Event Debrief: Montreal AI Symposium

 
Montreal Symposium.png
 

In a world where data is abundant and varied, it is not access to data, but access to information and insight that becomes critical. Most unstructured information is hard and inefficient for humans to parse through; creating opportunities for more effective machine driven solutions. At IMRSV Data Labs, we develop competitive NLP technologies for our clients by conducting AI research that employs both machine learning and deep learning algorithms. In particular, text summarization has been a consistent topic of focus in our research at IMSRV from day one.

Given our background in legal information services and legal analytics, it made sense to start tackling the problem of text summarization in that domain. Specifically, case search is a routine practice that we identified as an area with major opportunity. This activity requires legal professionals to sift through a database of cases, commonly using a search tool, and analyze the results to select the most pertinent to their case. This analysis phase requires a judgement to be made regarding the topic, the facts, and the case outcome by “reading” through it. Although law professionals are experts at sifting through cases, this still takes a significant amount of time that could be saved with automated summaries. Very few of the available legal cases online are accompanied by case summaries, as they are generally written by legal professionals and are typically only written for landmark cases. By using an automated case summarization tool, we could reduce costs and improve the efficiency of lawyers and self-represented litigants conducting legal research.

This is what we started to research back in January 2017. After many months, we developed a novel algorithm that yields accurate summaries of legal texts by cleverly combining two popular, but not previously explored together, machine learning techniques: Dimensionality reduction from classical machine learning and word embedding from the deep learning era. By combining these two techniques, we were able to extract the salient information that is worth displaying to the legal case researcher from long legal documents. The intelligent algorithm is capable of convincingly selecting the key sentences from the long document to compose the summary. Following this achievement, we set to share those early results with the AI community. As such, in September 2017 we successfully submitted an application to the inaugural Montreal AI symposium and were invited to present our work.

The symposium was organized by leading figures in AI research: Senior chair Hugo Larochelle from Google who heads the Google Brain Research in Montreal (pioneered AI research under Yoshua Bengio), Joelle Pineau from McGill who is the newly appointed Head of Facebook’s AI Research in Montreal, Adam Trischler who is the Research Manager from Maluuba (Microsoft Research Lab), and Nicolas Chapados who is the Chief Scientist at Element AI & a Co-founder of Imagia.

The event sought to bring AI researchers, experts, and professionals from the Greater Montreal Area under the same roof to discuss and share outstanding AI research progress and AI industry opportunities.

We received a lot of interest and engagement from the 300 attendants (capacity was unfortunately capped due to the size of the venue), both at the poster session for our research and at our booth for the live demonstrations of AI applications. The booth interactions were also an opportunity to interact and address questions from attending researchers and professionals alike.

Since then, we’ve built on these successes, extending our core algorithm to support automatic extraction of keywords and key concepts from documents, thereby complementing the summarization. We’ve also worked on adjacent tasks, such as sentence classification, document classification, semantic search, and transcription-to-text.

Our summarization technology can be used to summarize any long form document: from legal documents to written essays or news articles. It’s even been tested on research papers. To view our program in action, request a free demo or view a summary of this very post shown below.  


SAMPLE SUMMARY (30% Compression) 

In a world where data is abundant and varied, it is not access to data, but access to information and insight that becomes critical. Most unstructured information is hard and inefficient for humans to parse through; creating opportunities for more effective machine driven solutions. At IMRSV Data Labs, we develop competitive NLP technologies for our clients by conducting AI research that employs both machine learning and deep learning algorithms. Given our background in legal information services and legal analytics, it made sense to start tackling the problem of text summarization in that domain. This activity requires legal professionals to sift through a database of cases, commonly using a search tool, and analyze the results to select the most pertinent to their case. By using an automated case summarization tool, we could reduce costs and improve the efficiency of lawyers and self-represented litigants conducting legal research. After many months, we developed a novel algorithm that yields accurate summaries of legal texts by cleverly combining two popular, but not previously explored together, machine learning techniques: Dimensionality reduction from classical machine learning and word embedding from the deep learning era. The event sought to bring AI researchers, experts, and professionals from the Greater Montreal Area under the same roof to discuss and share outstanding AI research progress and AI industry opportunities. We received a lot of interest and engagement from the 300 attendants (capacity was unfortunately capped due to the size of the venue), both at the poster session for our research and at our booth for the live demonstrations of AI applications.

Simon Hicks