EIVA

01 — Overview

Research into whether redundant video evidence distorts video-level authentication confidence, and whether correlation-aware aggregation improves robustness without losing sensitivity to informative frames.

02 — Technology stack

• Python
• PyTorch
• scikit-learn

Taken from the project's manifests and architecture documentation.

03 — Technical highlight

Pre-registered hypotheses and leakage rules: split videos before perturbation, keep the test set out of model selection, record seeds and versions.

04 — Evidence

Scope and stack documented in source · reviewed 8 Nov 2026

• Private repository
• Architecture document (private repository)

Private repositories are available to review on request.