The Schwartz Report

Blog archive

Microsoft Aims at Smarter AI with Maluuba Acquisition

Microsoft is taking another key step to move forward to advance artificial intelligence (AI) and machine learning with its agreement to acquire Maluuba, a Montreal startup that promises to squash the limitations of search and AI. Maluuba, founded in 2011, has some of the most advanced deep learning and natural language research and datasets in the world, according to Microsoft.

The Maluuba team will become part of Microsoft's Artificial Intelligence and Research group. Announced last fall, the new group brings together 5,000 engineers and computer scientists from the company's Bing, Cortana, Ambient Computing and Robotics, and Information Platform groups. Microsoft has made advancing AI, machine learning and conversational computing a key priority. Cortana, Microsoft's digital assistant in Windows, is a key pillar of that effort.

But as Maluuba CTO and Cofounder Kaheer Suleman said last year at a demonstration in New York, today's digital assistants are confined to a limited number of buckets or domains.  This is why if you ask a question to Siri, Cortana or Google Now that they are designed to be asked, they work well. But if you ask anything outside of that they fall flat on their faces,  Suleman said (see the 17-minute video and YouTube demo here).

"Maluuba's vision is to advance toward a more general artificial intelligence by creating literate machines that can think, reason and communicate like humans -- a vision exactly in line with ours," said Harry Shum, executive VP of Microsoft's AI and Research Group, in a blog post announcing the deal. "Maluuba's impressive team is addressing some of the fundamental problems in language understanding by modeling some of the innate capabilities of the human brain -- from memory and common sense reasoning to curiosity and decision making. I've been in the AI research and development field for more than 20 years now and I'm incredibly excited about the scenarios that this acquisition could make possible in conversational AI."

The company's founders, CEO Sam Pasupalak and Suleman, met in an AI class at the University of Waterloo and share a common goal of making AI more intuitive by developing machine learning techniques optimized for natural interactions.  Professor Yoshua Bengio, described as "one of Deep Learning's founding fathers," was key in supporting their efforts and will advise Microsoft's Shum while maintaining his role as head of the Montreal Institute for Learning Algorithms.

"Understanding human language is an extremely complex task and, ultimately, the holy grail in the field of AI," the two said in a blog post announcing the Microsoft deal. "Microsoft provides us the opportunity to deliver our work to the billions of consumer and enterprise users that can benefit from the advent of truly intelligent machines."

The two also pointed to Microsoft's Azure public cloud as the optimal choice for applying the datasets the company has developed because of the GPUs and field-programmable gate arrays (FPGAs) used to bolster its global infrastructure, which the company revealed back at last fall's Ignite conference in Atlanta. Maluuba's most recent advance came last month when the company announced the release of two natural language datasets.

The first dataset is called NewsQA, which the company said can train algorithms to answer complex questions that typically require human comprehension and reasoning skills. It uses CNN articles from the DeepMind Q&A Dataset, described as a collection methodology based on "incomplete information and fostered curiosity." The questions in the dataset "require reasoning to answer, such as synthesis, inference and handling ambiguity, unlike other datasets that have focused on larger volumes yet simpler questions. The result is a robust dataset that will further drive natural language research," according to the company.

The other dataset, called Frames, addresses motivation and is based on 19,986 turns designed to train deep-learning algorithms to engage in natural conversations. The model simulates text-based conversations between a customer interacting with a travel agent.

Terms of the deal were not disclosed but the company was backed by Samsung Ventures.

Posted by Jeffrey Schwartz on 01/13/2017 at 12:40 PM


comments powered by Disqus

Subscribe on YouTube