The recent developments in the research area have shown a significant shift towards enhancing the robustness, interpretability, and safety of machine learning models, particularly in critical applications such as autonomous driving, fault diagnosis, and risk assessment. There is a growing emphasis on integrating uncertainty quantification into model predictions, which is evident in the use of Bayesian neural networks and ensemble learning methods. These approaches aim to provide more reliable and interpretable outputs, especially in scenarios where data is noisy or incomplete. Additionally, there is a notable trend towards incorporating system-level safety requirements into perception models through reinforcement learning, ensuring that model predictions not only improve accuracy but also align with broader safety objectives. The field is also witnessing advancements in the development of tools for probabilistic verification of neural networks, which are crucial for assessing and synthesizing safety guarantees in real-world applications. Overall, the research is moving towards creating more resilient and trustworthy AI systems that can operate effectively in uncertain and dynamic environments.
Enhancing Robustness and Safety in Machine Learning Models
Sources
On the Unknowable Limits to Prediction
A Multi-Loss Strategy for Vehicle Trajectory Prediction: Combining Off-Road, Diversity, and Directional Consistency Losses
Risk-Averse Certification of Bayesian Neural Networks
Probabilistic Prediction of Ship Maneuvering Motion using Ensemble Learning with Feedforward Neural Networks
Predictive Inference With Fast Feature Conformal Prediction
Calibration through the Lens of Interpretability
Towards Robust Interpretable Surrogates for Optimization
Uncertainty-Aware Artificial Intelligence for Gear Fault Diagnosis in Motor Drives
Learning Ensembles of Vision-based Safety Control Filters
Crash Severity Risk Modeling Strategies under Data Imbalance
Artificial Expert Intelligence through PAC-reasoning
SAVER: A Toolbox for Sampling-Based, Probabilistic Verification of Neural Networks
Incorporating System-level Safety Requirements in Perception Models via Reinforcement Learning
Risk-aware Classification via Uncertainty Quantification
Reinforced Symbolic Learning with Logical Constraints for Predicting Turbine Blade Fatigue Life
Safe Adaptive Cruise Control Under Perception Uncertainty: A Deep Ensemble and Conformal Tube Model Predictive Control Approach
Semi-automated transmission control for motorcycle gearshift: design, data-driven tuning and experimental validation
An In-Depth Examination of Risk Assessment in Multi-Class Classification Algorithms