RIT Internship Reflection

Today was the day of the data collection. In the morning my two mentors, their professor, and me carried the large, heavy cases containing the camera, the tripod, and the camera's computer across campus to the roof of a building near the bus loop. The work was labor intensive and time consuming, but once the camera was set up the actual collect was quite relaxing. We waited for the cars to come into our frame and then packed up once we were done. I was tired at the end of the day but looked forward to working with the new data.

Today I ran my adapted code. It worked, but not very well. I had used a suboptimal loss function and thus my backpropagation was incorrect. I didn't quite understand where exactly to implement each of the functions I learned about. Therefore I went back to studying to learn what function is best used in what application and actual calculations each function is doing. After a while I eventually understood the concept and was able to optimize the code. Ironically I found a code that accomplished the same task as mine after I finished. I compared the two and while the online code was more compact, they both especially did the same thing. I was proud and felt accomplished to have learned the concepts so quickly.

I was given the task by my mentor to convert a linear neural network into a convolutional neural network. It was to test my understanding of the fundamentals. It was challenging but I was able to change it. However, it was at this point that an issue rose up; my computer isn't powerful enough to run the code. I had to establish a link to my mentors computer, establish an account, recreate a virtual network, and rewrite the code. Unfortunately by this point I had run out of time and had to return home.

We are planning to have a data collect of the bus loop next week. We are planning to use the 12 bit hyperspectral camera from another lab and situate it on the roof of a nearby building. Another professor is having a data collect with the drones at the bus loop so we will gather data at the same time to guarantee that we have the data we need. Today we set up the camera in the lab and then took it down to make sure we understood how the camera works. I look forward to next week.

Today I continued learning about neural networks through the Stanford class. I am confident in my understanding of the foundational concepts of convolutional neural networks; from the basic system to more advanced concepts such as drop out and loss functions. I feel confident in my ability to understand the underlying theory of the code we are working on and was ready to be tested by my mentor.

Now that I understood Python as a language, it was time for me to learn about neural networks. Stanford offers videos for their cs231n class, which serves as a very good introduction into convolutional neural networks. My mentor advised me to look at this course. For the next few days I will spend my time exploring these classes. It should give me a solid foundation for the coding aspect of the research.