AIAI (Artificial Intelligence for Advancing Instruction) Challenge - Phase 1

Classify instructional activities using multimodal classroom data. #education

$70,000 in prizes
4 days left
27 joined

About the sponsor

The challenge is sponsored by researchers at the University of Virginia and is part of the Artificial Intelligence for Advancing Instruction (AIAI) project. The purpose of the AIAI project is to explore how models can be used to classify instructional activities and discourse in videos of elementary mathematics and English language arts (ELA) instruction.


About the data

The data for this competition comes from the Development of Ambitious Instruction (DAI) study and Artificial Intelligence for Advancing Instruction (AIAI) project. The DAI study was a mid-scale, longitudinal study of elementary teacher preparation and instructional practice that collected classroom observation data of mathematics and English language arts (ELA) lessons from 83 teachers in their first three years of teaching. In preparation for the competition, the AIAI project annotated over 160 hours of classroom observation video data with 24 instructional activity labels and bounding boxes around the teachers’ activities; a subset of over 50 hours of classroom observation audio transcript data was annotated with 19 discourse labels.

Additional resources

  • Read our journal article in Computers and Education: Artificial Intelligence describing our automatic classification of activities in classroom videos results based solely on video annotations (i.e., no bounding box data). The results were achieved from a subset of approximately 244 hours of video recordings from the DAI dataset.
  • Read our publication from 2024 IEEE Southwest Symposium on Image Analysis and Interpretation describing our transformer network with multi-semantic attention. We presented an innovative activity recognition pipeline designed explicitly for instructional videos, leveraging a multi-semantic attention mechanism. The novel pipeline uses a transformer network that incorporates several types of instructional semantic attention, including teacher-to-students, students-to-students, teacher-to-object, and students-to-object relationships.
  • Read our book chapter from Oxford University Press’s Uses of Artificial Intelligence in STEM Education arguing that deep-learning neural networks have potential to help researchers save time and money in cataloging and analyzing large quantities of classroom videos by detecting features of instruction in videos. We evaluated the performance of three neural networks in detecting instructional activities in classroom videos. The video dataset included a total of forty-six hours, which was evenly split between elementary mathematics and English language arts lesson recordings.
  • Please visit these links for more information on the Protocol for Language Arts Teaching Observation (PLATO) and the Mathematics Scan (M-Scan), which our instructional activity and discourse labels were based on.