HFC

Hybrid Forecasting Competition

Intelligence Value

The HFC program developed and tested hybrid geopolitical forecasting systems, which integrated human and machine forecasting components to create accurate, flexible, and scalable forecasting capabilities.

Summary

Intelligence analysts are faced with the daunting task of developing intellectually rigorous, well-supported assessments of geopolitical events using volumes of data that are often beyond their capability to fully ingest. Machine-driven analytical systems can process large amounts of data quickly and efficiently but lack the nuanced cognitive capabilities of human analysts. The goal of HFC was to integrate the strengths of human cognitive and reasoning abilities with those of machine-driven systems to produce maximally accurate forecasts of geopolitical and economic events. HFC-defined success outcomes include training methods for human forecasters to weigh human and machine judgments, predictive models that incorporate machine and human judgments, and crowdsourcing platforms that enable human and machine forecasters to interact effectively. HFC completed nearly three years of research in May 2020.

Accomplishments

  • Beat the state of the art for human-only forecast systems with a 10 percent increase in accuracy over baseline
  • Developed platforms to integrate human and machine forecasts
  • Created and evaluated hybrid forecasting training materials
  • Conducted foundational research on human trust of machine models

Related Publications

To access HFC program-related publications, please visit Google Scholar.

HFC Logo

Contact Information

Program Manager

Main Office

Related Program(s)

Broad Agency Announcement (BAA)

Link(s) to BAA

IARPA-BAA-16-02

Solicitation Status

CLOSED

Proposers' Day Date

February 3, 2016

BAA Release Date

September 12, 2016

BAA Question Period

October 3, 2016 — October 14, 2016

Proposal Due Date

November 17, 2016

Program Summary

Testing and Evaluation Partners

  • MITRE

Prime Performers

  • Raytheon BBN
  • University of Southern California Information Sciences Institute