Improving End-to-End Models for Form Understanding with Synthetic Ground Truth Pairs
Methods for improving document understanding models using synthetic ground truth pairs of form data...
Document understanding, form parsing, and end-to-end visual models with synthetic ground truth.
1 active projects →Generating synthetic ground truth pairs for training robust document and form understanding models.
1 active projects →Building foundational components for long-horizon autonomous computer use tasks.
1 active projects →Sensor fusion and data collection for visual inertial odometry in aerospace applications.
1 active projects →Building and expanding foundational components critical to enabling long-horizon autonomous computer use tasks.
Data collection and sensor fusion for visual inertial odometry on dynamical kinetic systems (drones).
Like the legendary pirate utopia, Libertalia is built on principles of freedom and collective endeavor. We believe the most important problems in AI require open collaboration, reproducible research, and tools that belong to everyone.
All research, code, and datasets published under permissive licenses.
Every paper includes complete training configs and evaluation scripts.
Research directions decided collectively by active contributors.
Methods for improving document understanding models using synthetic ground truth pairs of form data...
Benchmarking LLMs for security log analysis. Introduces LLM4Sec pipeline achieving 0.998 F1-Score...
Neural architecture optimization focusing on channel configurations for improved CNN efficiency...
Libertalia is led by researchers and engineers with experience across ML, security, and startups.

Co-Founder
YC W24 Founder. Ex-Twitch, Ex-Microsoft ML. CVPR, ICCV.

Co-Founder
VAKTROS CEO. MSc CS (Security & ML). Canadian CTF Champion.

Co-Founder
Deep Tech Fund Advisors. YC'24. Compound (acquired).