Distilling Knowledge from Deep Networks with Applications to Healthcare Domain

By Zhengping Che, Sanjay Purushotham, Robinder G. Khemani, Yan Liu. 2015

147 citations recorded by the source corpus.

Exponential growth in Electronic Healthcare Records (EHR) has resulted in new opportunities and urgent needs for discovery of meaningful data-driven representations and patterns of diseases in Computational Phenotyping research. Deep Learning models have shown superior performance for robust prediction in computational phenotyping tasks, but suffer from the issue of model interpretability which is crucial for clinicians involved in decision-making. In this paper, we introduce a novel knowledge-distillation approach called Interpretable Mimic Learning, to learn interpretable phenotype features for making robust prediction while mimicking the performance of deep learning models. Our framework uses Gradient Boosting Trees to learn interpretable features from deep learning models such as Stacked Denoising Autoencoder and Long Short-Term Memory. Exhaustive experiments on a real-world clinical time-series dataset show that our method obtains similar or better performance than the deep learning models, and it provides interpretable phenotypes for clinical decision making.

Subfields: Artificial Intelligence, Efficiency & Compression, Medical & Health AI, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Efficiency & Compression

In Medical & Health AI

In Signal Processing

Citations in the corpus

Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI) (2018, background): “[123] introduced in their paper a knowledge-distillation approach called Interpretable Mimic Learning, to learn interpretable phenotype features for making robust prediction while mimicking the performance of deep learning models.”
RETAIN: An Interpretable Predictive Model for Healthcare using Reverse Time Attention Mechanism (2016, background): “But The recurrence is on the attention generation components ( h i or g i ) while the hidden state v i is generated by a simpler more interpretable output. from use of RNNs is at the cost of model output that is notoriously difﬁcult to interpret.”
Reinforcement Learning in Healthcare: A Survey (2019, background): “models in healthcare settings [33, 241], how to develop interpretable RL solutions to increase the robustness, safety and correctness of learned strategies in healthcare domains is still an unsolved issue.”
Opportunities and obstacles for deep learning in biology and medicine (2017, methodology): “[364] used gradient boosted trees to learn interpretable healthcare features from trained deep models.”
Doctor AI: Predicting Clinical Events via Recurrent Neural Networks (2015, background): “Although, several recent attempts have been made to interpret the parameters of RNNs, such as [12] and this problem remains an active line of research [22, 7].”
Deep Learning for Health Informatics (2017, background): “[101] highlighted that although DNNs outperform conventional machine learning approaches in their ability to predict and classify clinical events, they suffer from the issue of model interpretability, which is important for clinical adaptation.”
Improving the Interpretability of Deep Neural Networks with Knowledge Distillation (2018, methodology): “In the health care domain, two pipelines [41] [4] are proposed to distill the knowledge from a DNN to Gradient Boosting Trees (GBTs) [39] [40].”
An Interpretable Classification Framework for Information Extraction from Online Healthcare Forums (2017, background): “Second, when categorizing post sentences into different classes, it is difficult tomake the tradeoff between classification accuracy and interpretability [4, 5].”
Interpreting Black-Box Models: A Review on Explainable Artificial Intelligence (2023, methodology): “Che et al. proposed Interpretable Mimic Learning, a model distillation-based method that has the capacity to learn interpretable phenotypic traits for providing potent predictions [104].”
Evaluating Explanation Without Ground Truth in Interpretable Machine Learning (2019, methodology): “For instance, in work [5], the authors conducted experiments with several sets of teacher-student models, where the teacher is the target model and the student is the proxy model.”
Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review (2018, background): “91,115 The main idea is to use the complex model to generate more soft-labeled examples to train a simpler model.”
What Clinicians Want: Contextualizing Explainable Machine Learning for Clinical End Use (2019, background): “Model distillation (Hinton et al., 2015) refers to learning a “simpler” model class that perform at par with an complex ML model and can be considered as an attempt toward transparent design (Che et al., 2015).”
Explaining Therapy Predictions with Layer-Wise Relevance Propagation in Neural Networks (2018, methodology): “[34, 35] apply knowledge distillation to predict mortality and ventilator-free days for patients with acute hypoxemic respiratory failure.”
Learning Feature Engineering for Classification (2017, background): “Moreover, features derived by neural networks are often not interpretable which is an important factor in certain application domains such as healthcare [Che et al., 2015].”
Benchmark of Deep Learning Models on Large Healthcare MIMIC Datasets (2017, background): “Recent research has shown that deep learning methods achieve state-of-the-art performance in analyzing healthrelated data, such as ICU mortality prediction [6], phenotype discovery [7] and disease prediction [36].”
Multi-layer Representation Learning for Medical Concepts (2016, background): “However, despite their outstanding predictive performance, RNNs are difficult to interpret [6] which limits their applications in healthcare.”