Something that stood out in my meeting with Professor Santiago Segerra was the new strategies he suggested for expanding and testing my project on the Thucydides Trap. Since my sample size was so small, he mentioned that I should test my curve by removing several points to see if their outcomes could be predicted. Although, since the survival risks change over time, the chance of a war breaking out at any given time is more fluid. Another potential issue I could run into with testing is that since the risk changes from each month to the next, calculating a "real" risk of war at any given point in history is difficult. The reason is there are many factors at play and most sources have the bias of knowing the result of conflicts.
We also discussed authorial attribution, an area that Professor Segerra had stopped work on several years before. The idea of using machine learning to recognize the style of a piece of literature and identify its author fascinated me, especially because Professor Subramanian mentioned earlier that it is difficult for machines to convert words and equations back and forth.
Overall, my main takeaways were 1) machine learning is still very much in development; 2) tasks that seem extremely complicated can actually be very basic in the grand scheme of things, especially in a world such as machine learning. I also realized that I have not considered many methods of model testing previously, because they are usually not very effective when working with small data sets. Now hearing Professor Segerra's strategy, I am thinking about using it to test my project's prediction capability.
Σχόλια