Interpretability With Accurate Small Models

Ghose, Abhishek; Ravindran, Balaraman

doi:10.3389/frai.2020.00003

Cited by 3 publications

(3 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While decision trees (DT) are generally considered interpretable (Letham et al, 2015), trees of arbitrarily large depths can be difficult to understand (Ghose and Ravindran, 2020) and simulate (Lipton, 2018). A sufficiently sparse DT is desirable and considered interpretable (Lakkaraju et al, 2016).…”

Section: Icct Architecturementioning

confidence: 99%

Learning Interpretable, High-Performing Policies for Autonomous Driving

Paleja¹,

Niu²,

Silva³

et al. 2022

Robotics: Science and Systems XVIII

View full text Add to dashboard Cite

Gradient-based approaches in reinforcement learning (RL) have achieved tremendous success in learning policies for autonomous vehicles. While the performance of these approaches warrants real-world adoption, these policies lack interpretability, limiting deployability in the safety-critical and legally-regulated domain of autonomous driving (AD). AD requires interpretable and verifiable control policies that maintain high performance. We propose Interpretable Continuous Control Trees (ICCTs), a tree-based model that can be optimized via modern, gradient-based, RL approaches to produce highperforming, interpretable policies. The key to our approach is a procedure for allowing direct optimization in a sparse decisiontree-like representation. We validate ICCTs against baselines across six domains, showing that ICCTs are capable of learning interpretable policy representations that parity or outperform baselines by up to 33% in AD scenarios while achieving a 300x-600x reduction in the number of policy parameters against deep learning baselines. Furthermore, we demonstrate the interpretability and utility of our ICCTs through a 14-car physical robot demonstration. 1) We propose Interpretable Continuous Control Trees (IC-CTs), a novel tree-based model that can be optimized via gradient descent with modern RL algorithms to produce high-performance, interpretable continuous control policies. We provide several extensions to prior DDT i x k −b i > 0). As each leaf node is represented as a probability mass function over output classes in prior work, leaf nodes, l, must be modified to produce

show abstract

Section: Icct Architecturementioning

confidence: 99%

Learning Interpretable, High-Performing Policies for Autonomous Driving

Paleja¹,

Niu²,

Silva³

et al. 2022

Robotics: Science and Systems XVIII

View full text Add to dashboard Cite

show abstract

“…Therefore, interpretable models are preferably small in size, as well as of sufficient high-performance. In order to have high explanation complexity, there is a significant need for shrinkage methods for ML models [ 5 ]. For example, a decision tree of depth = 5 is easier to understand than one of depth = 50.…”

Section: Introductionmentioning

confidence: 99%

Evolutionary Algorithm for Improving Decision Tree with Global Discretization in Manufacturing

Jun

2021

Sensors

View full text Add to dashboard Cite

Due to the recent advance in the industrial Internet of Things (IoT) in manufacturing, the vast amount of data from sensors has triggered the need for leveraging such big data for fault detection. In particular, interpretable machine learning techniques, such as tree-based algorithms, have drawn attention to the need to implement reliable manufacturing systems, and identify the root causes of faults. However, despite the high interpretability of decision trees, tree-based models make a trade-off between accuracy and interpretability. In order to improve the tree’s performance while maintaining its interpretability, an evolutionary algorithm for discretization of multiple attributes, called Decision tree Improved by Multiple sPLits with Evolutionary algorithm for Discretization (DIMPLED), is proposed. The experimental results with two real-world datasets from sensors showed that the decision tree improved by DIMPLED outperformed the performances of single-decision-tree models (C4.5 and CART) that are widely used in practice, and it proved competitive compared to the ensemble methods, which have multiple decision trees. Even though the ensemble methods could produce slightly better performances, the proposed DIMPLED has a more interpretable structure, while maintaining an appropriate performance level.

show abstract

“…Alternatively, there are methods applied while generating a model that aim to find a trade-off between model accuracy and complexity [ 10 ]. At this stage, optimal sampling techniques or model structures that lead to higher accuracy and lower complexity might be determined [ 11 , 12 ].…”

Section: Introductionmentioning

confidence: 99%

Post-Analysis of Predictive Modeling with an Epidemiological Example

et al. 2021

View full text Add to dashboard Cite

Post-analysis of predictive models fosters their application in practice, as domain experts want to understand the logic behind them. In epidemiology, methods explaining sophisticated models facilitate the usage of up-to-date tools, especially in the high-dimensional predictor space. Investigating how model performance varies for subjects with different conditions is one of the important parts of post-analysis. This paper presents a model-independent approach for post-analysis, aiming to reveal those subjects’ conditions that lead to low or high model performance, compared to the average level on the whole sample. Conditions of interest are presented in the form of rules generated by a multi-objective evolutionary algorithm (MOGA). In this study, Lasso logistic regression (LLR) was trained to predict cardiovascular death by 2016 using the data from the 1984–1989 examination within the Kuopio Ischemic Heart Disease Risk Factor Study (KIHD), which contained 2682 subjects and 950 preselected predictors. After 50 independent runs of five-fold cross-validation, the model performance collected for each subject was used to generate rules describing “easy” and “difficult” cases. LLR with 61 selected predictors, on average, achieved 72.53% accuracy on the whole sample. However, during post-analysis, three categories of subjects were discovered: “Easy” cases with an LLR accuracy of 95.84%, “difficult” cases with an LLR accuracy of 48.11%, and the remaining cases with an LLR accuracy of 71.00%. Moreover, the rule analysis showed that medication was one of the main confusing factors that led to lower model performance. The proposed approach provides insightful information about subjects’ conditions that complicate predictive modeling.

show abstract

Interpretability With Accurate Small Models

Cited by 3 publications

References 55 publications

Learning Interpretable, High-Performing Policies for Autonomous Driving

Learning Interpretable, High-Performing Policies for Autonomous Driving

Evolutionary Algorithm for Improving Decision Tree with Global Discretization in Manufacturing

Post-Analysis of Predictive Modeling with an Epidemiological Example

Contact Info

Product

Resources

About