Skip to content

roadmap.md

Latest commit

 

History

History
32 lines (24 loc) · 1.75 KB

roadmap.md

File metadata and controls

32 lines (24 loc) · 1.75 KB

VERONA Roadmap

This roadmap outlines the planned development and priorities for VERONA.

1. Vision

VERONA aims to become the standard framework for scalable and reproducible neural network robustness experiments.

2. Near-Term Goals (Sept 2025 – Jan 2026)

  • Roadmap meeting – Finalise this draft in an open meeting with all council members.
  • Certified defense with randomized smoothing – Support for certified defense (yielding statistical robustness certificates) methods for image classifiers.
  • Vehicle integration for local robustness.

3. Mid-Term Goals (2026)

  • Vehicle integration for complex properties – Extend to properties with tree structures requiring GPU communication in parallel execution. Add an estimator for this.
  • Reduce dependency on AutoVerify.
  • Building dedicated interfaces to support adv attacks from foolbox and adversarial-attacks-pytorch
  • Support for tree-based models – Add support for decision trees and random forests (based on Marie’s bachelor project + student work).
  • Docker support – Provide Docker images for reproducibility and paper-specific setups.

4. Long-Term Goals

  • Maintain AutoVerify / Create lean version.
  • Platform independence for AutoVerify – Currently Linux-only.
  • Benchmarking – Use VERONA as the benchmarking tool for VNN-COMP.
  • Model card integration – Add robustness distributions to Hugging Face model cards.

5. Community & Contributions

  • [TODO] Add links to PRs open for contributions.

6. Updates

  • PyPI release: ada_verona v1.0.0 on [DATE] ([LINK])
  • First draft of roadmap released on [DATE].