Board 425: Using Neural Networks to Provide Automated Feedback on Elementary Mathematics Instruction

Peter Youngs; Scout Beron Crimmins; Jonathan Foster; Matthew Korban; Scott T. Acton; Ginger S. Watson

Download Paper | Permalink

Conference: 2023 ASEE Annual Conference & Exposition
Location: Baltimore , Maryland
Publication Date: June 25, 2023
Start Date: June 25, 2023
End Date: June 28, 2023
Conference Session: NSF Grantees Poster Session
Tagged Topic: NSF Grantees Poster Session
Page Count: 6
DOI: 10.18260/1-2--42768
Permanent URL: https://strategy.asee.org/42768
Download Count: 148

Request a correction

Paper Authors

biography

Peter Youngs University of Virginia

visit author page

Peter Youngs is a professor in the Department of Curriculum, Instruction, and Special Education at University of Virginia. He conducts research on ways that neural networks can be used to (a) provide automated feedback to elementary teachers on their mathematics instruction and (b) make the process of analyzing videos of mathematics instruction more efficient and less time-consuming. His research has been funded by Institute of Education Sciences, National Science Foundation, Spencer Foundation, and William T. Grant Foundation. He currently serves as co-editor of American Educational Research Journal (2019 to 2024).

visit author page

biography

Scout Beron Crimmins University of Virginia

visit author page

Doctoral Student, UVA

visit author page

biography

Jonathan Foster University of Virginia

visit author page

Dr. Jonathan Foster is a Postdoctoral Research Associate at the University of Virginia, School of Education and Human Development. He holds a Ph.D. in Mathematics Education from University of Georgia and a B.S. in Mathematics from Wofford College. After getting his undergraduate degree, Jonathan taught a wide range of high school mathematics courses before embarking on his doctoral studies. As a former mathematics teacher, he brings a unique perspective to his interdisciplinary research on teacher education, discourse, and AI.

visit author page

biography

Matthew Korban University of Virginia

visit author page

Matthew Korban received his BSc and MSc degree in Electrical Engineering in 2013 from the University of Guilan, where he worked on sign language recognition in video. He received his PhD in Computer Engineering from Louisiana State University. He is currently a Postdoc Research Associate at the University of Virginia, working with Prof. Scott T. Acton. His research interest includes Human Action Recognition, Early Action Recognition, Motion Synthesis, and Human Geometric Modeling in Virtual Reality environments

visit author page

author page

Ginger S. Watson Old Dominion University

author page

Scott T. Acton California State University, Channel Islands

Download Paper | Permalink

Abstract

In recent years, several researchers have begun to use neural networks (i.e., a form of artificial intelligence) to provide automated classification of instructional activities in early childhood, elementary, and secondary classrooms (e.g., Author, 2022; Dale et al., 2022; Jacobs et al., 2022; Kelly et al., 2018; Ramakrishnan et al., 2019) in order to provide feedback to teachers on their instruction. For instance, neural networks can be trained to determine the percentage of time that teachers engage in whole group instruction, small group instruction, and individual seatwork during individual lessons in elementary classrooms (Author, 2022). Another promising research development involves efforts to train neural networks to summatively evaluate different aspects of instruction in ways that are consistent with those of trained human raters. For example, Ramakrishnan et al. (2019) provide evidence regarding the promise of neural networks to engage in summative evaluation of videos of early childhood classrooms. But there has been little attention to applying neural networks to summative classroom observation measures aligned to elementary mathematics instruction. In addition, there has been little systematic research on how features of mathematics instruction may be associated with ratings of ambitious instruction by trained humans. This study is designed to address these shortcomings in the research literature. We utilized 125 hours of videos of elementary mathematics lessons that had previously been rated by individuals trained in the use of the Mathematics-Scan (M-Scan) instrument (Berry et al., 2013). In particular, we examined correlations between the four M-Scan domains (i.e., mathematical tasks, discourse, representations, and coherence and (a) human annotations and (b) neural network classifications of individual features of instruction and combinations of aspects of instruction in the observed lessons. In addition, we have developed a teacher dashboard for use in providing automated feedback to teachers on their mathematics instruction. In this study, we report on qualitative data collected from six experienced elementary teachers on their perceptions of and experiences with teacher dashboard data (based on neural network analyses of videos of their own mathematics instruction).

Citation
Format

Youngs, P., & Crimmins , S. B., & Foster, J., & Korban, M., & Watson, G. S., & Acton, S. T. (2023, June), Board 425: Using Neural Networks to Provide Automated Feedback on Elementary Mathematics Instruction Paper presented at 2023 ASEE Annual Conference & Exposition, Baltimore , Maryland. 10.18260/1-2--42768

TY - CPAPER
AB - In recent years, several researchers have begun to use neural networks (i.e., a form of artificial intelligence) to provide automated classification of instructional activities in early childhood, elementary, and secondary classrooms (e.g., Author, 2022; Dale et al., 2022; Jacobs et al., 2022; Kelly et al., 2018; Ramakrishnan et al., 2019) in order to provide feedback to teachers on their instruction. For instance, neural networks can be trained to determine the percentage of time that teachers engage in whole group instruction, small group instruction, and individual seatwork during individual lessons in elementary classrooms (Author, 2022).
Another promising research development involves efforts to train neural networks to summatively evaluate different aspects of instruction in ways that are consistent with those of trained human raters. For example, Ramakrishnan et al. (2019) provide evidence regarding the promise of neural networks to engage in summative evaluation of videos of early childhood classrooms. But there has been little attention to applying neural networks to summative classroom observation measures aligned to elementary mathematics instruction. In addition, there has been little systematic research on how features of mathematics instruction may be associated with ratings of ambitious instruction by trained humans.
This study is designed to address these shortcomings in the research literature. We utilized 125 hours of videos of elementary mathematics lessons that had previously been rated by individuals trained in the use of the Mathematics-Scan (M-Scan) instrument (Berry et al., 2013). In particular, we examined correlations between the four M-Scan domains (i.e., mathematical tasks, discourse, representations, and coherence and (a) human annotations and (b) neural network classifications of individual features of instruction and combinations of aspects of instruction in the observed lessons.
In addition, we have developed a teacher dashboard for use in providing automated feedback to teachers on their mathematics instruction. In this study, we report on qualitative data collected from six experienced elementary teachers on their perceptions of and experiences with teacher dashboard data (based on neural network analyses of videos of their own mathematics instruction).

AU - Peter Youngs
AU - Scout Beron Crimmins
AU - Jonathan Foster
AU - Matthew Korban
AU - Ginger S. Watson
AU - Scott T. Acton
CY - Baltimore , Maryland
DA - 2023/06/25
PB - ASEE Conferences
TI - Board 425: Using Neural Networks to Provide Automated Feedback on Elementary Mathematics Instruction
UR - https://strategy.asee.org/42768
DO - 10.18260/1-2--42768
ER -