The field of machine learning is moving towards a greater emphasis on causal discovery and explainability. Researchers are developing new methods to identify causal relationships in complex systems, such as neural networks and time series data. These methods include the use of generative models, causal graphs, and ensemble learning techniques. Additionally, there is a growing interest in explaining machine learning predictions, particularly in high-stakes domains like healthcare. Techniques like SHAP and causal feature attribution are being developed to provide more transparent and interpretable models. Noteworthy papers in this area include TranCIT, which introduces a comprehensive analysis pipeline for quantifying transient causal interactions, and Causal SHAP, which integrates causal relationships into feature attribution. Other notable papers include Causal Sensitivity Identification using Generative Learning and Causal Representation Learning from Network Data, which demonstrate the effectiveness of generative models and graph neural networks in identifying causal relationships.
Causal Discovery and Explainability in Machine Learning
Sources
TranCIT: Transient Causal Interaction Toolbox
Missing Data Imputation using Neural Cellular Automata
Causal SHAP: Feature Attribution with Dependency Awareness through Causal Discovery
Causal Sensitivity Identification using Generative Learning
Evaluating the stability of model explanations in instance-dependent cost-sensitive credit scoring
Effects of Distributional Biases on Gradient-Based Causal Discovery in the Bivariate Categorical Case
Causal representation learning from network data
Ensemble Learning for Healthcare: A Comparative Analysis of Hybrid Voting and Ensemble Stacking in Obesity Risk Prediction
Meta-Imputation Balanced (MIB): An Ensemble Approach for Handling Missing Data in Biomedical Machine Learning
An Empirical Evaluation of Factors Affecting SHAP Explanation of Time Series Classification
Nonnegative matrix factorization and the principle of the common cause
A Primer on Causal and Statistical Dataset Biases for Fair and Robust Image Analysis
Interpretable Clustering with Adaptive Heterogeneous Causal Structure Learning in Mixed Observational Data
Causal Debiasing Medical Multimodal Representation Learning with Missing Modalities