Globally renowned research

Tackling reasoning, decision-making and communication in machines through groundbreaking research.

Neural Network 1.jpg

Maluuba datasets for machine literacy

Maluuba's sophisticated human-generated datasets are freely available to the Artificial Intelligence research community. Learn more about their design and development, read the related papers and access the datasets.

Access datasets

Hybrid Reward Architecture (HRA)

 Ms. Pac-Man is regarded as one of the hardest games from the Atari games set for AI to learn, due to the many unique situations and the limited number of lives. Maluuba's algorithm achieved the maximum possible score of 999,990 points.

learn more


"We're solving fundamental challenges in deep learning and natural language to power a new era of artificial intelligence."

- Kaheer Suleman, CTO and Co Founder.

With several patents, twenty peer-reviewed papers – and more coming soon - we combine advanced research with industry to further enhance our AI Platform. We seek to motivate and support the AI research community by sharing datasets and regularly attending leading events such as NIPS, ICML, ICLR and ACL. Oh, we're hiring too!

Keheer Quote.png

Featured Publications

Read our featured peer-reviewed publications, or view all publications.


May 2017

Advancing Machine Comprehension through question generation

While asking a question may seem straightforward, it is the process of asking the right question that can drive better understanding of concepts and information. While many QA datasets are geared to training for answering questions – an extractive task – the process of asking questions is comparatively abstractive: it requires the generation of text that may not appear in the context document. Asking ‘good’ questions involves skills beyond those needed to answer them.

Read paper


April 2017

Multi-Advisor Reinforcement Learning

In the first Maluuba paper following our acquisition by Microsoft, we deal with a novel branch of Separation of Concerns, called Multi-Advisor Reinforcement Learning (MAd-RL), where a single-agent RL problem is distributed to n learners, called advisors.

exploring  task decomposition and using multiple agents to handle subtasks 

View research


January 2017

Towards AGI: Building information-seeking agents

Our new paper and video outline a suite of tasks that teach artificial agents to accomplish tasks through efficient information-seeking behaviour. Such behaviour is a vital research step towards Artificial General Intelligence.

View research


December 2016

Frames: A Corpus For Adding Memory To Goal-Oriented Dialogue Systems

In this paper, study the role of memory in goal-oriented dialogue systems and we share a new dataset for the AI research community.

View publication

Access dataset



December 2016

Improving Scalability of Reinforcement Learning by Separation of Concerns

In this paper, we propose a framework for solving a single-agent task by using multiple agents, each focusing on different aspects of the task.

View publication


Our focus and approach

Machine reading comprehension aims at teaching artificial agents to read and understand natural language.

Almost all human knowledge can be distilled from stored text. From stone tablets to printed books to the internet, we write our ideas down. A machine capable of assimilating this information would be incredibly useful. A truly literate machine could answer users' questions about arbitrary subjects in open domains like an oracle -- and wouldn't need an engineer to feed it all useful information in advance (an impossible task, anyhow). A machine that reads on its own could incorporate dynamic information in real time from the world around it.

Advanced Conversational Systems

Without effective natural-language interaction, the benefits of literate machines will be largely inaccessible to most users. Building a conversational agent involves solving many problems, from understanding natural language and decision making to generating natural language. Our goal is to build systems which are knowledgeable and that can exchange information with users to help users accomplish tasks or gain knowledge.

Reinforcement Learning

An intelligent machine should be able to make decisions and learn from environmental feedback similar to humans. In contrast to supervised learning the agent does not require examples of correct or incorrect behaviour in reinforcement learning. Instead, it can improve its behaviour by itself by interacting with the environment and observing the rewards it gets for its actions. At Maluuba, we do fundamental research in scalability of Reinforcement Learning to allow machines to perform complex tasks in the real world.