Auto-Differentiation: At the Intersection of Nifty and Obvious

Channel:
United States Google TechTalks
Subscribers:
348,000
Published on ● Video Link: https://www.youtube.com/watch?v=Fntkxhcs3mU


Duration: 47:03
985 views
23

A Google TechTalk, 2021/1/29 , presented by Alan Christopher
ABSTRACT: Automatic differentiation, or autodiff, is a technique for computing derivatives of numerical computations exactly (up to machine precision). This talk covers the mathematical basis of autodiff, describes the forward and backward mode algorithms, and briefly touches on the practical value of the technique.

https://github.com/alan-christopher/autodiff-edu

About the speaker: Alan is a former Google SRE. This presentation shares his recent exploration of autodiff.



Other Videos By Google TechTalks


2021-04-212021 Blockly Developer Summit: Operator Precedence
2021-04-212021 Blockly Developer Summit: Blockly on npm
2021-04-212021 Blockly Developer Summit: Blockly Roadmap
2021-04-212021 Blockly Developer Summit: Blockly Codelabs
2021-04-212021 Blockly Developer Summit: Toolbox APIs
2021-04-212021 Blockly Developer Summit: Developer Tools
2021-04-212021 Blockly Developer Summit: How to Build a Plugin
2021-04-212021 Blockly Developer Summit: Plugins Overview
2021-04-212021 Blockly Developer Summit: Blockly Year in Review
2021-03-15Managing Digital Accessibility at Universities During the COVID 19
2021-03-15Auto-Differentiation: At the Intersection of Nifty and Obvious
2021-03-15Beyond Submodular Maximization via One-Sided Smoothness and Meta-Submodularity
2021-02-16Statistically Valid Inferences from Privacy Protected Data
2021-02-16Two Case Studies in Private Data Analysis
2021-02-16Uncertainty Estimation via (Multi) Calibration
2021-02-16(Nearly) Optimal Algorithm for Private Online Learning
2021-02-16DP-SGD Privacy Analysis is Tight!
2021-02-16Efficient Alternatives to Min-Max Models
2021-02-16When is Memorization of Entire Examples Necessary for High-Accuracy Learning?
2020-09-28Pathways to Innovation: A Fireside Chat with Google Research
2020-09-12Dominant Truthful Multi-task Peer Prediction Mechanism with a Constant Number of Tasks