discuss the importance of causal models, as well as some of the most state of the art methods for reasoning, including CausalNex – QuantumBlack’s open-source Bayesian Network tool, that we use to deliver highly interpretable models and powerful counterfactual analysis: [https://causalnex.readthedocs.io][https://causalnex.readthedocs.io]
The talk will be a mixture of lecture and discussion on CausalNex, where we will explore:
Motivation & examples applied to client projects
- Introduction to Bayesian Network theory
- Structure learning
- Probability fitting
Paul Beaumont is a Principal Data Scientist in the Singapore office of QuantumBlack. He works on statistical models for explanatory, predictive and prescriptive problems.
His role involves leading analytics teams of data engineers, scientists & translators, designing mathematical models to help clients understand pertinent questions about their data, as well as analyzing results from modelling to delivering insights.
Paul is also the co-lead for CausalNex – QuantumBlack’s open-source Bayesian Network tool, that we use to deliver highly interpretable models and powerful counterfactual analysis.
Safe At Any Speed
Building a performant, safe, maintainable packet processor A fast, safe and readable packet-processing system – How can we get all three?
At Jane Street, we’ve been building systems to trade electronically for over a decade. As technology advances and the scale of the markets grows, we need our systems to be able to process ever growing amounts of data in ever shorter time windows. In this talk, we’re going to explore how to build a highly optimized single-core packet processing system that is capable of processing millions of messages per second. We will see how to bridge the gap between the high-level abstractions we’ve come to love when structuring code, and efficient machine-level execution necessary to process messages at line-rate.
Malcolm Karutz earned his BS and MS in Computer Science from the University of Melbourne, Australia. Since 2019 he has been a software developer at Jane Street in Hong Kong where he works on frameworks, tools, and infrastructure for developing, running, and monitoring high-performance trading systems.