SYS-CON MEDIA Authors: Yeshim Deniz, Kevin Benedict, Gilad Parann-Nissany

News Feed Item

Catalyst Files Patent for Next-Generation Technology Assisted Review Based on 'Reinforcement Learning'

New Research Validates That Continuous Learning Methods Improve Savings and Results in Technology Assisted Review

DENVER CO -- (Marketwired) -- 06/11/14 -- Catalyst Repository Systems -- a pioneer in developing secure, cloud-based software to help corporations and their law firms take control of e-discovery, compliance and regulatory matters -- today announced it has applied for a patent on the type of continuous learning capability it invented for its next-generation technology assisted review (TAR 2.0) platform, Insight Predict.

Described in the patent application as "reinforcement learning based document coding," Catalyst's TAR technology is able to continuously learn from actions taken by the review team throughout the review process. With reinforcement learning, certain actions -- such as coding a document as responsive or not or adding additional documents -- enable the system to continue to grow "smarter" in its ability to select relevant documents.

What is Reinforcement Learning?

Reinforcement learning differs from older TAR 1.0 systems which require training by a high-level attorney. This expensive and time-consuming approach requires the senior attorney to first review and code an initial training set of randomly selected documents. With Catalyst's reinforcement learning technology, the full review team can begin right away. As reviewers' judgments are fed back into the system and new documents added, the system's selection and ranking of relevant documents continuously improves.

A new, peer-reviewed study by two leading experts in e-discovery validates the effectiveness of continuous learning technologies in e-discovery. In a paper they will present at the Association of Computing Machinery Special Interest Group on Information Retrieval (SIGIR) international conference in July 2014, "Evaluation of Machine-Learning Protocols for Technology-Assisted Review in Electronic Discovery," Gordon V. Cormack and Maura R. Grossman conclude that non-random training methods using continuous active learning "require substantially and significantly less human review effort" and yield "generally superior results."

Why is Catalyst's Approach Unique?

Even among continuous learning systems, Catalyst's method is unique for its use of reinforcement learning rather than active learning. Active learning systems are geared towards optimizing the quality of the classifier, the algorithm that labels documents as relevant or not. By contrast, reinforcement learning is designed to optimize for the goal the user seeks to achieve, which is generally to find as many relevant documents as possible. In this way, reinforcement learning helps users reach that goal more quickly.

"In contrast to the 'one bite of the apple' approach of earlier TAR engines, Insight Predict is able to use judgmental seeds and relevance feedback to continuously learn and rank throughout the review process, while avoiding the problems of bias and incomplete coverage through its use of contextual diversity," said John Tredennick, Catalyst's founder and CEO. "This is a major benefit to our clients because it eliminates the need for subject-matter experts for training, allows the review to get started sooner, accommodates rolling uploads, and ultimately delivers savings in time and costs."

Catalyst's unique reinforcement learning system was developed by Dr. Jeremy Pickens, Catalyst's senior research scientist, and Bruce Kiefer, Catalyst's vice president, platform. Pickens, one of the world's leading search scientists and a pioneer in the field of collaborative exploratory search, has a number of patents and patents pending in the field of information retrieval.

Overcoming the Five Myths of TAR

Catalyst's technology upends a number of common misconceptions about TAR -- that training is finite based on an initial seed set, that documents for training must be selected at random, that subject matter experts are required to train the system, that training cannot start until all documents on hand, and that it does not work for non-English documents.

To read more about the myths surrounding TAR and how advanced systems disprove them, see John Tredennick's Law Technology News article, Five Myths About Technology Assisted Review.

About Catalyst Repository Systems

A pioneer in cloud-based litigation technology, Catalyst provides global corporations and their counsel with secure, hosted document repositories to manage discovery, regulatory inquiries and other complex legal matters. Clients use Insight, Catalyst's "Big Discovery" platform, and Insight Predict, our advanced technology assisted review engine, to reduce discovery costs and associated risks. Corporations gain greater control and predictability over the discovery process and greater visibility across all their legal matters.

For more information, visit Catalyst at www.catalystsecure.com or follow us on Twitter at: http://twitter.com/catalystsecure.

For more information, press only
Shana Graham
Plat4orm PR
Email Contact
206.661.6336

More Stories By Marketwired .

Copyright © 2009 Marketwired. All rights reserved. All the news releases provided by Marketwired are copyrighted. Any forms of copying other than an individual user's personal reference without express written permission is prohibited. Further distribution of these materials is strictly forbidden, including but not limited to, posting, emailing, faxing, archiving in a public database, redistributing via a computer network or in a printed form.