It’s been a whirlwind of a first month working on Udacity’s Self Driving Car Nanodegree. We’ve covered everything from finding lane lines to neural networks to TensorFlow. I’ve struggled with switching from Ruby to Python, applying basic mathematics, and picking up TensorFlow syntax, but wouldn’t take back a second of it all. I think the most satisfying part is seeing my progress in a very short amount of time. Every lesson, video, piece of code typed gives me more extensive knowledge on what it will take to be a self driving car engineer.
Currently, I am about to build my first Convolutional Neural Network using the LeNet architecture. I’ll be using purely Python, TensorFlow, and an Amazon EC2 g2.2xlarge gpu to train the network. Out of the box, the LeNet architecture trains fairly sufficiently – hitting around 96% accuracy. It’ll be my job to tune the network even further, by experimenting with concepts such as dropout, augmenting the dataset, or visualizing each layer just to name a few. I’ll be spending the next week training the network and attempting to increase the accuracy of the model as best I can.
It’s really been an incredible year learning the basics of programming and now diving deeper into a very promising field in AI and self driving cars. I can’t wait to see what’s in store for the rest of the program, especially when we dive into C++ during Terms 2 and 3. I’ll try to post more frequently, but no promises.
Cheers!