Advancements in Safety-Critical Systems and Out-of-Distribution Detection

The field of safety-critical systems is witnessing significant advancements, driven by the need for reliable and robust solutions. A key area of focus is the development of effective out-of-distribution (OOD) detection techniques, which are crucial for ensuring the safety and performance of autonomous systems. Researchers are exploring innovative approaches to OOD detection, including the use of statistical methods, machine learning algorithms, and formal verification techniques. These efforts aim to address the challenges posed by distribution shifts, anomalies, and uncertainties in real-world environments. Noteworthy papers in this area include:

  • DEEDEE: Fast and Scalable Out-of-Distribution Dynamics Detection, which introduces a two-statistic detector that achieves state-of-the-art performance with reduced computational costs.
  • Taming Silent Failures: A Framework for Verifiable AI Reliability, which proposes a novel framework that synergizes formal synthesis and runtime monitoring to detect silent failures in AI systems.
  • Eigen-Value: Efficient Domain-Robust Data Valuation via Eigenvalue-Based Approach, which presents a plug-and-play data valuation framework that provides improved OOD robustness and stable value rankings across real-world datasets.

Sources

Aircraft Collision Avoidance Systems: Technological Challenges and Solutions on the Path to Regulatory Acceptance

Out-of-Distribution Detection for Safety Assurance of AI and Autonomous Systems

DEEDEE: Fast and Scalable Out-of-Distribution Dynamics Detection

Taming Silent Failures: A Framework for Verifiable AI Reliability

A method for outlier detection based on cluster analysis and visual expert criteria

Eigen-Value: Efficient Domain-Robust Data Valuation via Eigenvalue-Based Approach

Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations

Shift is Good: Mismatched Data Mixing Improves Test Performance

Built with on top of