Why can't you get a taxi near the Chinese New Year? Probability theory to the rescue!
Big Data Digest Subtitles
Musk has sent his Tesla sports car flying, yet those of you on the ground trying to get a cab may find it increasingly difficult to hail one as the Chinese New Year draws near.
Overtime requires caution
When it comes to taking a taxi, Uber is one of the most popular taxi-hailing apps in the world. Uber must constantly match riders and drivers and get them to their destinations as quickly as possible. Each step of this simple task needs to be optimized, such as deciding on the optimal path, the best time to arrive ......
However, the real world is full of uncertainty. Rain, snow, accidents, many events can occur that can affect the supply and demand of vehicles.
Many current machine learning models are still based on determinism, but real-world data is largely incomplete or imperfect in some sense. This makes probability theory very useful when making forecasts (such as predicting vehicle supply and demand).
Bayesian inference enables us to have a priori assumptions about the real world before making predictions, and to continuously update the model based on observations.
Not long ago, to better adapt to real-world uncertainty, Uber's team open-sourced Pyro, a programming language. Pyro is a probabilistic language, built using Python and PyTorch. With it, we can build a Bayesian deep learning model that is both scalable and very efficient.
There are so many programming languages, why develop one at all?
This is because the language can accommodate uncertainty modeling - for the same input values twice in the model, there may be two different results. It is simply too difficult to use probability in traditional programming languages, and probability can only exist in obscure subroutines. By contrast, Pyro puts probability distributions at the heart of programming. The basic unit of the Pyro program is the random function, which helps us to explicitly compute the output probability given the input.
Will the introduction of Pyro alleviate the problem of difficult taxi rides?
Today, let's take a look at a video by YouTube sensation Siraj Raval: A Primer on Uber Pyro Probabilistic Programming. He will explain the principles and uses of Pyro and will also apply a practical example of weather data. The video comes with the code to follow along and practice!
8 minutes in length
With Chinese subtitles
Click to watch
Pyro is open sourced by Uber's AI lab. The lab comes from an unusual background too, it started as a startup acquired by Uber, Geometric Intelligence, a company that was founded by NYU professor Gary Marcus, though he quit a few months after joining. Yes, the same Gary Marcus who stirred up questions about the future of deep learning and sparked a polemic among Yann LeCun and other AI people from all walks of life. (⊙⊙)
In addition to predicting the demand for rides and vehicle supply in a few hours, Pyro is also used by Uber for financial budgeting, where it can predict the financial figures for the next few weeks. Meanwhile, Uber's other lines of business are also utilizing Pyro, such as Uber Eats using it to predict food preparation and shipping times; it also contributes to the work of the unmanned vehicle division.
Σ (ﾟдﾟlll) So Uber has come up with something this awesome!
Guess who gets to fly first, Uber or Tesla's car? (not counting those taken up by rocket into outer space)
Our journey is to the stars and the sea
This picture is still so lit!
So long, and thanks for all the fish!
Original video by Siraj Raval, translated with permission from Big Data Digest
Link to original video.
https://www.youtube.com/watch?v=ATaMq62fXno
Translation. White the dog, Barbara, singing vegetables
Proofread by. Jiang Chang
Timeline. Jiang Baoshan (1929-), former Guomindang leader
Post. Angie.
Coordination. Long Mu Xue (1980-), Northern Ireland
[Machine Learning Concept of the Day]
Have a Great Defination