Annotate to Crawl, Adapt to Walk, See NLP Run

Human Language Technology (HLT) cannot be “future proofed” because the target is organic, ever changing, human language. Like a weather forecast, a static model is only accurate at a point in time. What we need is the NLP equivalent of a weather app pulling in streaming live weather data.

Adaptation is the future of HLT: continuously learning models that are quickly spun up with an annotation effort to “crawl”; then adapted to learn new domains to “walk”; and receive the fine-tuning and corrections in the field from non-technical subject matter experts who will see HLT run!

Today, models of human language can learn and evolve in the same way as children, and adaptation systems are key to making that all work. Come to HLTCon to learn more about these innovations that will keep your analytics accurate and relevant to your mission.

Day 1: Ensuring Training Data Quality
March 16, 2021  •  11 AM – 1 PM  ET
KEYNOTE: Doug Reynolds

Day 2: Combating Misinformation
March 17, 2021 • 11 AM – 1 PM ET
KEYNOTE: Joe Campbell & Danelle Shah

Day 3: Ensuring Training Data Quality
March 18, 2021 • 11 AM – 1 PM ET
KEYNOTE: Kfir Bar

Agenda

Day 1: Ensuring Training Data Quality

March 16, 2021 • 11 AM – 1 PM EDT
KEYNOTE: Doug Reynolds

Human labor is the slowest link in the journey from an idea for machine augmentation of analysts to a real, effective NLP system. Human-annotation of training data for machine learning systems dwarfs the engineering model building on a scale of more than 20:1. NLP experts from government, industry, and MIT with experience in production-ready systems, will introduce new strategies and technologies that are drastically decreasing human labor while also increasing the quality of the end product. Continuous feedback empowers non-technical users to correct and increase the accuracy of deployed models.

Day 2: Combating Misinformation

March 17, 2021 • 11 AM – 1 PM EDT
KEYNOTE: Joe Campbell & Danelle Shah

Information Operations is a particularly timely area of intelligence work that leverages NLP over a variety of media and languages. Players in disinformation are more elusive and difficult to address head-on, creating gaps in intelligence whether they represent misinformation or InfoOps players. Speakers will discuss the problem space of “you don’t know what you don’t know” to set the stage for day three of HLTCon.

Day 3: New Capabilities for the Mission

March 18, 2021 • 11 AM – 1 PM EDT
KEYNOTE: Kfir Bar

New NLP technology making progress towards solving the age-old intelligence problem: “How do you discover that which you don’t know to look for?” Event extraction can reveal patterns of actions involving players that were previously under the radar. What if a machine augmented intelligence gathering workflow that collected data on known persons of interest, also tracked peripheral yet-unknown persons (“ghosts”) to eventually build up full identity profiles? Learn how NLP is bringing the theoretical to life.