Page 7 - Computer Power User - February 2017

Research Aims To Make New System Could Lead To Fully

Big Data More Meaningful Automated Speech Recognition
In many cases, big data represents the future of technology and Speech-recognition systems today rely on machine learning:
helping consumers and companies make better decisions. But poring through millions of audio files and transcriptions to
one challenge is finding systems that can store, manage, and learn the acoustic features that correspond with typed words,
interpret all the data. Researchers according to researchers at MIT. But a new system could reduce
at MIT and Israel’s University the cost and time needed to develop such a system by instead
of Haifa are working to solve analyzing the correspondence between images and spoken
that. At a high level, the coreset- descriptions of those images. “The goal of this work is to try to
generation technique identifies get the machine to learn language more like the way humans
a small subset of your data that do,” said Jim Glass, a senior research scientist at CSAIL.
preserves the salient mathematical Researchers tested the system using a database of 1,000 images
relationships of the entire data and associated recordings of free-form verbal descriptions. They
set, according to MIT. You can then gave the system one of the recordings and asked it to
then use that smaller data set and retrieve 10 images that best matched the recording; the system
analyze it in ways that would be was able to return the correct image 31% of the time. Given
prohibitively time-consuming on that, Glass said the system under development is “taking baby
the full data set. MIT reports that steps” and has a long way to go. “But it’s an encouraging start.”
the technique is tailored to data
analysis tools with applications
in areas such as natural-language
processing, computer vision,
recommendation systems,
weather prediction, and more.
To test the technique, researchers
worked with a matrix mapping Image source: MIT News
every article against every word in the English version of Wikipedia
and were able to come up with clusters of words that denote the
100 most common topics on Wikipedia.

SOFTWARE SHORTS

YouTube VR App Lets You Experience and Byrne wrote in the Official YouTube Blog. As such, they’ve

Content In A More Immersive Way already been working with content creators to learn about,
experiment with, and create virtual reality video. Current
If you have a Daydream-ready phone such as a Pixel and a content includes news, gaming, and cooking videos, along
Daydream View headset and controller, Google is offering with videos that provide experiences such as swimming with
a new way to experience sharks, traveling to places such
content with its YouTube as Belize, or hiking a trail, the
VR app. “Every single video blog states. The YouTube VR
on the platform becomes an app is currently only available
immersive VR experience, on Daydream View, which
from 360-degree videos that is Google’s VR platform for
let you step inside the content Android the company launched
to standard videos shown on a virtual movie screen in the new late last year. In addition to the YouTube VR app, which has
theater mode,” according to Product Manager Erin Teague features for voice searching and the ability to follow channels
and Director of YouTube Creators Jamie Byrne. “More than you subscribe to and check out your playlists, Daydream lets
just an amazing new technology, VR allows us to make deep, you experience other Google apps in VR, including Google
human connections with people, places, and stories,” Teague Photos, Google Street View, and Google Play Movies.

CPU / February2017 7