Welcome! This is the first issue of my newsletter. I am still figuring things out but thought it would be a nice way to share new things I found, as well as some of my thoughts on academia, my PhD journey, machine learning and NLP.
Last week I started writing my thesis. Scary task, but surely, if I am consistent I would get there. Well, so far the progress is slow, with about 10 hours of work on the thesis, I successfully removed 142 words from the initial version of the thesis. This doesn't show the full picture, as part of the efforts are related to structuring work, rewording, editing and paraphrasing, etc. I aim at having a bigger update in February, with more stats incoming in this section! In the meantime, I spent some time working in lovely coffee shops around Cambridge (reach out if you would be interested in my cáfe tier list). Also, I am trying to incorporate exercise before writing, which lead me to rowing in some freezing weather (if you zoom in on the picture below, you could see ice forming on the oar).
After I started the final push to finish the PhD and started working more than 40h/week, I noticed that I may have lost some mental agility. To measure this, I play about 2-3 games of blitz chess a day - looking at my ELO score, it seems that it is indeed not going well. Once I am done with my PhD, I will have to rethink my chess career!
In this section, I share papers you would unlikely find on your Twitter timeline, as they are not necessarily related to the current research hype. In other words - this section is dedicated to the odd papers, cool results, insightful comments, and unpopular opinions. Also, it's a highly subjective list of things I found curious based on my current research and work.
Question Asking During Tutoring, by Arthur C. Graesser and Natalie K. Person, 1994. On the importance of asking good questions in a teaching environment.
Large Dual Encoders Are Generalizable Retrievers, Jianmo Ni, Chen Qu, Jing Lu, Zhuyun Dai, Gustavo Hernández Ábrego, Ji Ma, Vincent Y. Zhao, Yi Luan, Keith B. Hall, Ming-Wei Chang, Yinfei Yang. Very decent sentence embeddings. Tried them on a few tasks and they performed quite well. Interestingly, the paper is not super popular and based on the huggingface website, the model is not downloaded that often. If you are doing some sort of text classification, or analysis, make sure to give them a try (and cite the authors). https://huggingface.co/sentence-transformers/gtr-t5-xl
Seeing through walls with Wi-Fi signals? The first time I heard about this, I thought this can't really work, can it now? Turns out - this is an active area of research, and it seems to be working quite well. And in terms of timing - it is here, probably a few years away (or much less) from hitting mass-scale applications. Last month, I found this paper, which seems like a strong contribution in that direction: (https://arxiv.org/pdf/2301.00250.pdf). There is also a group in CSAIL@MIT that seemed to work on this a while ago: https://people.csail.mit.edu/fadel/wivi/.
The science behind it is fascinating, it was an interesting read for me. That said, the elephant in the room is - can this be used for unethical applications? And in my opinion, the answer is yes. Imagine living in a world, where your router can detect how many people are there, in what room, and possibly what are you doing. Certainly, this wouldn't be abused by corporations, surveillance agencies, and governmental organisations./s
Leaving the gloomy attitude behind, together with these technological breakthroughs, another area of research would be monitoring, regulating, and counteracting this.
This section would include some neat tricks that I learnt about in the past month. These tricks are not necessarily advanced or obscure, but I found them fun and useful!
If you are using Python >= 3.10, you can do Type aliases like this:
from typing import TypeAlias
Factors: TypeAlias = list[int]
This is quite cool and could be very useful for improved code readability and expressiveness, while not writing 10 classes for boilerplate code.
Thank you all for joining my mailing list. If you feel like someone would benefit from the mildly educational content here - feel free to forward them the mailing list.
I will be experimenting with what topics might be interesting, so if anyone has any comments or suggestions - do reach out! In the meantime - I wish you a happy February!
PhD project: https://www.delibot.xyz/