Maluuba

Maluuba was a Canadian artificial-intelligence research company founded in 2011 by four University of Waterloo undergraduate students: Zhiyuan Wu, Joshua Pantony, Sam Pasupalak, and Kaheer Suleman. The company began with a voice-based flight-search proof of concept, reflecting an early focus on enabling computers and consumer devices to interpret spoken requests and return useful answers rather than simply directing users to conventional search results. The company developed natural-language-understanding technology for applications in smartphones, smart televisions, automobiles, and connected devices. From 2013 onward, Maluuba partnered with companies operating in these sectors, and its technology was ultimately used by several consumer-electronics brands across more than 50 million devices. Its commercial proposition centered on voice interfaces, question answering, dialogue management, and the conversion of unstructured human language into actionable tasks. Maluuba also pursued an ambitious research program. In 2012 it obtained seed financing from Samsung Ventures, and in 2015 it raised a Series A round from Nautilus Ventures and Emerllion Capital. The same year, it opened a research-and-development laboratory in Montreal, advised by Yoshua Bengio and Richard Sutton and connected with research communities at the University of Montreal's MILA laboratory and McGill University. By 2016, the company had more than fifty employees and had published fifteen peer-reviewed papers in language understanding and related fields. Its best-known research work involved machine reading comprehension. In 2016 Maluuba demonstrated systems that answered questions about the content of J. K. Rowling's Harry Potter and the Philosopher's Stone and reported results on the MCTest benchmark that exceeded simpler word-matching methods. Its EpiReader system was presented as competitive with, and in some tests ahead of, systems associated with larger technology companies. Maluuba also released the NewsQA and Frames datasets, supporting wider research into reading comprehension and conversational language understanding. The company published work on dialogue-state tracking, natural-language generation, deep-reinforcement-learning dialogue policies, and scalable reinforcement learning; in 2017 its research team reported an agent that surpassed the human benchmark for Ms. Pac-Man on the Atari 2600. Microsoft announced the acquisition of Maluuba in January 2017. The transaction brought Maluuba's researchers and language-understanding work into Microsoft's broader artificial-intelligence activities. Reports later indicated that the company's Kitchener-Waterloo office closed and employees were moved to the Montreal operation in July 2017. As a standalone company, Maluuba is therefore best understood as an acquired AI research brand whose principal legacy is its contribution to voice interfaces, machine reading comprehension, dialogue research, open datasets, and reinforcement-learning methods.

Maluuba was established in 2011 by four University of Waterloo undergraduates: Zhiyuan Wu, Joshua Pantony, Sam Pasupalak, and Kaheer Suleman. Its early proof of concept enabled users to search for flights through voice commands. This starting point shaped the company's subsequent emphasis on natural-language understanding, speech-driven interfaces, and systems that could interpret requests in context. In February 2012, Maluuba received seed funding from Samsung Ventures. From 2013, it formed partnerships in smartphone, smart-TV, automotive, and Internet-of-Things markets. These relationships gave the company a route to commercial deployment while it continued developing underlying language technologies. Its products and research were intended to let users ask questions or perform tasks conversationally, producing direct answers or actions rather than merely presenting a list of web links. Maluuba raised a Series A investment in August 2015 from Nautilus Ventures and Emerllion Capital. In December of the same year, it opened a research-and-development laboratory in Montreal, Quebec. The lab was advised by Yoshua Bengio of the University of Montreal and Richard Sutton of the University of Alberta, and Maluuba collaborated with the University of Montreal's MILA laboratory and McGill University. By 2016, the company had more than fifty employees and had published fifteen peer-reviewed papers focused on language understanding. A major part of Maluuba's research program concerned machine reading comprehension. In March 2016, the company demonstrated a system on the MCTest benchmark and reported an advantage over word-matching approaches. Later that year it presented EpiReader, which used neural reading methods and large datasets including the CNN/Daily Mail collection and the Children's Book Test. Maluuba also demonstrated a machine-reading system able to answer questions about Harry Potter and the Philosopher's Stone. Its work contributed to a broader shift from keyword retrieval toward systems that attempted to understand passages and infer answers. The company expanded this work by releasing NewsQA, a dataset for question answering over news articles, and Frames, a human-generated conversational corpus. Its dialogue research addressed natural-language understanding, state tracking, natural-language generation, and dialogue-policy learning with deep reinforcement learning. Maluuba also investigated reinforcement learning more broadly, including methods intended to improve the scalability of intelligent agents. In June 2017, the research team reported that it had produced an agent that surpassed the human benchmark on Ms. Pac-Man for the Atari 2600. Microsoft announced the acquisition of Maluuba on January 13, 2017. The acquisition integrated the company's researchers and technology into Microsoft's artificial-intelligence work. Later reports said that Maluuba's Kitchener-Waterloo office closed in July 2017 and that employees moved to Montreal. The independent company consequently ceased to operate as a separate business, while its research, personnel, datasets, and language-understanding methods became part of the wider Microsoft AI ecosystem.

• 2017

  Acquired by Microsoft

  [1]Maluubaen.wikipedia.org

  Microsoft announced the acquisition of Maluuba, bringing its research team and language-understanding work into Microsoft.

• 2017

  Reported relocation of employees to Montreal

  [1]Maluubaen.wikipedia.org

  Reports said that Maluuba's Kitchener-Waterloo office closed and employees moved to the Montreal office.

• 2016

  Machine-reading and dataset research becomes prominent

  [1]Maluubaen.wikipedia.org

  Maluuba demonstrated machine-reading systems, including EpiReader, and released the NewsQA and Frames datasets.

• 2015

  Series A financing and Montreal research laboratory

  [1]Maluubaen.wikipedia.org

  Maluuba raised Series A financing and opened a Montreal research-and-development laboratory later that year.

• 2013

  Expansion into device and automotive partnerships

  [1]Maluubaen.wikipedia.org

  The company began partnering with businesses in smartphones, smart televisions, automotive systems, and connected devices.

• 2012

  Seed funding from Samsung Ventures

  [1]Maluubaen.wikipedia.org

  Maluuba secured its first reported institutional financing from Samsung Ventures.

• 2011

  Maluuba founded

  [1]Maluubaen.wikipedia.org

  Zhiyuan Wu, Joshua Pantony, Sam Pasupalak, and Kaheer Suleman founded Maluuba after developing an early voice-based flight-search proof of concept.

Applied and fundamental artificial-intelligence research focused on language understanding, voice interaction, machine reading, dialogue systems, and reinforcement learning.

• Voice interfaces

  Voice assistant and natural-language-understanding technology

  Maluuba developed language-understanding systems that allowed users to speak requests to devices and receive direct answers or initiate tasks. The technology was targeted at smartphones, smart televisions, automotive systems, and Internet-of-Things products, and was reported to have been used by several consumer-electronics brands.

• Machine reading comprehension · 2016

  EpiReader

  EpiReader was a machine-reading-comprehension system presented by Maluuba in 2016. It was designed to read text and answer questions using large-scale datasets such as CNN/Daily Mail and the Children's Book Test, forming part of the company's effort to move beyond keyword matching.

• Research dataset · 2016

  NewsQA

  NewsQA is a dataset released by Maluuba for studying question answering and reading comprehension over news articles. It supported research into systems that must interpret an article's meaning and identify an answer rather than rely on direct word matching.

• Dialogue dataset · 2016

  Frames

  Frames is a human-generated conversational dataset released by Maluuba for research into dialogue understanding. It was associated with the company's work on dialogue state tracking, natural-language understanding, and systems that learn policies for multi-turn interactions.

• 2017

  Microsoft acquisitionM&A

  Background

  Maluuba had developed commercial language-understanding applications while building a substantial research program in machine reading, dialogue, and reinforcement learning.

  Action taken

  Microsoft announced that it had acquired Maluuba and would integrate its people and technology into Microsoft's artificial-intelligence activities.

  Aftermath

  Maluuba no longer operated as an independent company. Reports later described the closure of its Kitchener-Waterloo office and the relocation of employees to Montreal.

  Source

1. 1Maluubaen.wikipedia.org