OpenAI announced it has developed an unsupervised sentiment neuron. According to an OpenAI blog post, it “learns an excellent representation sentiment, despite being trained only to predict the next character in the text of Amazon reviews.”
The representation contains a distinct sentiment neuron that contains almost all of the sentiment signal, according to OpenAI. The team explained its system beats other approaches on the Stanford Sentiment Treebank, while using less data.
“Our results are a promising step towards general unsupervised representation learning,” the team wrote. “We found the results by exploring whether we could learn good quality representations as a side effect of language modeling, and scaled up an existing model on a carefully-chosen dataset. Yet the underlying phenomena remain more mysterious than clear.”
More information on their system can be found here.
VersionOne releases 11th annual State of Agile Report
VersionOne released its 11th annual State of Agile Report, which includes insights on the most used agile tools, agile maturity in organizations across the globe, and the challenges companies face when scaling agile.
“The 11th annual State of Agile Report shows that enterprise agility is increasing throughout organizations and across almost all industries at an accelerated rate,” said VersionOne CEO and cofounder Robert Holler. “However, the survey also highlights that there is still a lot of opportunity for growth and that the momentum is far from slowing.”
The report revealed while agile adoption is still growing, 94% of respondents said that more than half of their organizations’ teams are no practicing agile. The report also stated that 71% of respondents’ organizations have current or planned DevOps initiatives.
DeepMind open sources Sonnet
It’s been about a year since DeepMind made the decision to use TensorFlow (TF) throughout the research organization. During this time, DeepMind built its own framework for neural network modules with TF, which they decided to open source this week.
Sonnet is a library that uses an object-oriented approach, allowing modules to be created to define the forward pass of some computation, writes DeepMind in a blog. Sonnet is also designed to work with TF, so it doesn’t prevent teams from accessing the underlying details such as Tensors and variable scopes. In addition, models written in Sonnet can be freely mixed with raw TF code, and other high level libraries.
More information as well as source code for Sonnet can be found here.
GM, SAE International announce competitors in AutoDrive Challenge
SAE International, SAE World Congress Experience, and General Motors have announced eight North American universities to compete in the upcoming AutoDrive Challenge, a three-year competition where students will focus on autonomous technologies.
The challenge’s goal is to develop and demonstrate a fully autonomous passenger vehicle. The competition’s main goal is to navigate through an urban driving course in an automated driving mode as described by the SAE Standard Level 4 definition by year three.
The following universities were selected for this challenge:
- Kettering University
- Michigan State University
- Michigan Tech
- North Carolina A&T University
- Texas A&M University
- University of Toronto
- University of Waterloo
- Virginia Tech
Shoutem launches new React Native-based app builder
Shoutem, a mobile app development platform provider, launched its fifth generation platform based on React Native and React.js. The React Native-based app builder is designed to let developers build native and cross-platform apps using small building blocks or extensions.
All extensions and design templates of the app builder are written with React Native, which gives JavaScript developers a “head start” when building their native iOS and Android apps, according to the company. With the launch of its new platform, Shoutem also will migrate 5,000 Android and iOS apps built with previous Shoutem platforms to React Native.