Whitepapers from our advisory council

Original research from Verlo's lead advisor on the intersection of artificial intelligence and evidence-based litigation. New papers are added as they are published.

May 2019Bai, Bolonkin, Burgoon, Chen, Dunbar, Singh, Subrahmanian, Wu — arXiv:1905.08617By our lead advisor

Automatic Long-Term Deception Detection in Group Interaction Videos

Introduces a method for detecting prolonged deception in group interactions using the Resistance game (5–8 players, 2–3 hidden spies, 30–65 minute sessions). An ensemble model combines low- and high-level video features with a novel LiarRank feature class, achieving AUCs above 0.70 in a fully automated setting — a step beyond isolated single-person deception detection.

Read paper →

December 2017Wu, Singh, Davis, Subrahmanian — arXiv:1712.04415By our lead advisor

Deception Detection in Videos

A multi-modal system for automated deception detection in courtroom trial videos, combining vision (micro-expressions, Improved Dense Trajectory features), audio (MFCC), and text. The fully automated system reaches an AUC of 0.877 in 10-fold cross-validation, outperforming prior methods that rely on human annotations by roughly 5%.

Read paper →