Teaching FTC High School Students Tensorflow Lite and Model Training Strategies

I tweaked it to strengthen model training.

It is reasonably straight forward to teach high school students how to leverage a pre-leaned Tensorflow Lite Models. That is, assuming that they are familiar with implementing autonomous programs in Java for FTC (FIRST Tech Challenge). What is surprising is that some teams take the next step to reteach the supplied model to improve its performance and to enable their gaming strategy. This is not straight forward. It requires developing inferencing strategies and training their own models.

This seminar will investigate a strategy for approaching this year’s competition, outlining the tools required to train a model to accomplish it and show how to write programs to utilize trained models. It is targeted at anyone interested in using Android Tensorflow Lite in FTC (FIRST Tech Challenge) competitions, learning about what you can do to take it further and introducing machine learning. We’ll potentially leave the detailed training process for a later time.

Dr. Edward Epp created 6 machine learning high school robotics workshops impacting 95 students and their mentors. The materials included demos, Tensorflow Lite code walkthroughs, hands-on programming exercises, identifying machine learning limitations, convolutions and other machine learning internals We know that Machine Learning is an important technical skill. LinkedIn identified the Machine Learning Engineer as one of the five top emerging jobs in 2018. It is his hope that attending TensorFlow and machine learning experts are also interested in exploring connections between machine learning and STEM. For example, convolutions reinforce fundamental array manipulation and image representations. This is a link to one of the 6 workshops Ed presented:https://meetings.vtools.ieee.org/m/207937.

Bio: Dr. Edward C. Epp is a retired Intel senior software architect and former University of Portland associate professor. He has experience writing Digital Rights Management and Content Protection patents, supporting content security, leveraging software architectures, writing hardware robustness requirements, implementing software/firmware, and doing performance analysis. He authored a Java text book while at the University of Portland, and teaching object oriented design principles. Ed is currently the Oregon IEEE RAS chair.