ML Models for Safety-Critical Systems with Lucas García - 705

W-Model Workflow

The W-model is a crucial framework for integrating AI models into safety-critical systems. Lucas García explains that it begins with defining requirements for the AI model, followed by data management and learning process management, which involves selecting suitable AI architectures and experimenting with hyperparameters 1. The model training phase requires appropriate hardware, such as GPUs, to develop a trained model. Verification and validation (V&V) are key components of this process, ensuring the model meets design specifications and performs effectively in real-world scenarios 2.

Verification is about checking if the model is correctly implementing the intended design, while validation ensures it meets its intended use.

--- Lucas García

The W-model's structure allows for a comprehensive approach to AI integration, addressing both verification and validation challenges.

   

Implementation Challenges

Implementing AI models in safety-critical systems presents unique challenges, particularly regarding hardware constraints. Lucas García highlights the need for translating AI models into code suitable for resource-constrained processors like microprocessors or FPGAs 3. This process involves ensuring that the translated code meets all initial requirements without introducing unintended behaviors. Manual conversion for each use case can be labor-intensive, adding complexity to the implementation phase 4.

You have to manually convert it for every particular use case.

--- Sam Charrington

These challenges underscore the importance of meticulous planning and execution in deploying AI models within critical systems.