About Me
I am a fourth-year Ph.D. student in Computer Science at Ohio State University’s Computer Vision Lab, working under the guidance of Dr. Jim Davis. My research centers on evaluating and mitigating uncertainty in multimodal deep learning models, with a particular emphasis on vision-language models.
I have developed and applied Reject Option Classification methods to enhance user confidence in unimodal n-way classification models, which established a foundation for my current focus on understanding uncertainty in multimodal vision-language modeling. Additionally, I have conducted research in Machine Translation, proposing new metrics to assess the effectiveness of multimodal translation models in visual disambiguation.
I am actively seeking a research internship to leverage my interdisciplinary skills in a practical setting, aiming to contribute to innovative AI projects while expanding my research expertise.
I hold a B.S. in Computer Science from Ohio State University with a minor in Mathematics, which has provided a strong foundation in theoretical and applied computer science, driving my research and future aspirations in the field.
Resume
Experience
Graduate Research Associate
Dept. of Computer Science and Engineering, Ohio State University
2021 - Present
Graduate Teaching Associate
Dept. of Computer Science and Engineering, Ohio State University
2023 - 2024
Research Intern
Air Force Research Laboratory, Wright State University / University of Dayton
2020 - 2024
Undergraduate Research Associate
Dept. of Computer Science and Engineering, Ohio State University
2020 - 2021
Summer Research Intern
Sii Canada, Concordia University
2019
Undergraduate Teaching Associate
Dept. of Computer Science and Engineering, Ohio State University
2018 - 2019
Publications
Here are some of my recent publications:
Learning When to Say “I Don’t Know”
N. Kashani Motlagh, J. Davis, T. Anderson, and J. Gwinnup
International Symposium on Visual Computing, October 2022.
[Best Paper Award]
Abstract: We propose a new Reject Option Classification technique to identify and remove regions of uncertainty in the decision space for a given neural classifier and dataset. Such existing formulations employ a learned rejection (remove)/selection (keep) function and require either a known cost for rejecting examples or strong constraints on the accuracy or coverage of the selected examples. We consider an alternative formulation by instead analyzing the complementary reject region and employing a validation set to learn per-class softmax thresholds. The goal is to maximize the accuracy of the selected examples subject to a natural randomness allowance on the rejected examples (rejecting more incorrect than correct predictions). We provide results showing the benefits of the proposed method over naïvely thresholding calibrated/uncalibrated softmax scores with 2-D points, imagery, and text classification datasets using state-of-the-art pretrained models. Source code is available at this https URL.
A Framework for Semi-automatic Collection of Temporal Satellite Imagery for Analysis of Dynamic Regions
N. Kashani Motlagh, A. Radhakrishnan, J. Davis, and R. Ilin
Learning to Understand Aerial Images (ICCV Workshop), October 2021.
Abstract: Analyzing natural and anthropogenic activities using re-mote sensing data has become a problem of increasing interest. However, this generally involves tediously labeling extensive imagery, perhaps on a global scale. The lack of a streamlined method to collect and label imagery over time makes it challenging to tackle these problems using popular, supervised deep learning approaches. We address this need by presenting a framework to semi-automatically collect and label dynamic regions in satellite imagery using crowd-sourced OpenStreetMap data and available satellite imagery resources. The generated labels can be quickly verified to ease the burden of full manual labeling. We leverage this framework for the ability to gather image sequences of areas that have label reclassification over time. One possible application of our framework is demonstrated to collect and classify construction vs. non-construction sites. Overall, the proposed framework can be adapted for similar change detection or classification tasks in various remote sensing applications.