Local Attribution

Local attribution methods aim to explain individual predictions made by complex machine learning models, identifying which input features or training data points most influenced the outcome. Current research focuses on improving the efficiency and accuracy of these methods, particularly for challenging scenarios like low signal-to-noise data and diverse model architectures (including neural networks and Siamese encoders), often employing techniques like surrogate modeling and integrated gradients. This work is crucial for enhancing the transparency and trustworthiness of machine learning systems across various applications, from image generation to time series analysis, by providing insights into model behavior and potential biases.