A Little Bit of Everything

26 Jan, 2026

want to actively gather information, focusing on the "problem to be tackled"rather than the theories themselves.

I think the problem is that jargons are often heavily used. Few people would talk about the problems without being too technical. So finding explanation of the problem and the related tools often requires delibrate searching, or resorting to llm chatbots.

pole/zero and response

How to relate the pole/zero location and the response of a system? We now see a few examples of this engineering method.

real pole

If the poles are real, then the response is either eventually decaying or unbounded. Intuitively, real poles do not contribute to the frequency.

Next, we inspect pole at the origin.

ffmpeg

2026-01-09 I tried to burn the subtitle using ffmpeg but failed to do so. it might have to do with my version of ffmpeg

Thi is a text file for recording my journey of learning latin.

I am desperate finding job. Having failed the master entrance exam, I am deperate wanting to prove myself in the latin exam in May.

Instead of staring at texts, i should assess myself more. e . g . do the questions without refering to declension tables.

So, put it plainly, my way of learning latin is drilling. There is nothing special. But using grammar drills as a benchmark.

After grinding declensions for several days, I find that i actually enjoy reading texts from Latin From Ovid. It also helps solidify the declensions.

I need to try hardr in order to get over the current stage.

learn ml

I think reading is way demanding than doing exercises. I haven't seen anyone who discussed this before. Reading is more information dense, requiring the reader to process information as fast as the book delivers.

Today, having slept only 5 hours, i can hardly handle reading.

After the cafeine intake, I feel more alert and productive.

My goal is to complete making a linear model in two weeks(today is Dec 28 2025). slow or not, this is my currently planned pace of learning. I think that the difficulty or the type of a problem does not matter, and it is worth doing as long as I can learn something through working on it. and it does not matter how i figure it out.

A week later

It is Jan 4 2026 now. A week has passed. I have not touched linear models, or anything related to machine learning ever since.

I have not been intrigued much, although ski-py brought my attention. On the other hand, algorithm questions on leetcode and computational aspects of algebra indeed slightly intrigues me recently.

Having ideas in mind, I have not actually initiated to implement them in code.

By the way, I need to be used to type 2026 when talking about the current year. Also I try to get used to type 0 instead of ^ to navigate to the start of a line in vim. I consider prioritize Leetcode, not having firm decisions.

Fermat's SOS representation of primes

The answer is cornacia algorithm. A subalgorithm of which is solving $r^{2} \equiv - 1 (\mod p)$ , which is sufficient for our purpose.

So i plan to look at how cornacia solves this equation. This is not needed. It can be done in finite steps.

The remaining step is to do the extended euclidean algorithm on $(r, p)$ , stopping once $r_{k}^{2} < p$ .

I used to think understanding is more important. But now I realize that you do not need a thorough understanding to take actions.