This week was a lot of fun, and I feel that I have already learned a lot! Machine learning has been a really interesting topic, and I’m looking forward to the rest of the summer.
My lab partner, Chris, and I coded a program that uses 10 perceptrons (for digits 0-9) to identify handwritten numbers in the MNIST database. From here, we incrementally worked toward using a webcam to take videos of our own handwritten digits and make a prediction. This final challenge required the use of ROS.
I am really happy with everything that I have learned so far. However, I have also been having a lot of technical difficulties. First, I had trouble working from the Ubuntu kernel running on my Windows laptop when it came to working with ‘matplotlib’. Then, I moved onto VirtualBox, which had a lot of trouble detecting the webcam that was necessary for the ROS part of the program. Although Pedro and Ajay (graduate students) tried to help, it got to the point where VirtualBox was too slow, and I could not type anything. Because of these struggles, I have been working on an Odroid since then. The Odroid is working well now, but it also took a lot of effort getting to this point. The Odroid cannot wirelessly connect to the Internet, and preset settings on the computer created difficulties. From there, I had to make sure that the previous code that I had written (which was in python3) would run on the Odroid, which runs a version of ROS that uses python2. After overcoming these difficulties over a stretch of days, I have been able to finally settle down and work on the machine learning activities.
I am increasingly realizing that the graduate students here are really cool. The graduate student that I primarily work with, Pedro, is extremely helpful. I have had a lot of technical difficulties throughout the week, and he was always willing to help me out. The atmosphere of the lab is pretty light, and this makes the work environment enjoyable.
Today, we had a brief lecture on more advanced neural networks for machine learning. This weekend, I am going to look into TensorFlow using a Jupyter Notebook activity.