Tuesday, October 26

[soid]

 Woke up around 9:15am, strangely fresh despite sleeping less than 5 hours. Strange thing, I can sleep 8 hours and wake up tired, and 5 hour and wake up fresh. An apricot danish with coffee, shower, and I went to the lecture by 10.

10-11:30am Middle East lecture. Moving on to World War II. Main outcomes: establishing a new world order that lasted until 1990; nation self-determination is a key principle of new order; US leads, Europe fades; the world is dominated by the US and Soviets. At the start of the war in August 1941 the US and Britain set their goals after the war in what is called the Atlantic Charter: no territorial gains after the war, principle of self-determination, and other. These principles laid out the foundation of the Declaration of United Nation, signed by a wider range of allied countries on January 1, 1942. Question: why the same Churchill, who rejected Wilsonian self-determination 20 years earlier, now agreed to it with Roosevelt? Answer: he had no choice, he desperately needed the US to enter the war. Self-determination is explicitly laid out in the charter of the UN. The UN starts deconstructing colonialism – with the help of self-determination principle. However, the action goes on for decades. Some countries gain their independence only in the 1970s.

2pm Office hours with Lelia. I asked about the comparison prompt ("Compare causes and outcome of Iranian and Ottoman constitutional revolutions"), about how to write something meaningful and interesting in this form and not just list differences and commonalities. She said comparison is only a tool, but this tool should be used to deliver a wider message. This message in turn can be found by looking at facts and finding a question about them, then thinking what could be the answer that becomes the message of the paper, use the comparison tool to support this message. Sounds great – I was reading the prompt too literary, as this comparison was the goal by itself. But too late, the paper is due tonight.

2:40-4pm Topology. Peano curve for constructing a surjective continuous function from I to I^2. It's like Hilbert curve, but Peano curve is harder to construct and easer to work with in proofs. Interesting. I've seen Hilbert curve is useful for linearizing a picture, so the same 2D neighborhood is within the same neighborhood on the linearized 1D scale. I think it's used in JPEG. The Peano curve is not injective: some 2D points are mapped from multiple 1D points. The proof of subjectivity uses complete metric spaces and sequences of functions. Interesting proof too – example of when complete metric spaces are useful (compact metric spaces are complete; then Peano curve represented as a sequence of functions is Cauchy, converges to I^2). Song was lecturing instead of the prof. Actually, he's great at explaining things. Students are often better than professors at explaining things.

5-6pm Middle East discussion group. To think about: all the history books that we read in the class are written by men because women are too busy to write books, apparently. 

Read another paper for NLG in the evening: Image Captioning (ClipCAP). Instead of glueing image and text neural networks and fine-tuning (most popular approach until last year), they extract latent features from images, run an attention layer to find meaningful features, and then convert it to transformer tokens. Strange feeling about this class. Reading some interesting ideas, but need to practice them. But no way I can practice all of them. Have to focus on a few, and other things don't really matter that much. It's good to see one or two novel ideas in each paper, but I shouldn't worry too much about getting too deep in each of them.

Feeling tired early, bed around 11pm.