My independent project for this semester aims to create a report which compares several of the usual deep learning architecture using a set of grocery pictures. I started by watching and reading through open course materials and videos, which I found on the website of Stanford University, course CS231n. I also get in touch with a mentor from my hometown and sought his guidance on this topic. He is a graduate student at Fudan University, Shanghai. We arranged meetings every other week via Zoom and he would be helping me to better understand open course materials.
After a week of introduction, he encouraged me to set up a computational environment so I could run the machine learning code in it. With his advice, I rent a cloud computing host offered by Alibaba, a Chinese cloud service provider. The set-up process took me quite some time: the install procedures require me to download and install numerous related programs and customize the environment specifically for my cloud host. At the same time, I obtained the image resource I need under the direction of my mentor from a public image database.
Then, I started to experiment with what I have learnt and put them into practice. Tensorflow is the programming language I use to build deep learning model. Though based on Python, one of my familiar computer programming language, Tensorflow took me quite amount of time to get used to, especially for me to understand the underlying structure and concepts. I have just started to implement basic machine learning models with Tensorflow, and I am looking forward to using it in my research.
I am currently consistent with my timeline, but the initial learning part, especially when I learn to use Tensorflow, took a bit too long. I did not expect to meet such difficulties putting what I have learnt from open courses into practice. Also, there are other issues: the connection to and from the cloud server, for example, is frequently disturbed. I had to figure out a way to keep my working session intact without being disturbed by the interruptions in the connections to my cloud host. Still, I learnt much from the progress I have made. I learnt about many basic tactics when setting up a cloud host in another operating system I am not familiar with. Moreover, I learnt about where to look for when I faced a new challenge. Before working on this project, I was limited to consulting my mentor or looking around on the Internet aimlessly. Now, I will join online communities to solve problems I face and read through open source projects documentation.
To achieve my goal over the rest of the semester, to produce a research report that is, I will keep working on my timeline planed. Starting from the Thanksgiving break, I will move away from how to implement deep learning models to the actual process of experiment and comparison among different deep learning models.
Brokaw, Alex. “Google’s TensorFlow is now available for iOS.” The Verge, The Verge, 8 June 2016, http://www.theverge.com/2016/6/8/11885924/google-tensorflow-release-ios-magenta-neural-network.