Part 7/12:
Kumar underscored the importance of model reliability and explainability. He shared that ISRO develops methods to understand and validate AI outputs, such as using entropy measures like XC (Explainability Coupling Variable), which indicate the certainty of model decisions. This approach allows ISRO to confidently deploy AI tools in critical missions, ensuring robustness and trustworthiness.
He presented a visualization showing how training improves class separation in models, thus enabling ISRO to determine precise thresholds for fault detection. Such explainability is crucial not only for operational effectiveness but also for fostering confidence among engineers and stakeholders.