Research

I have previously worked in the areas of information-theoretic learning (ITL) and explainable artificial intelligence (XAI). I worked on investigative techniques for improving the interpretability of learning models’ training and results. This is done through “probing” the insides of the DL models or through creating models that may be modularized and are inherently more interpretable than the existing state-of-the-art. I still have an interest in this area and hope to utilize ITL in future work to explain learning models and guide DL design choices to help in the creation of state-of-the-art models.

Presently, I am working on semantic segmentation in remote sensing datasets and statistical texture analysis. Throughout my dissertation, I investigated the potential of combining learnable histogram features¹ with deep learning convolutional methods. Additionally, I have enhanced statistical texture feature learning in deep convolutional networks and explored the nuances of embedding histogram layers to maximize their impact within feature extraction pipelines. One application of these embedded histogram layers is for root-soil segmentation in mini-rhizobox environments², where the objects-of-interest are highly textured (eg. soil).

Publications

S. Chang, R. Chowdhry, A. Zare. Paper under double-blind review.

Y. Song, G. Sapes, S. Chang, R. Chowdhry, T. Mejia, A. Hampton, S. Kucharski, T. M. S. Sazzad, Y. Zhang, B. L. Tillman, M. F. R. Resende Jr, S. Koppal, C. Wilson, S. Gerber, A. Zare, W. M. Hammond. Hyperspectral signals in the soil: Plant-soil hydraulic connection and disequilibrium as mechanisms of drought tolerance and rapid recovery, Plant Cell Environ., Jun. 2024, doi: 10.1111/pce.15011.

S. J. Chang, R. Chowdhry, Y. Song, T. Mejia, A. Hampton, S. Kucharski, TM Sazzad, Y. Zhang, S. J. Koppal, C. H. Wilson, S. Gerber, B. Tillman, M. FR Resende Jr., W. M Hammond, A. Zare. HyperPRI: A Dataset of Hyperspectral Images for Underground Plant Root Study. 2023. Computers and Electronics in Agriculture, Volume 225, 2024, 109307, ISSN 0168-1699, https://doi.org/10.1016/j.compag.2024.109307.

S. Duan, S. Chang, and J. C. Principe, Labels, Information, and Computation: Efficient Learning Using Sufficient Labels, J. Mach. Learn. Res., vol. 24, no. 31, pp. 1–35, 2023. JMLR v24, 22-0019, also at arXiv.2104.09015.

S. Chang and J. C. Principe, Explaining Deep and ResNet Architecture Choices with Information Flow, in 2022 International Joint Conference on Neural Networks (IJCNN), Jul. 2022, pp. 1–6. doi: 10.1109/IJCNN55064.2022.9892065.

Theses

• Brain Tumor Classification Using Hit-or-Miss Capsule Layers (MS Thesis, 2019)