After 19 amazing years at Iowa State University Dept. of Computer Science,I will be moving to Tulane University as the next Dean of the School of Science and Engineering. My family are I are excited about our life in the vibrant city of New Orleans.

Dr. Ali Ghanbari and I have received the Distinguished Paper Award at the 38th IEEE/ACM International Conference on Automated Software Engineering (ASE 2023) for our paper titled “Mutation-based Fault Localization of Deep Neural Networks.” ASE conference is a top-tier venue for research on techniques for automating software engineering practices such as software bug detection and localization.

In the realm of academia, the most profound indicator of our success lies in the caliber of students we mentor and prepare for the future. This year, it has been my privilege to collaborate closely with two exceptional individuals, guiding them to the successful completion of their doctoral and postdoctoral training programs.

My students and I have continued to make progress on exploring decomposing of neural networks into modules. Starting with our pioneering work on this topic that appeared at ESEC/FSE 2020 and subsequent work that showed the virtue of and methods for decomposing Convolutional Neural Networks at ICSE 2022, we have expanded our results to Recurrent Neural Networks (RNNs). During our recent research project, titled “Decomposing a Recurrent Neural Network into Modules for Enabling Reusability and Replacement,” my students and I introduced a groundbreaking method for reengineering recurrent neural networks (RNNs).Our findings, which my PhD student Sayem Imtiaz presented at the 45th International Conference on Software Engineering, highlight a novel technique to decompose RNNs into reusable and replaceable components, eliminating the need for retraining the entire model in some cases.

SIGPLAN perspectives has published our essay that provides a perspective on Hybrid SPLASH 2021 conference. SPLASH 2021 was the first hybrid conference at this scale, and one of the first ACM conferences after pandemic. This perspective is authored by Jonathan Aldrich, Sumon Biswas, Steve Blackburn, Benjamin Chung, Youyou Cong, Alex Potanin, Hridesh Rajan, and Talia Ringer. The authors jointly served as hybridization co-chairs of SPLASH 2021, except Rajan who was the general chair of the SPLASH 2021 conference.

I am celebrating the graduation of Sumon Biswas, Rangeet Pan and Iris Ma. Sumon and Rangeet were PhD students in my lab, and Iris was an undergraduate research assistant. After graduation Sumon Biswas will start a postdoctoral position at Carnegie Mellon University, Rangeet Pan will start as a researcher at IBM TJ Watson Research Center, and Iris Ma will start her PhD program at the University of California, Irvine.

In this paper, we propose an approach to decompose a convolutional neural network (CNN) model used for image classification into modules, in which each module can classify a single output class. Furthermore, these modules can be reused or replaced in different scenarios. Also, we evaluate our approach against the 𝐶𝑂2𝑒 consumption of models created from scratch and reusing or replacing decomposed modules.

This work attempts to inform the art and practice of designing data science (DS) pipeline. Our investigation suggest that DS pipeline is a well used software architecture but often built in an ad hoc manner. We demonstrated the importance of standardization and analysis framework for DS pipeline following the traditional software engineering research on software architecture and design patterns. We contributed three representations of DS pipelines that capture the essence of our subjects in theory, in-the-small, and in-the-large that would facilitate building new DS pipelines.

Sumon Biswas has successfully defended his MS thesis entitled “Understanding Unfairness and its Mitigation in Open-Source Machine Learning Models”. The abstract of the thesis is as follows:

This work proposed a causal method to measure fairness of components in machine learning pipeline. Specifically, it evaluates many preprocessing stages in the pipeline and shows that those stages can be root cause of societal bias.