Continuous Multimodal Learning Adds Powerful New Features and Flexibility to Workflows
Written by Brainspace on December 13, 2017
Designed for fast moving applications in investigations, e-discovery, information governance, national intelligence, and beyond, CMML makes machine learning a natural part of searching for just the key information. It lets you discover novel insights in your data at a fraction of the time and cost of traditional methods.
TAR (Technology Assisted Review) software, including Brainspace 6’s market leading Predictive Coding capability, has reduced the costs of document review in litigation, antitrust second requests, and similar applications. Users code batches of documents, teaching the software how to cull or prioritize a much larger data set, thus greatly reducing manual review effort.
Next-generation text analytics applications have emerged, however, that demand a faster and more flexible approach to text analytics than batch-oriented machine learning. Investigations, early case assessment (ECA), analysis of inbound productions, and other tasks focus on getting to the key information, not processing documents.
Brainspace’s new CMML technology makes machine learning a natural byproduct of assembling just the evidence you need. Tag documents for follow up as you find them, and use them to power predictive modeling at any time to find even more crucial material. Work completely within Brainspace, or in combination with a review platform if desired.
CMML uses the same supervised learning and active learning algorithms supported by Brainspace’s world class Predictive Coding capability. CMML also supports iterative relevance feedback: the newly popular continuous strategy of training on top-ranked documents from predictive models to find particularly rare documents.
Next-generation text applications can vary more in their workflows than do traditional document reviews. Investigators have different styles, data sets have different constraints, and a new piece of information may take a project in an unpredicted direction.
Brainspace designed CMML for this flexible style of work. Users employ our new tagging and notebook functions to build up incremental results, test new lines of inquiry, and build multiple predictive models. They can quickly try out new topics and predictive models, and develop or abandon them as needed.
Conceptual search, clustering, communication analysis, metadata search, and other text analytics tools can be brought to bear on an entire dataset, or focused on a single set of documents of interest. Predictive modeling results can power any of Brainspace’s analytics tools, and these tools in turn can provide training data for predictive modeling, closing a virtuous circle that augments your efforts.
With CMML, machine learning serves you, not the other way around.
Brainspace 6 tailors management tools to the varied applications of CMML workflows. A graphical display shows progress at finding documents of interest in the entire data set, while optional statistical sampling provides a check on whether anything has been missed. The notebook facility displays progress on tagging particular sets of documents, supporting a lightweight review capability.
Decisions on training are supported by metrics that track the consistency and stability of supervised learning. And users have the option of taking a batch-oriented training approach when needed.