Maluuba open-sources two datasets to improve AI’s reading comprehension and conversational skills

Briefing

Maluuba open-sources two datasets to improve AI’s reading comprehension and conversational skills

March 15, 2017

Briefing

  • Two Training Datasets – Canadian deep learning startup Maluuba released two datasets that can train AI programs to learn reading comprehension and conversational dialogue
  • NewsQA Dataset – Has more than 110,000 training questions on reading comprehension taken from CNN news articles, contributed by one group of online workers
  • Frames Dataset – Contains 1,368 dialogues based on online chat conversations for booking vacations, challenging algorithms to engage in back-and-forth dialogue, while retaining memory of conversation
  • AI Training – Used in training Maluuba's own deep learning algorithms, and made them public to help other researchers improve existing machine comprehension technologies
  • Potential Applications – Include smarter personal assistants, online chatbots, and intelligent robots that can match or exceed human intelligence

Accelerator

Sector

Information Technology

Organization

Maluuba

Source

Original Publication Date

December 1, 2016

Leave a comment