2025 Q1 Summary

Sat, 03 May 2025 10:56:16 +0800

New Insights:

Connect knowledge into a network rather than isolated nodes; forming a graph structure allows you to gain new insights from each of your own insights.
Use tools appropriately, don’t blindly pursue programmability and reusability. This is a habit developed as a software engineer, but when solving problems, tailor the approach to the specific issue.

Previous Phase

Below is a summary of Q1 activities from the perspectives of work, learning, and research.

Research

I originally planned to publish a conference paper in Q1 and follow up on some of the latest research progress, but I found that this task was not actually completed. I believe this was due to preparing the SWR paper. I supplemented a large number of experiments on SWR and rewrote much of the code that I previously considered unreliable. Submitted the SWR paper.
I refactored part of the meter-viewer content; preparing to reintroduce new methods for building more suitable datasets.
I reimplemented SWR based on Lightning.
Considered domain adaptation models for meter recognition problems.
Explored some active learning algorithms.

Work

Further design and foundational thinking on meterhub.
- Spent most of January improving meterhub.
- Feedback features.
- Login with email.
Released django-login-mail v0.6.1.
Started a new project: githubManager.
Upgraded the Hugo theme and update process for HIGH, using the stack theme and corresponding GitHub CI.

Learn

Experimented with OpenAI’s SDK and simple calls to LlamaIndex.
Revisited MAE.
Read CLIP code.
OpenManus and Docker.
ComfyUI testing.
MCP server.
CUDA-related problems…
Using OpenRouter.
Sharpened basic programming skills.
- Rethinking about functional programming.
- Where to Put Your Data Folder.md
- Using PDM: https://svtter.cn/p/dynamic-version-in-pdm.md/
Operations and maintenance.
- SSL certificates: https://svtter.cn/p/certbot-self-signed.md/

Plans for the Next Phase

Focus on writing the overall framework for the paper.
- Since I’m not someone who can sit still for long periods, I need to pay extra attention to staying focused.
Attempt to publish a conference paper; the level is not important.
Refine the outline of the graduation thesis and write the details.
Write a new paper.

2024 Q3 Summary

Mon, 14 Oct 2024 11:08:56 +0800

This is a summary for July 2024 to September 2024.

This quarter passed quite quickly; it doesn’t seem long since the last review. There truly is a gap in intelligence between people; I feel like I’m a not-very-hardworking fool.

When conducting a review, try not to modify the content of the journal as much as possible. Otherwise, it will be troublesome to trace back later.

Life #life

Most of my energy during this period has been focused on taking care of my pregnant wife. The child was born safely, and the mother is doing well and happy 😆. For friends preparing for pregnancy and childbirth, I recommend this book: Pregnant with My Wife. Pregnancy is quite challenging for both men and women. Influenced by hormones, emotions can often fluctuate easily, and the man needs to balance work and family well.

Financially, prepare sufficient funds and budget for family expenses in advance. If your job is not stable enough, I don’t recommend rushing to have children. Although the old saying goes, “Start a family before establishing a career,” starting a family doesn’t mean having children immediately. During pregnancy, the emotional needs of the woman will be higher, so be mentally prepared. It’s best to seek advice from slightly older friends or relatives. Also, pay special attention to the location for childbirth. The ideal situation is to have your own independent family. Otherwise, you might have to deal with a series of complicated matters, which can be exhausting. Adding work pressure on top of that makes it even more difficult. In short, if conditions aren’t right, don’t have children; don’t make things hard for yourself. If you’re unhappy, your family will likely be unhappy too.

Read Rich Dad Poor Dad 2 but couldn’t finish it. Made a small profit in stocks and cashed out; couldn’t keep up with the market trend anymore. In investment and financial management, too many people want to make quick money, so designing strategies directly aimed at quick profits can make money. Leverage is absolutely not advisable; it’s just used to deceive gamblers.

I wonder how the friend who borrowed money to buy stocks a couple of days ago is doing.

Research #Research

Although a new network was constructed, it couldn’t be transformed into a publishable paper.

Experiments and New Research

New experiments verified that CRNN is still the best method in certain scenarios; formed an article specifically discussing CTC. However, it’s not yet at the level for publication.

Tried many multimodal approaches, different encoding methods, word embedding, and one-hot. Spent a lot of time learning about transformers. Completed querynet, a new structure for solving multimodal problems. However, it still hasn’t solved the problem I proposed.

In terms of mid-term progress, the content is insufficient. Although I’ve done many experiments now, relatively few can be solidified into theories, and the overall coherence is lacking. This is not very consistent with the original plan.

Regarding submissions, just received feedback on the new paper SWR; the new paper has been rejected and is still in rebuttal. But overall, it’s still okay.

I feel that the current research progress is fine, but there haven’t been significant achievements.

Some Thoughts

If researching a field, spend time exploring, organizing materials, and increasing understanding of the industry.
Organizing thoughts #thinking

Extensively adopted functional programming
Deeply explored the React framework, understood useState, useEffect, and fixed many bugs in the framework.
Deeply explored react-router.

Developed a relatively universal front-end and back-end underlying framework. Front-end based on React, back-end based on Django Ninja.
- After so many years of development, I’ve finally solidified something. By using this framework, many problems brought by the native Django and React frameworks can be avoided. This is a kind of account for all these years.
- The framework still has many areas for improvement, with the biggest help being efficiency improvement. If there’s an opportunity, it will be open-sourced.

Next Quarter

Adjust SWR well, submit it, and strive to get one article accepted.
Find a way forward for querynet and the new research problem proposed, respectively.
Launch meterhub.

On top of completing the above, finish the projects at hand and then organize the graduation thesis.

Summary on Svtter's Blog