Forney, A., & Korpusik, M., & Kitamura, M. (2022, August), Pacman Trainer: Classroom-Ready Deep Learning from Data to Deployment  Paper presented at 2022 ASEE Annual Conference & Exposition, Minneapolis, MN. 10.18260/1-2--40897

\bibitem{asee_peer_40897}
  Forney, A., \& Korpusik, M., \& Kitamura, M. (2022, August), \emph{Pacman Trainer: Classroom-Ready Deep Learning from Data to Deployment}
  Paper presented at 2022 ASEE Annual Conference \& Exposition, Minneapolis, MN.
  10.18260/1-2--40897

Andrew Forney, Mandy Korpusik, and Masao Kitamura.  "Pacman Trainer: Classroom-Ready Deep Learning from Data to Deployment".  2022 ASEE Annual Conference & Exposition, Minneapolis, MN, 2022, August.  ASEE Conferences, 2022.  https://strategy.asee.org/40897 Internet.  23 Jul, 2024

\bibitem{asee_peer_40897}
  Andrew Forney, Mandy Korpusik, and Masao Kitamura.
  "Pacman Trainer: Classroom-Ready Deep Learning from Data to Deployment".
  \emph{2022 ASEE Annual Conference \& Exposition, Minneapolis, MN, 2022, August}.
  ASEE Conferences, 2022.
  https://strategy.asee.org/40897 Internet.  23 Jul, 2024

@INPROCEEDINGS{asee_peer_40897
author = "Andrew Forney, Mandy Korpusik, and Masao Kitamura"
title = "Pacman Trainer: Classroom-Ready Deep Learning from Data to Deployment"
booktitle = "2022 ASEE Annual Conference \& Exposition"
year = "2022"
month = "August"
address = "Minneapolis, MN"
publisher = "ASEE Conferences"
note = {https://strategy.asee.org/40897}
number = {10.18260/1-2--40897}
}

TY  - CPAPER
AB  - Deep learning has seen a meteoric rise in the machine learning community and has vastly changed the landscape of many fields like computer vision and natural language processing. Yet, for all of its successful applications, connecting deep learning theory to practice remains a challenge in the classroom, with students typically only seeing parts of the data-collection, training, and deployment process at a time. To address these difficulties, this work presents Pacman Trainer, a web application that can be used as a class activity in which students provide the best move for Pacman to take in a given maze and prompt, thus generating a labeled dataset for supervised deep-imitation learning. By experiencing first-hand how training data is labeled, sanitized, vectorized, used during training, and then deployed to control Pacman in the same environment, students grasp the entirety of the deep learning pipeline concretely. This experience also highlights the shortcomings of deep imitation learning, which segues to discussions of overfitting, generalizability, and reinforcement learning alternatives, in which Pacman agents can be trained online in the same environment to juxtapose learning paradigms. Classroom-ready instructions, examples, and accessory exercises are provided using Pytorch, complete with clonable repositories suitable for GitHub Classroom integration.
AU  - Andrew Forney
AU  - Mandy Korpusik
AU  - Masao Kitamura
CY  - Minneapolis, MN
DA  - 2022/08/23
PB  - ASEE Conferences
TI  - Pacman Trainer: Classroom-Ready Deep Learning from Data to Deployment
UR  - https://strategy.asee.org/40897
DO  - 10.18260/1-2--40897
ER  -