dl-workshop-2022

---

This was a full-day in-person workshop which drew 21 participants, many of whom were industry professionals and academics. The workshop was supported by the IEEE St. Louis chapter and UMSL’s computer science department.

Learning Objectives

• Build, train, and evaluate convolutional networks using Tensorflow/Keras
• Learn several hacks such as regularization and data augmentation
• Learn to apply deep transfer learning to train models
• Understand how deep learning works, and deep learning engineers do

Morning session

⌚	⏳	Topic
9:00 AM	10 min	Introductions (organizers and participants)
9:10 AM	15 min	🖥️   Demo: Google’s teachable machine
		✏️   Task: Learn how to use teachable machine
9:25 AM	25 min	📖   Logistics, your day-long project, and dataset options
		📖   Introduction to (Multi-)Classification, evaluation, and baseline accuracy
		✏️   Score a 100% on quiz: (multi-)classification, evaluation, and baseline accuracy
9:50 AM	10 min	📖   An image is a two- or three-dimensional matrix
		🖥️   Demo: Reading images with Python (Tensorflow)
10:00 AM	30 min	✏️   Task: Pick a dataset for your project and practice reading the images
10:30 AM	30 min	🖥️   Demo: Using Xception network to make a prediction / elephant.jpg
		✏️   Task: Practice using Xception network, and decode predictions
		✏️   Optional Task: What does the model predict for Dwayne’s picture? Read: why it fails!
11:00 AM	20 min	📖   Introduction to convolutional neural network (CNN)
		🖥️   Demo: Build and train a basic CNN
11:20 AM	40 min	✏️   Task: Build your CNN model and obtain a high accuracy

🥪 Lunch - Noon to 1 PM

Afternoon session

⌚	⏳	Topic
1:00 PM	20 min	📖   Model regularization
		🖥️   Demo: Regularization
		✏️   Task: Score a 100% on quiz: Regularization
1:20 PM	30 min	✏️   Task: Try improving your model’s accuracy using dropout
1:50 PM	10 min	📖   Data augmentation
		🖥️   Demo: Applying data augmentation to your training, Notebook
2:00 PM	30 min	✏️   Task: Improve your model’s accuracy using data augmentation
2:30 PM	10 min	📖   Deep Transfer learning
		🖥️   Demo: Transfer learning using DenseNet121 model, Notebook
2:40 PM	30 min	✏️   Task: Improve your model’s accuracy using transfer learning
3:10 PM	10 min	✏️   Task: Complete evaluation survey
3:20 PM	30 min	✏️   Make a concept map and send it to Kate (kma9q7@mail.umsl.edu)
3:50 PM	10 min	🎓   Pictures and certificate distribution

Pre-workshop

• Flyer for advertising
• Pre-workshop announcement for registered participants
• Media coverage by UMSL Daily

Concept maps drawn by participants

Photos

Participants

Adam Brkljach, Alhassan Sahad, Bryan Garcia, Clara Lebow, Diana Hobbs, Ehab Ur Rahman, Georgy Sinitsyn, Hao Lan, Jaccob Stanton, Jackie Herbstreit, James McElhannon, Joshua Meppiel, Kenan Oestreich, Larry Kremer, Lasanthi Gamage, Manu Bhandari, Nilima Kafle, Norah Vii, Prathyusha Velupula, Princewill Okorie, and Sang Mai

Participants by profession:
Manager-1, Architect-1, Associate Professor-1, Computational Scientist-1, CTO-1, Data Scientist-1, Electrical Engineer-1, Machine Learning Engineer-1, Neuroimaging Engineer-1, Professor-1, Research assistant-1, Research Technician II-1, Senior Laboratory Techician-1, Software Engineer-3, and Student-5.

Participants by institution/organization:
Ameren-1, Arkansas State University-1, Boeing-1, Donald Danforth Plant Science Center-3, FinLocker LLC-2, Hubbell-1, Missouri S&T-1, Mizzou-2, UMSL-6, Washington University in St Louis-2, and Webster University-1.

Acknowledgements

We are thankful to:

• the organizers of the STL CyberCon for allowing us to advertise our workshop at the event,
• Sheila Stiles at UMSL Catering for helping us at the last moment,
• Dr. Cezary Janikow, the chair of the CS Department, for supporting us in several ways,
• Kimberly Stanger at the UMSL CS Department for helping us with several ideas, and
• the IEEE St. Louis section for sponsoring the morning coffee and bagels.

Organizers

Dr. Badri Adhikari

Shaney Flores

Colton Fitzjarrald

Amulya Reddy Lakku

Kate Arendes