AWS Public Sector Blog

What do Birds and Social Media Have in Common? AWS Machine Learning Support

Machine learning enables computers to make predictions from data without having been explicitly programmed to do so. Developers and researchers use AWS to develop and refine custom machine learning algorithms to help solve complex problems like classifying images and text.

In this post, we profile two academic researchers who used the AWS Cloud Credits for Research program to support their groundbreaking research. Learn how these researchers use the AWS Cloud to learn about language through 140 characters and identify different species of birds with just a photo.

Social Media Sentiment

Noah Smith, Associate Professor of Computer Science & Engineering at the University of Washington, designs algorithms for automated analysis of human language. He and his collaborators study social media to see what tweets and posts can reveal about language in today’s networked world.

“AWS supported a range of projects in the fields of computational linguistics and natural language processing. Most notably, we studied the link between sentiment expressed by social media users to public opinion surveys and language variation across the United States,” said Noah. “We also developed new tools for automated linguistic analysis of social media text.”

Social media messages offer new insight into regional language variation and change over time.  For example, new methods developed using AWS resources inferred a “subway map” of American cities, showing how new words on social media expand to usage in new places.

All About Birds

Serge Belongie, Professor in the Department of Computer Science at Cornell University and Cornell Tech, used image classification algorithms for a practical application. His team created the Merlin Bird App to offer quick identification help for beginning and intermediate bird watchers to learn about North America’s most common birds.

The app users can identify birds with just a photo or by answering five questions. Described as “like Shazam, but for birds,” users simply snap a photo of a bird, or pull one in from their camera roll, and Merlin Photo ID will offer a short list of possible matches. Merlin draws upon more than 370 million observations from the eBird citizen-science project and customizes the list to the species most likely to have been seen at the location and time of year.

“Our ability to use cloud storage and cloud computing on AWS has been instrumental for us to deploy the Merlin Bird Photo ID system to the public,” Serge said.

To build a successful visual recognition engine for birds, Serge and team created custom machine learning algorithms. The more data they collect, the more the algorithm is able to identify patterns. Those patterns are then used to improve Merlin’s performance and accuracy.

AWS Services for Artificial Intelligence

While these researchers used the power of AWS to develop their own Machine Learning algorithms, you do not need to be an expert in Artificial Intelligence (AI) to add Machine Learning algorithms to your applications. Amazon AI services enable you to:

  • Build conversational interfaces to your applications using voice and text
  • Add image analysis to your applications
  • Turn text into lifelike speech that you can store and play back
  • Build smart machine learning applications quickly and easily.

AWS also hosts several Public Datasets for Machine Learning research, including Common Crawl, Multimedia Commons, Landsat, and the Amazon Bin Image Dataset.

Are you an academic researcher interested in experimenting with ML and AI? Visit our AWS Cloud Credits for Research program.