This week I worked more on my machine learning again. While machine learning confuses me more, I implemented a way to record images with a webcam via an imported library so that JavaScript could work with python. I found this a lot easier than the rest of the machine learning for the most part, I was surprised with the amount of HTML I remembered while testing the JavaScript before I used it in my python code. I don’t remember using the promise function much, but other than that it was fairly simple to code.
Later on, I visited the Laboratory Media studio. I loved the studios goal of trying to inspire disadvantaged youths when standard education won’t work for them. I really enjoyed looking around at all the work the students had made,
on Wednesday, I had two lectures. From the first lecture, we had a discussion on how we could use creative skills to try and achieve the UN’s Sustainable Development Goals (SDG’s). Even though I was not the best at coming up with ideas on the spot, I felt really inspired by some answers that other people gave for ways to help reach these goals; for instance, on an anonymous answer form, one person suggested that we could use performance art or murals to bring awareness to the UN’s goals as an interactive and subtle way to inspire conversations about inequality. I thought it was a very beautiful way to go about it.
Jamie also said that studies show that murals depicting sensitive topics can help bring awareness to physical, mental and interpersonal issues and can promote positive discussion. I really like the idea that raising awareness doesn’t have to be something overly complicated, as I feel as though awareness should be something relatively easily understood, (unless its purpose is to make people think, but even then I find that it should be easily able to be understood by the general public.) This lecture really made me think of the impact I have on influencing change as a creative, I’ve always wanted to make a positive different but didn’t know how. After this lecture I feel as though I have more of an idea now.
During the second lecture, Diana mentioned the failed metaverse project in Seoul – South Korea. I already knew parts about the metaverse, but not specifically to Seoul. I was surprised that so much money went into the project, especially since not many people even own a VR headset. I did find it interesting to discuss other reasons why it failed as well, such as the lack of trust in VR and general discomfort about wearing the headset for prolonged periods of time.
LLM were also mentioned, and how they need to make sure they are fully tested. The example we were given was a chatbot that helped assisting refugees by giving them resources to help escape their current living situations. There are a lot of specific worried that I didn’t even consider. We were given an example of a response where we discussed it, and I hadn’t even realised that the information given on how to travel to the area given was illegal! This really opened my eyes. Legal issues with LLM’s need to be considered more heavily than I initially thought.
On Thursday, I tried to work on my ML project, while I understand the concept of machine learning reasonably well, I struggle heavily with understanding the code. I definitely need to do more further reading. In the afternoon, I had a 1-1 session where I got some help with understanding the code better, and I feel as though I understand many of the keywords (batches, epochs, etc.) better than I did prior. I also got help with some tweaks to my code (Relative to my dataset, my batch size was far too small, So I increased it.
Despite having a batch size of 200 and learning rate of 0.1 and 20 epoch, after 2 rounds of 20 epochs, (around 8 hours worth of training) , the results didn’t improve. (1.4% highest accuracy at the end of it, only fluctuating about ~0.5% accuracy). I think I either didn’t save my training property, (or loaded it incorrectly) or that my learning rate is too high (I set it at 0.1)