Rahul Mehrotra is a Product Manager at Maluuba, a Canadian AI company that’s teaching machines to think, reason and communicate with humans (acquired by Microsoft in January 2017). Based in the AI epicenter of Montréal, Maluuba applies deep learning techniques to solve complex problems in language understanding. Rahul works across Maluuba’s three research areas (Machine Comprehension, Dialogue Systems and Reinforcement Learning) and helps advance breakthrough research by providing real-world problems and use cases. Rahul leads product initiatives to bring cutting-edge academic research to robust product pipelines. Rahul holds a B.ASc in Systems Design Engineering from the University of Waterloo.
Building Literate Machines
Advances in AI research have led to great innovations based on image and voice recognition, and 2017 will see further advances in the field of language, including the creation of more literate machines—those that can comprehend and communicate with humans but also machines that begin to model innate human-like skills.
In this talk, Rahul Mehrotra will explore how advances in deep and reinforcement learning are being applied to solve language understanding problems. You will gain a deeper understanding of the research fundamentals as well as implications and opportunities that language understanding AI services will bring. Rahul will outline how researchers are seeking to equip machines with higher level cognitive skills like common-sense reasoning, information seeking, transfer learning, and decision-making.
He will explain how these capabilities are being applied in enterprise, using practical examples across a range of business functions. These use cases are transformative.
To give just one example, knowledge workers and employees would no longer need to desperately search through an organization’s directories, repositories, emails, and other channels to find a specific document. Instead, the employee would communicate with an AI agent leveraging machine comprehension capabilities. The agent would be capable of answering the question in a security-compliant manner by having a deep understanding of the contents of the organization’s documents instead of simply retrieving based on keywords.
The talk will provide audience with key takeaways on the underlying research as well as the current and future applications of using language understanding AI in enterprise.
7. A look at the A.I. space.
CBInsights, 2016https://www.cbinsights.com/reports/CB-Insights-Artificial-Intelligence-Webinar.pdf
8. Speech Recognition
Object Recognition and Detection
Machine Translation
Reinforcement Learning
Reasoning and Memory
Natural Language Understanding
Application areas of Artificial Intelligence
9. Speech Recognition
Object Recognition and Detection
Machine Translation
Reinforcement Learning
Reasoning and Memory
} Perception
Natural Language Understanding
10. Speech Recognition
Object Recognition and Detection
Machine Translation
Reinforcement Learning
Reasoning and Memory
} Perception
Natural Language Understanding
}Intelligence
21. 0
1
2
3
4
Virtual agents are limited in intelligence today
Abilities
Depth
Today’s
Performance
Users expect this
Current assistants support 20 or so abilities,
each with limited depth.
Limitations of AI
22. 1 Trained, pre-programmed models for single, narrow domains
2 Requirement of huge in-flux of training data
The reality …
3 Lack of fundamental reasoning or transfer learning capabilities
Question Answering: Introduction
25. Ontological/
Rule based
Predictable and
limited keyword
based queries
Statistical
Machine
Learning
Queries with high
grammatical
diversity
Deep
Language
Learning
Models do not
require traditional,
task-specific feature
engineering.
The Past The Present The Future
Underlying Technology