haotianblog

haotianblog

This is a bilingual technical blog for algorithm notes, programming write-ups, project logs, and downloadable companion materials. The goal is to publish durable references that remain useful after the initial work is done.

How to enter the site from a problem

Explore next

• Network Fundamentals Visualized from DNS, TLS, and HTTP/3 to proxy tunnels, load balancing, and shared caches with reproducible labs
• Battery Modeling for AI Data for reading PyBaMM, EISSimulation, aging simulations, and trainable labeled datasets
• AI Security Engineering from threat modeling to adversarial examples, poisoning, privacy, and LLM/RAG/Agent controls
• Deep Learning / CNN tutorials from neural network basics to CIFAR-10 Tiny CNN image classification
• Machine Learning From Scratch for K-means, Iris, feature processing, and evaluation
• Algorithm Visualization for eight queens, backtracking, bitmasks, and the playground
• Animation Asset Lab for turning local character images into SVG, layered PNGs, and an asset manifest
• Student AI Projects from a handwritten digit C classifier to browser demos and CNN work
• Share Center with tracked direct links, copy-ready blurbs, and social platform actions
• Tools for browser-side image conversion into truecolor Unicode quadrant ANSI files
• Algorithm playground for running the eight queens search, Iris K-means, and handwritten digit demos
• Knowledge map for connecting projects, posts, resources, and learning routes
• Projects for timelines, topic hubs, and the next planned notes
• Downloads for code, data, and diagrams
• Transformer Attention Math for hand-calculating Q/K/V, softmax, masks, multi-head attention, and KV cache
• HTTP/2, HTTP/3, and CDN Cache Waterfall comparing protocol overhead, cache state, and request timing
• Data Poisoning and Backdoor Defense covering poison rate, trigger behavior, attack success rate, and pipeline controls

What this site focuses on

• Algorithm and data-structure articles with implementation details and debugging notes
• C and Python code walkthroughs that explain why the implementation works
• Project notes covering deployment, maintenance, and site-building decisions
• Downloadable code, diagrams, datasets, and supporting files that belong to each article

Where to start

The current site is strongest on technical blog posts and downloadable materials. Start here if you want concrete examples:

• Network fundamentals series calculating request paths, proxy boundaries, queues, and cache state with code and figures
• Battery modeling for AI series using PyBaMM to generate traceable aging, impedance, and SOH/RUL label examples
• AI foundations series from core concepts to a Python classification practice
• AI security series connecting NIST, MITRE ATLAS, OWASP, robust evaluation, privacy, and safe toy labs
• Handwritten digit project series from CSV structure and the C softmax classifier to the browser playground
• Convolution and Receptive Field Math deriving output size, receptive fields, padding, stride, and im2col layout
• Transformer Attention Math from single-head attention arithmetic to multi-head structure and KV cache
• Tools for downloadable .ans output that can be displayed with cat in a truecolor terminal
• Animation Asset Lab as the first step in 2D animation asset preparation
• Share Center for direct links you can send to classmates or readers
• Downloads for code, data, and diagrams
• K-means clustering on the Iris dataset in C
• Eight queens with classic backtracking
• Bitwise optimization for the eight queens problem

Publishing direction

The Chinese side currently contains the most complete long-form articles. The English side is being expanded into a clearer landing surface for summaries, project context, and future translated or original posts.

Contact

If you find problems in articles, code, downloadable resources, or bilingual links on Haotian Blog, email haotianblog@gmail.com.

For a reproducible report, include the page URL, the exact location, the error message you saw, or the environment used to run the experiment. For code issues, include the operating system, Python or C compiler version, command, and the relevant output snippet when possible.

Useful Feedback

• Formula, code, protocol explanation, or experiment output errors in an article.
• Broken download files, diagrams, sample datasets, or resource links.
• Incorrect Chinese-English page pairing, or an English summary that does not match the Chinese article.
• Missing boundary notes in machine learning, algorithm implementation, network protocols, AI security, or battery modeling content.

How to Send a Technical Correction

The most useful message is specific enough that the issue can be reproduced without guessing. For an algorithm article, include the input, expected output, actual output, and the step where the trace first diverges. For a C example, include the compiler, command line, operating system, warning text, and whether the behavior changes with optimization flags. For a Python or machine learning article, include the Python version, package versions, random seed if relevant, data shape, and the smallest code fragment that reproduces the problem.

For network protocol notes, include the request URL, status code, relevant headers, local environment, and whether the result came from a browser, curl, a reverse proxy, or a local simulation. For AI security material, keep reports defensive and bounded: describe the toy dataset, local script, metric, and mitigation question rather than sending instructions against a real third-party system. The goal is to improve the public explanation, not to exchange unsafe operational instructions by email.

Content Priorities

Corrections that affect reproducibility, safety boundaries, or reader trust are handled before cosmetic requests. Examples include code that no longer runs, a broken downloadable file, a diagram that contradicts the text, a metric label that can be misread, an incorrect bilingual link, or a security lab that needs a clearer disclaimer. Requests for new topics are useful when they include a concrete learning problem, such as a confusing compiler error, a machine learning evaluation trap, or a networking behavior that is hard to observe directly.

Not every request will become a new article. A topic is more likely to be added when it fits an existing route and can be supported by original explanation, runnable material, or a reproducible observation. General requests such as “write about AI” are too broad; focused requests such as “compare K-means initialization effects on Iris SSE” or “explain why an HTTP cache revalidates with ETag but not with Last-Modified” are easier to turn into durable site content.

What Happens After a Report

Reports are reviewed as editorial maintenance work, not as private support tickets. If the issue is confirmed, the public page may be corrected, expanded, redirected, marked noindex, or moved behind a clearer disclaimer depending on the problem. A small typo may be fixed silently, while a reproducibility issue normally requires checking the code path, command, dataset, and explanation together.

For content-quality reports, the preferred outcome is a stronger public page. That may mean adding a missing assumption, documenting a failed run, explaining why an experiment is only educational, or linking a hub page to a more complete article. The site does not treat a thin page as solved merely because it has more links; the page should help a first-time reader understand what the material is for and how to evaluate it.

If a report affects several pages, the fix may be applied to the hub, the article, the download description, and the navigation text together so the same confusion does not reappear in another route.

Response Expectations

Because the site is maintained as a technical archive, response time depends on the type of issue and the evidence included. A broken public link or incorrect bilingual route can often be checked quickly. A code correction, machine learning metric issue, or network protocol claim may require rerunning a command, checking a dataset, comparing article text with source files, and updating related hub pages.

If you do not receive a direct reply, the report may still be used for maintenance. Public corrections are usually more important than private correspondence. When a correction is accepted, the visible result should be a clearer page, a stronger disclaimer, a fixed download, a redirected weak page, or a better route through the related material.

Signals That Help Improve a Page

When you send feedback, describe the reader problem rather than only the surface symptom. For example, “the page is too short” is less useful than “the page links to a dataset but never explains the label column, split strategy, or expected output.” A good report identifies the public URL, the confusing claim, the missing evidence, and the type of reader affected.

The same rule applies to bilingual issues. If the English page is only a route summary while the Chinese page carries the full derivation, say which concept or example is missing from the English side. That makes it easier to decide whether the page needs translation, a stronger summary, a canonical link to the article, or removal from indexing.

Useful reports often include a suggested verification step. For a broken code block, that might be the exact command that should pass. For a networking article, it might be the expected header, status code, or packet field. For a machine learning page, it might be the metric, split, or random seed that should be visible. This keeps the discussion tied to evidence instead of opinion.

Collaboration and Boundaries

haotianblog is open to technical corrections, references to primary documentation, reproducible examples, and suggestions for clearer bilingual terminology. If you want to reference an article in class, a project note, or a personal learning plan, attribution and a link back to the relevant page are appreciated. If you notice a resource that appears to contain private, copyrighted, unsafe, or incorrectly attributed material, report the exact URL so it can be reviewed.

Email is not a support channel for account recovery, private infrastructure administration, bypassing platform controls, attacking systems, or completing homework on someone else’s behalf. The site can discuss defensive engineering, academic learning, and reproducible toy experiments, but it will not provide help that turns educational material into unauthorized activity or a substitute for independent academic work.

Response Scope

This site prioritizes the quality of public technical content. Feedback that affects reproduction, understanding, or safety boundaries is handled first, such as code that no longer runs, unavailable resource links, inaccurate terminology, or security experiments that need clearer limits.

The site does not provide production operations outsourcing, access-control bypass assistance, attacks against real systems, homework completion, or unauthorized data-processing services by email. Technical discussion should stay within education, academic research, personal learning, and defensive engineering practice.

If your message is about an AdSense, SEO, or site-quality concern, include the exact public URL and describe what a first-time reader cannot understand from the page. Useful reports focus on content value: missing explanation, unclear navigation, weak translation, duplicated wording, thin context around a download, or a page that looks like a tool without enough editorial framing. Those reports directly help decide which pages should be expanded, merged, or kept out of search indexing.

If the issue is urgent because it affects privacy, safety, attribution, or public trust, put that concern in the first sentence so it is not mistaken for a general topic suggestion.

For privacy or takedown concerns, identify the material, the reason it should be reviewed, and the ownership or safety issue involved. Do not send sensitive credentials, private datasets, access tokens, personal identity documents, or exploit targets by email. A short description and the public URL are enough to start a review.