# AI News, Ask Peter Norvig

When we first began working on Leada, we sought to better understand the data science industry by interviewing professionals in the field.

Some experts took longer to contact than others (I emailed Hal Varian over 8 times) but you would be surprised who you can get 25 minutes of time to help inform others.

As a conclusion to The Data Analytics Handbook, we thought it would be fun to see what questions our readers had about data science, and then had them answered by the #3 ranked most powerful data scientist in the world today by Forbes and Director of Research at Google, Peter Norvig.

So take a University class if that is convenient, or an online class if that makes more sense for you, but then get to work on some real data and figure out how to apply what you learned to a data set, then keep at that.

Obviously there are differences in the size of problems: the number of rows (examples) and columns (features) can vary from dozens to billions, but there are other differences that are less obvious: stationarity (are the examples changing over time), transfer (can we train on one data set and apply what we learned to a different set), sparsity (how many of the possible examples are represented in the data), structure (can the problem be represented as a vector of real numbers, or is some other representation necessary), and so on.

Judea Pearl’s book Probabilistic Reasoning in Intelligent Systems, and before the book his papers and hearing him talk in person, demonstrated to me why I was having so much trouble trying to build systems based on Boolean logic, and showed that probability is the right foundation for dealing with any situation that involves uncertainty.

We want to train some function to set parameters to minimize an expected loss function, and whether the function you are training is called a “neural network” or not just seems like an unimportant detail.

The fact that they are “semi-parametric” – they have a very large number of parameters, but do not rely on keeping all data points around – is certainly important, and I think the semi-parametric space is a very important one.

How Machines Learn

How do all the algorithms around us learn to do their jobs? Bot Wallpapers on Patreon: Discuss this video: ..

11. Introduction to Machine Learning

MIT 6.0002 Introduction to Computational Thinking and Data Science, Fall 2016 View the complete course: Instructor: Eric Grimson ..

Bias-Variance Tradeoff - Breaking down the learning performance into competing quantities. The learning curves. Lecture 8 of 18 of Caltech's Machine Learning ...

Lecture 02 - Is Learning Feasible?

Is Learning Feasible? - Can we generalize from a limited sample to the entire space? Relationship between in-sample and out-of-sample. Lecture 2 of 18 of ...

1. Introduction to Statistics

NOTE: This video was recorded in Fall 2017. The rest of the lectures were recorded in Fall 2016, but video of Lecture 1 was not available. MIT 18.650 Statistics ...

6. Monte Carlo Simulation

MIT 6.0002 Introduction to Computational Thinking and Data Science, Fall 2016 View the complete course: Instructor: John Guttag ..

We're building a dystopia just to make people click on ads | Zeynep Tufekci

We're building an artificial intelligence-powered dystopia, one click at a time, says technosociologist Zeynep Tufecki. In an eye-opening talk, she details how the ...

Controlled Experiments: Crash Course Statistics #9

We may be living IN a simulation (according to Elon Musk and many others), but that doesn't mean we don't need to perform simulations ourselves. Today, we're ...

Big Business: Unlocking Value from Big Data with Analytics

Executives and data scientists from Baidu, LinkedIn, and Foursquare discuss how to generate real value from Big Data, and the importance of business leaders ...

How to speak so that people want to listen | Julian Treasure

Have you ever felt like you're talking, but nobody is listening? Here's Julian Treasure to help you fix that. As the sound expert demonstrates some useful vocal ...