工程实践与系统学习

Wiki Note

wiki/工程实践与系统学习.md

工程实践与系统学习

[AI Synthesis] 由 5 份压缩摘要交叉孵化的新 L1 主题（mean conf 0.77，max 0.80）。

Core patterns

_(Incubated; evidence appended below.)_

Distillation (2026-06-16) - source: 2015-03-06-facerecognition_with_opencv

[Literal] This post summarizes approaches and ideas for face recognition using OpenCV, focusing on geometric features, Eigenfaces, and Local Binary Patterns Histograms (LBPH). >

[Literal] Geometric features utilize marker points like eyes, ears, and noses to build feature vectors based on distances and angles. >

[AI Synthesis] The core idea behind Eigenfaces is dimensionality reduction, assuming facial images can be represented in a lower-dimensional subspace capturing most of the variance.

Distillation (2026-06-16) - source: 2016-03-31-notes-on-android-sdk-dev

[Literal] Notes from sharing Android NDK workflow: glue C/C++ into shared libs via Android.mk and Application.mk.

[AI Synthesis] Emphasis is on linker mechanics and choosing static vs dynamic libraries.

[Literal] Android.mk defines per-module properties; Application.mk defines app-wide module properties.
[Literal] Good dev tooling matters—Eclipse called terrible for this task.

Distillation (2026-06-16) - source: 2019-04-07-interview

ML 面试基础：bias/variance 权衡；反向传播、dropout；word2vec predictive vs GloVe count-based。

Bias：欠拟合；Variance：过拟合→正则化。
NN 训练误差小但测试误差可能大；非线性激活；dropout 防过拟合。
word2vec 预测 context；GloVe 基于共现计数。
社交：返工前还要准备/学什么；System Analyst；物流供应链电商方向。 -

(Source: 2019-04-07-interview)

Distillation (2026-06-16) - source: HSMC

[Literal] The inventory covers professional competencies, general skills, health matters, and personal relationships.

[AI Synthesis] The structure suggests a self-assessment or goal-setting exercise.

[Literal] Hard skills identified include NLP + ML, programming + systems knowledge, and writing abilities.
[Literal] Soft skills and general goals listed include communication skills, fluency in Cantonese, improving English, learning to drive, swimming, music, and reading.

Distillation (2026-06-16) - source: work insight

[Literal] Achievement is defined as (Success Rate × Influence) × Speed.

[AI Synthesis] This suggests that sustained achievement requires optimizing the quality of execution alongside the scale of its impact.

[Literal] It is important to focus on the success rate of tasks undertaken, aiming for every action to yield some positive effect to build a foundation for larger future endeavors.

Distillation (2026-07-09) - source: 2017-05-19-nlp

Human language is a symbolic signaling system: words map signifier to signified; continuous encodings (voice/gesture/writing) open cognitive questions. Linguistic stack: phonetics/phonology (stress, schwa, intonation), morphology (morphemes), syntax, semantics (lexical), pragmatics (context). Word vectors: dense context-based reps (TF-IDF count-based vs distributed). Embeddings encode semantics via distributional similarity (Word2Vec: know a word from its neighbors). Language modeling: probabilistic word sequences

N-gram, automata/RNN, LSTM.

(Source: 2017-05-19-nlp)

Distillation (2026-07-09) - source: 2017-11-27-the-open-source-community

Unix standards reconcile APIs; on open-source Unixes, features are often engineered using published standards as the specification. IETF process is practice-driven with rigorous peer review (Internet-Draft to RFC to Proposed to Draft to Internet Standard). Specs as DNA, code as RNA: modularity favors scrap-and-rebuild; careful standardization captures best existing practice; prototyping and cycles of test/re-specification beat perpetual patching without a standard. Open-source implementations of published standards cut coding workload and give a forward-port path—practice defensive design, build on open source. Contributors are often volunteers rewarded by usefulness and reputation; process transparency and peer review are crucial.

(Source: 2017-11-27-the-open-source-community)

Distillation (2026-07-09) - source: 2023-10-06-programming-languages

Interpreter path: program to parse to AST to eval. Python Data Model: implement special methods so objects behave like built-ins (Pythonic). Sequences: mutable/immutable; flat (own memory: str/bytes/array) vs container (hold references: list/tuple/deque). list copies; tuple(t) may return same reference; array.array packs numeric bytes. dict/set driven by sparse hash tables (fast, not space-efficient). Variables are labels; function params are aliases. Functions as first-class objects enable functional style; decorators wrap functions. Duck typing; iterators vs generators (yield pauses and saves state). Context managers: enter/exit or @contextmanager. SML: syntax vs semantics; type-checking in static env then evaluate in dynamic env; idioms (recursion, let); immutable data = mapping not assignment; pattern-matching over one-of/each-of types; every function takes one argument (often a tuple pattern); type inference; tail-recursion reuses stack.

(Source: 2023-10-06-programming-languages)

Distillation (2026-07-09) - source: 2024-11-29-notes-on-llm

[AI Synthesis] Working mental model of LLM generation (sections mostly TODO): text to tokenizer to embedding lookup to transformer to unembedding to softmax to next token. Component stubs: Tokenization, Embeddings, Positional encoding, Attention, feed-forward network, next-token prediction.

(Source: 2024-11-29-notes-on-llm)

Distillation (2026-07-09) - source: docs

Tools I am using：

Rectangle / iTerm2 分屏；Spacemacs：pyenv/pyvenv、ipython REPL、pytest、grip markdown preview。
常用：SPC m V w（venv）、SPC m c c（exec file）、SPC m t t（pytest）。
goodbooks 锚点：The Lean Startup、The Great CEO Within、The Effective Executive。
(Source: docs)

Evolution

2026-06-16: Theme incubated from 5 compression file(s) (threshold: support≥3, mean≥0.75, max≥0.8).
2026-06-16: Distilled from raw source 2015-03-06-facerecognition_with_opencv.
2026-06-16: Distilled from raw source 2016-03-31-notes-on-android-sdk-dev.
2026-06-16: Distilled from raw source 2019-04-07-interview.
2026-06-16: Distilled from raw source HSMC.
2026-06-16: Distilled from raw source work insight.
2026-07-09: Distilled from raw source 2017-05-19-nlp.
2026-07-09: Distilled from raw source 2017-11-27-the-open-source-community.
2026-07-09: Distilled from raw source 2023-10-06-programming-languages.
2026-07-09: Distilled from raw source 2024-11-29-notes-on-llm.
2026-07-09: Distilled from raw source docs.

Sources