There are some research from my interests.
2025-08-09 ｜ #Words: 1287
Recently, I’ve been organizing some videos and realized I have too many. I wanted to shrink some videos that already needed transcoding. Previously, I increased the bitrate by 10-20% to compensate for transcoding losses, but when converting from H264 to HEVC (or other formats), it might not be necessary to increase the bitrate—you might even be able to reduce it.
2025-07-28 ｜ #Words: 4502
2025-07-27 ｜ #Words: 4252
Maximum Likelihood Estimation (MLE), also translated as maximum likelihood estimation, is a method for obtaining the optimal parameters of a model from sample data. For example, MLE can estimate the parameters of a probability density function (PDF). If a set of samples/datasets/statistics (likely) follow a normal distribution, we only need to find the mean and variance (or standard deviation) of the dataset to derive the probability density formula that fits the data.
2025-01-22 ｜ #Words: 720
Recently, while going through D2L (Dive into Deep Learning) code, I frequently noticed the "@" symbol appearing at the very top of classes or functions. Although Chapter 2 explains the comment `#@save` (indicating the following function is pre-defined in the package and can be used directly without redefinition), it does not explain the usage of the "@" symbol itself.
2025-01-21 ｜ #Words: 319
While reading a book today, I noticed that gradients are zeroed out at the end of the optimization algorithm. But the book didn’t provide an explanation. After checking, I found many optimizers also include this operation—why is that?
2025-01-19 ｜ #Words: 195
Today, while translating a blog post, I noticed Jekyll wasn’t generating HTML files as it usually does. After troubleshooting for a while, I found the cause: Jekyll cannot correctly recognize files encoded in **UTF-8 with BOM**.
2025-01-18 ｜ #Words: 2335
I’ve been using Ollama on macOS for a while without issues, but recently noticed it was running on the CPU after switching to Ubuntu. At first, I thought it was out of VRAM—on macOS, Ollama automatically falls back to CPU if it runs out of memory. However, I found that Llama3.2 3B only uses 3GB of VRAM, which is far from exceeding limits. I checked the command line options but found no relevant settings, so I did some research.
2025-01-14 ｜ #Words: 280
When processing the returned JSON today, I encountered the following error: JSONDecodeError: Extra data: line 2 column 1 (char 134)
2025-01-14 ｜ #Words: 1484
This article uses an Ubuntu environment. The version of FFmpeg installed directly via APT on Ubuntu supports CUDA acceleration. This demonstration uses this pre-installed version—your experience may vary if you compile FFmpeg yourself or install it from other sources.
2025-01-14 ｜ #Words: 3274
This is mostly a personal note to avoid re-researching solutions to problems I’ve already solved. However, since it took me about 20 days to write this article from start to finish, I can’t remember many details—so I may update it in the future.
2025-01-05 ｜ #Words: 235
If you check the top of the error message, you’ll find the root cause: the lookup path points to a `.cache/` directory (e.g., `xxxxxxx/.cache/`). This error persists regardless of whether you use an absolute or relative path.
2024-11-05 ｜ #Words: 1534
2024-10-31 ｜ #Words: 1069
Dynamic Programming (DP for short) is a programming paradigm, not a specific algorithm. It breaks down a large problem into smaller subproblems through recursion—mathematically speaking, it can be expressed as a polynomial. This allows solving problems of different scales with minimal changes to the code (a key advantage of recursion).
2024-10-10 ｜ #Words: 539
Many online articles lack clear explanations of complement graphs, isomorphic graphs, and self-complement graphs. Isomorphic graphs have the same points and edges but different positions, with the same degree sequence. Complement graphs share the same vertex set, with edges in one being those not in the other, forming a complete graph when combined.
2024-07-26 ｜ #Words: 3724
I was extremely curious about why this happens. The `0` on 32-bit systems is easy to understand, but why do we get random numbers on 64-bit systems? You can probably make a rough guess—it’s related to memory addresses (weird random numbers almost always point to address issues.
2024-07-22 ｜ #Words: 2622
I first learned about Hamming codes a few yedars ago when I watched 3Blue1Brown’s video. I then looked into Hamming himself and thought he was brilliant. Later, I was impressed to discover in *The UNIX Hater’s Handbook* that Hamming was actually Brian Kernighan’s (co-author of the C language bible *The C Programming Language*, commonly known as K&R) mentor at Bell Labs and the third Turing Award winner.
2024-07-14 ｜ #Words: 4801
2024-07-12 ｜ #Words: 2160
2024-07-12 ｜ #Words: 1911
SMA (String matching algorithms) have always been my weak spot, and I happened to have a need for them recently, so I took the opportunity to catch up. String matching algorithms primarily involve identifying a substring (pattern, which is sometimes translated as pattern in other contexts, hence sometimes called pattern matching) within a larger body of text, such as finding a word in a text editor or a specific sequence in a DNA strand.
2024-04-22 ｜ #Words: 881
MSRs (Model-Specific Registers) are specialized registers designed for various experimental functions, including debugging, program execution tracing, computer performance monitoring, simplified software programming, and power control.
2024-03-02 ｜ #Words: 2734
How to clone system disk to a smaller hard drive or SD card on Raspberry Pi, including partition, set up and repair boot files.
2023-11-07 ｜ #Words: 922
Raspbian recently upgraded from Bullseye to Bookworm last month, bringing many changes. As of writing this blog, less than 30 days have passed, so there aren’t many reference solutions available—we can only rely on previous versions or figure it out ourselves. In the future, the official team may fix this issue, or new problems may arise. If I encounter any, I’ll write a new article and link it here.
2023-11-05 ｜ #Words: 1680
This blog shares some reviews and tests of Raspberry Pi 4B. ZhongUncle compares it with Mac mini 2018 i5 in many aspects.
2023-08-14 ｜ #Words: 2112
This article aims to provide a low-level explanation of APFS, the file system used by Apple devices. Because I noticed that there was a lack of documentation on how APFS works in details, so I do some research and share my findings.
2023-07-19 ｜ #Words: 2150
In hard disk performance, IOPS is a very important parameter. In terms of speed, this is the most important parameter to pay attention to, rather than the sequential read and write performance or random read and write performance that many people measure.
2023-06-23 ｜ #Words: 1431
BDMV is the format used by Blu-ray discs. This format cannot be played using playback software, and a hard disk player, which is a specialized device, must be used. But the most classic ffmpeg can convert it to other formats and maintain the same bit rate and definition, so that it can be viewed easily.
2023-06-10 ｜ #Words: 2404
Recently, I found I was not very clear about the difference between GNU GCC and Clang. It affected implementation and learning somethings, so I spent free time in some days to research it carefully.
2023-05-31 ｜ #Words: 677
Recently, I saw some live broadcasts on Douyin that were really interesting, and I wanted to record them. So I studied how to record these live broadcasts. There are two methods: recording the screen and downloading from the push source.
2023-04-21 ｜ #Words: 2545
I encountered many problems when using VM Fusion to install Plan 9 on Mac. The official documents and some foreign blogs from the past two years were not clear, and even misleading (some situations have changed). So I am writing this article to help others who also encounter the same problem.
2023-03-16 ｜ #Words: 1123
ZhongUncle records what I find when using New Bing (GPT-4). Today New Bing called Bing Chat and can do more thing, but its mistakes are more often.
2023-02-17 ｜ #Words: 570
ZhongUncle records some performance of NN and know why bandwidth of gpu is important for Machine Learning.
2023-02-09 ｜ #Words: 755
I recently borrowed an A7R. This is my first time using a professional camera (not a card camera), and it is full frame tensor. So I spent some time to learn how to use it and how it works under various settings.
2023-02-03 ｜ #Words: 4222
When I was organizing my disk, I have a question: What is GIFs classified as? Static image files or video files?Because if you put it in a video, GIF is classified as a static image usually and is composed of images; if you put it in an image, it shows animations. Then I have new problem: Why GIFs shows animations, but be classified as static images?
2023-01-10 ｜ #Words: 512
Recently, I need to learn Java. Due to personal habits, I just want to use the terminal for development instead of using an IDE. But I never expected that I would be tricked by the official website once, and the downloaded Java could not be developed normally. So I wrote this article to explain how to configure it. It's very simple, just find the right official website.
2022-12-06 ｜ #Words: 797
Maybe you like me have a question: Why heap sorting using a binary tree, but it is called heap sorting, rather than tree sorting? Because the predecessor of heap sort is called Tree Selection Sorting, it also uses a tree structure, but is slightly simpler. Finally we will talk about why invent heap sorting to replace tree sorting.
2022-10-10 ｜ #Words: 1857
If you are studying or engaged in data analysis-related, you must have used or been told to use R language. But with so many languages for data analysis, why use R? Because R can output very nice charts for publication. It may not be necessary for blogs, but it is necessary for publications such as papers and books. After all, the charts in most publications are not in color or flashy.
2022-10-04 ｜ #Words: 2899
When I was learning quicksort algorithm, I found there are many style of code. I was very curious why. So I went to read Introduction to Algorithms to search for answer. Finally I found the reason be related to the development history of quicksort.
2022-09-03 ｜ #Words: 456
There are two kinds storage structures in data structures that are easily confused, namely index storage structure* and hash storage structure (hash storage structure) You can find corresponding keyword according to address. It can be understood as a yellow pages. You can find someone's phone number based on name. So index storage is also called direct addressing
2022-06-16 ｜ #Words: 2578
Do you wonder why Type-C in difference protocols is in difference speed? I will tell you why the Type-C have different bandwidths under different protocols from pin definitions, protocols, etc.
2022-02-13 ｜ #Words: 1425
Recently, I plan to use the open source and free SQLite as a database to save money. And due to personal habits, I prefer the command line, so I record how to use SQLite in command line.
2021-12-21 ｜ #Words: 1415
Recently, I needed to convert MKV files to MP4 format. Since Apple's Compressor doesn't support MKV, I decided to use FFmpeg. However, the video files were extremely large, and basic commands were too slow—so I wanted to use hardware acceleration. Due to macOS's uniqueness, the hardware acceleration method provided on the official FFmpeg website doesn't work. After some research, I put together this quick tutorial in hopes of helping others in need.