Skip to main content
SAGE
Search form
  • 00:00

    [MUSIC PLAYING][Deep Learning in Python--Introduction]

  • 00:09

    SOUBHIK BARARI: Hello, and welcome to this course.[Soubhik Barari, PhD Student in Political Science, IQSS,Harvard University] I'm your course instructor,Soubhik Barari.In this first video, I'm going to introduce youto one of the most powerful computational methodsfor pattern recognition that exists today,deep learning, and talk about why you should use itfor social science research and how this course willhelp you get started.

  • 00:31

    SOUBHIK BARARI [continued]: A little bit about me.I'm a data scientist and a PhD student in political scienceat Harvard University.As a data scientist, I especiallyfocus on leveraging techniques in computer science,such as deep learning, to better quantify and analyzecomplex phenomena in social science.If that sounds interesting to you,

  • 00:51

    SOUBHIK BARARI [continued]: I think you'll definitely enjoy this course.Before I explain what deep learning is,first, I want to talk about learning in general.The human brain on its own is really goodat learning abstract patterns and complex informationit finds in the world, such as sounds, images, and language,in both the written and spoken form.

  • 01:12

    SOUBHIK BARARI [continued]: By abstract patterns, I mean patternsthat are somewhat subjective in natureand that hierarchically arise from complex relationshipsbetween smaller patterns, which are themselves made upof even smaller patterns.For example, recognizing whether a face is happy.The brain can recognize a happy faceby first recognizing a smile, and it

  • 01:34

    SOUBHIK BARARI [continued]: can distinguish that it's a real smile and nota fake smile, maybe by the specific upward foldof the mouth and perhaps, other subtle facial cues.Another hint might be a look in the eyes, a look thatarises from maybe a combination of pupildilation, a gaze in a certain direction, and/or posture.In short, the brain recognizes an abstract idea, a happy face,

  • 01:59

    SOUBHIK BARARI [continued]: by identifying the presence and combinationof other abstract patterns.This is what makes humans really good at social scienceresearch, which is largely about discovering and explainingpatterns about human behavior.However, today, there is so much informationto collect about humans.We just don't have the time, energy, or the resources

  • 02:21

    SOUBHIK BARARI [continued]: to do all parts of the discovery entirely ourselves.Luckily, today's machine learning algorithmsare really good at efficiently recognizingstatistical patterns in huge amounts of data.But for the most part, it can only do so in very concreteand well-defined data--data that can be easily represented in vector or matrix

  • 02:44

    SOUBHIK BARARI [continued]: form and that has already been touched upand preprocessed by humans.For example, machine learning can accuratelyidentify topics in text documents, like newspapers,but only after they've been annotated for keywordsand cleaned of low information stop words.These kinds of data usually can't just

  • 03:06

    SOUBHIK BARARI [continued]: be used for machine learning in their original, highdimensional form.For this reason, a task like recognizinga strong emotional response in a human facecould be easily done by a human being,but probably can't be done by a machine learning algorithmunless there are lots of helpful hints about the face that

  • 03:27

    SOUBHIK BARARI [continued]: have been annotated by a human already.Thus, a task that both of these learning engines would fail atis detecting abstract patterns, such as facial emotion,in very large amounts of data.The respective limitations, unfortunately,render them as weak tools in getting the job done well.

  • 03:47

    SOUBHIK BARARI [continued]: This is a shame, since we're livingin the age of big data, where we could reallymake huge social scientific discoveries if we could onlyextract such abstract patterns.This is where deep learning comes in.Deep learning is a specific type of machine learningthat mimics how the human brain worksusing an artificial neural network model.

  • 04:09

    SOUBHIK BARARI [continued]: Just like the brain, such algorithmscan recognize abstract patterns by finding correlated lowerdimensional subpatterns in the original high dimensional data.It is a powerful example of artificial intelligence.Because of this, deep learning is a perfect toolto analyze large scale, complex datasets that are too big for individual humans to parse

  • 04:32

    SOUBHIK BARARI [continued]: and that are not clean or concreteenough for traditional machine learning algorithms.There are many really cool active applicationsof deep learning, specifically in social science,that demonstrate its capabilities, includingpredicting levels of poverty and economic developmentby detecting road networks and satellite data,

  • 04:53

    SOUBHIK BARARI [continued]: analyzing how politicians visually communicatethrough press releases, accurately forecastingthe fine grained movements of energy and financial markets,detecting emotional and political speech,studying the movements of tagged individuals in video,and much, much more.In this course, we are on our programdeep learning models using the Keras package in Python.

  • 05:16

    SOUBHIK BARARI [continued]: As such, I assume that you likelyhave intermediate level experiencein Python programming and beginnerto intermediate level experience with machine learning.Some exposure to high school level calculusmight be helpful.Don't worry if you don't have all of these things--you will still be able to follow along.However, you might want to take each video a little bit more

  • 05:36

    SOUBHIK BARARI [continued]: slowly in order to fully understand what's going on.The course is structured as follows.First, I'm going to show you how to set up your computerto do deep learning.Next, I'm going to give you an overview of deep learningmodels, along with practical applications.We'll learn how to use neural networkmodels to predict useful things on real world data.

  • 05:57

    SOUBHIK BARARI [continued]: Finally, we'll touch on some advanced topics,such as how to tune models for different applicationsand how to pick different deep learning libraries to use.All of the code for this course canbe downloaded from a GitHub repositoryat the link shown here.My main goal for this course is to not just convince youthat deep learning is a really cool method in computer

  • 06:18

    SOUBHIK BARARI [continued]: science, but that you can use it for really amazing applicationsin areas that you care about.To this end, it is important that youcan write functional code in orderto get your algorithms to work.However, it is not at all necessarythat you memorize how to call certain functionsor create certain objects.Similarly, while it is important to understandthe mathematical underpinnings of deep learning in order

  • 06:40

    SOUBHIK BARARI [continued]: to become an advanced or expert level user,it is also not necessary to get started.In short, by the end of this course,I want you to generally understandhow deep learning works and be comfortablewriting the foundational code neededfor practical applications in social science.So without further ado, let's get started.

  • 07:01

    SOUBHIK BARARI [continued]: [https://github. com/soukhikbarari/SAGE-Deep-Learning][MUSIC PLAYING]

Video Info

Series Name: Deep Learning in Python

Episode: 1

Publisher: SAGE Publications Ltd

Publication Year: 2019

Video Type:Tutorial

Methods: Deep learning

Keywords: abstract reasoning; algorithms; artificial intelligence; computer science; human brain; large-scale research; learning processes; neural networks; pattern analysis; programming and scripting languages; recognition; Social science research ... Show More

Segment Info

Segment Num.: 1

Persons Discussed:

Events Discussed:

Keywords:

Abstract

In this introductory module to Deep Learning in Python, Soubhik Barari, PhD student in Political Science, IQSS, at Harvard University, introduces the concepts of human learning, machine learning, deep learning, and provides a course outline.

Looks like you do not have access to this content.

Deep Learning in Python: Introduction to Deep Learning

In this introductory module to Deep Learning in Python, Soubhik Barari, PhD student in Political Science, IQSS, at Harvard University, introduces the concepts of human learning, machine learning, deep learning, and provides a course outline.

Copy and paste the following HTML into your website