Scientific Inference With Interpretable Machine Learning: Analyzing Models to Learn About Real-World Phenomena

Abstract

To learn about real world phenomena, scientists have traditionally used models with clearly interpretable elements. However, modern machine learning (ML) models, while powerful predictors, lack this direct elementwise interpretability (e.g. neural network weights). Interpretable machine learning (IML) offers a solution by analyzing models holistically to derive interpretations. Yet, current IML research is focused on auditing ML models rather than leveraging them for scientific inference. Our work bridges this gap, presenting a framework for designing IML methods-termed 'property descriptors' -- that illuminate not just the model, but also the phenomenon it represents. We demonstrate that property descriptors, grounded in statistical learning theory, can effectively reveal relevant properties of the joint probability distribution of the observational data. We identify existing IML methods suited for scientific inference and provide a guide for developing new descriptors with quantified epistemic uncertainty. Our framework empowers scientists to harness ML models for inference, and provides directions for future IML research to support scientific understanding.

The paper has been published at "Minds and Machines" and is accessible at https://doi.org/10.1007/s11023-024-09691-z

References in corpus (41)

• Towards A Rigorous Science of Interpretable Machine Learning
• To Explain or to Predict?
• Explaining Machine Learning Classifiers through Diverse Counterfactual Explanations
• All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously
• Meta-learners for Estimating Heterogeneous Treatment Effects using Machine Learning
• The frontier of simulation-based inference
• Explainable Machine Learning for Scientific Insights and Discoveries
• A Theoretically Grounded Application of Dropout in Recurrent Neural Networks
• Interpretable Machine Learning -- A Brief History, State-of-the-Art and Challenges
• This Looks Like That: Deep Learning for Interpretable Image Recognition
• Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)
• Uncertainty Estimation Using a Single Deep Deterministic Neural Network
• Forecasting Corn Yield with Machine Learning Ensembles
• Understanding Global Feature Contributions With Additive Importance Measures
• Strong Completeness and Faithfulness in Bayesian Networks
• Multi-Objective Counterfactual Explanations
• Multifaceted Feature Visualization: Uncovering the Different Types of Features Learned By Each Neuron in Deep Neural Networks
• Feature relevance quantification in explainable AI: A causal problem
• Interpretable Classification Models for Recidivism Prediction
• Quantifying Uncertainty in Random Forests via Confidence Intervals and Hypothesis Tests
• Visualizing the Feature Importance for Black Box Models
• Relating the Partial Dependence Plot and Permutation Feature Importance to the Data Generating Process
• Model-agnostic Feature Importance and Effects with Dependent Features -- A Conditional Subgroup Approach
• Double Machine Learning based Program Evaluation under Unconfoundedness
• Demystifying statistical learning based on efficient influence functions
• The no-free-lunch theorems of supervised learning
• Revisiting the Importance of Individual Units in CNNs via Ablation
• CXPlain: Causal Explanations for Model Interpretation under Uncertainty
• A Double Machine Learning Trend Model for Citizen Science Data
• Novel deep learning methods for track reconstruction
• A Guide to Feature Importance Methods for Scientific Inference
• Sampling, Intervention, Prediction, Aggregation: A Generalized Framework for Model-Agnostic Interpretations
• Shapley Flow: A Graph-based Approach to Interpreting Model Predictions
• Causal Reinforcement Learning using Observational and Interventional Data
• Selectivity considered harmful: evaluating the causal impact of class selectivity in DNNs
• Emergence of Concepts in DNNs?
• Improvement-Focused Causal Recourse (ICR)
• Explaining Hyperparameter Optimization via Partial Dependence Plots
• Generating Interpretable Counterfactual Explanations By Implicit Minimisation of Epistemic and Aleatoric Uncertainties
• Artificial Neural Nets and the Representation of Human Concepts
• DeDUCE: Generating Counterfactual Explanations Efficiently