Yashesh Dhebar

Black-box artificial intelligence (AI) induction methods such as deep reinforcement learning (DRL) are increasingly being used to find optimal policies for a given control task. Although policies represented using a black-box AI are capable of efficiently executing the underlying control task and achieving optimal closed-loop performance—controlling the agent from the initial time step until the successful termination of an episode, the developed control rules are often complex and neither interpretable nor explainable. In this article, we use a recently proposed nonlinear decision-tree (NLDT) approach to find a hierarchical set of control rules in an attempt to maximize the open-loop performance for approximating and explaining the pretrained black-box DRL (oracle) agent using the labeled state–action dataset. Recent advances in nonlinear optimization approaches using evolutionary computation facilitate finding a hierarchical set of nonlinear control rules as a function of state variables using a computationally fast bilevel optimization procedure at each node of the proposed NLDT. In addition, we propose a reoptimization procedure for enhancing the closed-loop performance of an already derived NLDT. We evaluate our proposed methodologies (open- and closed-loop NLDTs) on different control problems having multiple discrete actions. In all these problems, our proposed approach is able to find relatively simple and interpretable rules involving one to four nonlinear terms per rule, while simultaneously achieving on par closed-loop performance when compared to a trained black-box DRL agent. A postprocessing approach for simplifying the NLDT is also suggested. The obtained results are inspiring as they suggest the replacement of complicated black-box DRL policies involving thousands of parameters (making them noninterpretable) with relatively simple interpretable policies. The results are encouraging and motivating to pursue further applications of proposed approach in solving more complex control tasks.

Most evolutionary many-objective optimization (EMaO) algorithms start with a description of a number of predefined set of reference points on a unit simplex. So far, most studies have used the Das and Dennis’s structured approach for generating well-spaced reference points. Due to the highly structured nature of the procedure, this method cannot produce an arbitrary number of points, which is desired in an EMaO application. Although a layer-wise implementation has been suggested, EMO researchers always felt the need for a more generic approach. Motivated by earlier studies, we introduce a metric for defining well-spaced points on a unit simplex and propose a number of viable methods for generating such a set. We compare the proposed methods on a variety of performance metrics such as hypervolume, deviation in triangularized simplices, distance of the closest point pair, and variance of the geometric means to nearest neighbors in up to 15-dimensional spaces. We show that an iterative improvement based on Riesz s-Energy is able to effectively find an arbitrary number of wellspaced points even in higher-dimensional spaces. Reference points created using the proposed Riesz s-Energy method for a number of standard combinations of objectives and reference points as well as a source code written in Python are available publicly at https://www.egr.msu.edu/coinlab/blankjul/uniform.

For supervised classification problems involving design, control, and other practical purposes, users are not only interested in finding a highly accurate classifier but they also demand that the obtained classifier be easily interpretable. While the definition of interpretability of a classifier can vary from case to case, here, by a humanly interpretable classifier, we restrict it to be expressed in simplistic mathematical terms. As a novel approach, we represent a classifier as an assembly of simple mathematical rules using a nonlinear decision tree (NLDT). Each conditional (nonterminal) node of the tree represents a nonlinear mathematical rule (split-rule) involving features in order to partition the dataset in the given conditional node into two nonoverlapping subsets. This partitioning is intended to minimize the impurity of the resulting child nodes. By restricting the structure of the split-rule at each conditional node and depth of the decision tree, the interpretability of the classifier is ensured. The nonlinear split-rule at a given conditional node is obtained using an evolutionary bilevel optimization algorithm, in which while the upper level focuses on arriving at an interpretable structure of the split-rule, the lower level achieves the most appropriate weights (coefficients) of individual constituents of the rule to minimize the net impurity of two resulting child nodes. The performance of the proposed algorithm is demonstrated on a number of controlled test problems, existing benchmark problems, and industrial problems. Results on 2-500 feature problems are encouraging and open up further scopes of applying the proposed approach to more challenging and complex classification tasks.

Convolutional neural networks (CNNs) are the backbones of deep learning paradigms for numerous vision tasks. Early advancements in CNN architectures are primarily driven by human expertise and by elaborate design processes. Recently, neural architecture search was proposed with the aim of automating the network design process and generating task-dependent architectures. While existing approaches have achieved competitive performance in image classification, they are not well suited to problems where the computational budget is limited for two reasons: (1) the obtained architectures are either solely optimized for classification performance, or only for one deployment scenario; (2) the search process requires vast computational resources in most approaches. To overcome these limitations, we propose an evolutionary algorithm for searching neural architectures under multiple objectives, such as classification performance and floating point operations (FLOPs). The proposed method addresses the first shortcoming by populating a set of architectures to approximate the entire Pareto frontier through genetic operations that recombine and modify architectural components progressively. Our approach improves computational efficiency by carefully down-scaling the architectures during the search as well as reinforcing the patterns commonly shared among past successful architectures through Bayesian model learning. The integration of these two main contributions allows an efficient design of architectures that are competitive and in most cases outperform both manually and automatically designed architectures on benchmark image classification datasets: CIFAR, ImageNet and human chest X-ray. The flexibility provided from simultaneously obtaining multiple architecture choices for different compute requirements further differentiates our approach from other methods in the literature.