Ian Goodfellow, Google – Practical Methodology for Deploying Machine Learning #AIWTB Oct 2015

Ian Goodfellow, Google Research Scientist’s talk for the 1st edition of AI With The Best online conference Oct 2015 on Practical Methodology for Deploying Machine Learning.
To obtain good performance from a machine learning application, it is important to choose the right machine learning model, algorithm, parameters, and training set. In this talk, Ian describes:
1) how to set concrete goals for your machine learning system based on business objectives,
2) how to rapidly establish an end-to-end system that performs the task you desire,
3) how to use simple experiments to guide your modifications of this system until it meets your goals.

Ian’s talk ends with a brilliant Q&A
30:34 “Is there any interactive tool to choose the best AI algorithm?”
31:38 “How to reject non-positive data?”
32:39 “Is there a limit for the size of data set that different deep learning algorithms can handle?”
33:39 “Are neural networks still better for unstructured data?”
34:00 “Do you know of any library that can be used on Android?”
34:14 “Have you seen the tool KXEN that suggests the model based on the data you have?”
34:29 “What is an average day for Google like for you?”
35:25 “Any advice for deploying ML as a service?”
35:44 “Are CNNs limited in the number of classes?”
36:38 “Is there an LSTM RN algorithm for regression, not classification, prediction of numerical sequences?”
37:44 “Have you played with WetLabs AI that can find the best parameters like learning rates etc?”
39:39 “Do you think that Generative Adversarial Nets have a brighter future than adversarial inference?”


One comment

  1. Dusty Poster says:

    super straight-forward and practical guide. it’s hard to find this kind of
    simple but useful advice. thanks for sharing your wisdom!

Comments are closed.