DOCSEARCHbyAlgolia
Submit Docs

Knownoevil Study Notes

website·Indexed 29 days ago

https://shyu216.dpdns.org/knownoevil/

A collection of academic notes and research documentation covering computer science courses from CUHK and UniMelb, including CSCI1130 (Java), CSCI4180 (Cloud Computing), and COMP90014 (Genomics). The site also documents research on HoloLens 2, Unity projects, point clouds, and motion magnification.

Tokens
143.9K
Snippets
296
Records
1K
Agent score
50%

What's inside Knownoevil

  1. Overview of Network and System Security Components

    Defensive infrastructure and system security concepts include:

    • Firewalls: Devices or software that monitor and control incoming and outgoing network traffic based on security rules.
    • IDS (Intrusion Detection System): Systems that monitor network or system activities for malicious activities or policy violations.
    • DMZ (Demilitarized Zone): A physical or logical subnetwork that contains and exposes an organization's external-facing services to an untrusted network, usually a larger, unfiltered network such as the Internet.
    • Secure Protocols: Implementation of IPSec, TLS, and PGP for secure communication.
    • Wireless Security: Standards such as WEP, WPA, and WPA2.

  2. Overview of IERG4300: Large-scale Information Analysis

    IERG4300 (Large-scale Information Analysis) is a core course for Information Engineering and Computer Science students. It focuses on processing massive datasets and analysis techniques, specifically covering distributed computing frameworks such as MapReduce and Spark. The course serves as a bridge between foundational knowledge in distributed systems and databases and advanced applications in data science and machine learning.

  3. Overview of the Global LLM Landscape (May 2026)

    As of May 2026, the LLM market has entered a multi-polar competition era with a rapid release cycle (multiple models per week). The performance gap between top US and Chinese models has narrowed to approximately 2.7%. Key trends include the transition of reasoning capabilities from an optional mode to a default architecture in mainstream models and a heavy focus on 'Agent' capabilities (the ability of models to perform actual work/workflows).

  4. Overview of CSCI3160 Algorithm Design and Analysis

    CSCI3160 is a core computer science course focused on designing correct and efficient algorithms and analyzing their time and space complexity. It serves as a foundation for technical interviews, competitive programming (ACM, ICPC), and advanced fields like machine learning and cryptography.

  5. Overview of Lua for embedded scripting and game development

    Lua is a lightweight, high-performance dynamic scripting language designed primarily for embedding into applications to extend their functionality. It is widely used in game development (AI logic, scripting), embedded systems (firmware, IoT), and web infrastructure (Nginx extensions via OpenResty).

    Key technical advantages include:

    • Small Footprint: Core library is approximately 200KB.
    • Dynamic Typing: Variables do not require type declarations.
    • Concurrency: Supports collaborative multitasking via coroutines.
    • Metaprogramming: Uses a metatable mechanism to provide object-oriented capabilities.
    • C Integration: Features a concise C API for seamless integration with C/C++ code.
    • Memory Management: Includes automatic garbage collection.

  6. Overview of ESTR1006 Engineering Multivariable Calculus

    ESTR1006 (Engineering Multivariable Calculus) is a continuation of MATH1510, extending calculus from single-variable functions to multivariable functions. It provides the essential mathematical foundation for understanding physical field theory, fluid mechanics, and electromagnetism. The course is specifically designed for science students within the Faculty of Engineering (ESTR series) to provide advanced mathematical tools required for physics and engineering coursework.

  7. Overview of Global LLM Landscape (September 2026)

    As of September 2026, the LLM market has shifted from weekly releases to daily leaderboard updates, with over 580 models released in Q2 and Q3. Key trends include:

    • Agentic Evolution: Models have transitioned from simple text/code output to 'full-stack agents' with native browser access, terminal control, tool calling, long-term memory, and self-hosted sub-agents.
    • Performance Convergence: The performance gap between top US and Chinese models has narrowed to approximately 1.4%.
    • Edge AI: On-device and edge models are now first-tier contenders due to advancements in mobile NPUs and local inference frameworks from Apple and Qualcomm.

  8. Overview of Formal Languages and Automata Theory (CSCI3130)

    CSCI3130 is a core theoretical computer science course that explores the limits of computation. It provides the theoretical foundation for compiler design, computational complexity theory, and formal verification. The course is structured around four primary areas: Finite Automata (DFA, NFA, Regular Expressions), Context-Free Grammars (CFG, PDA), Turing Machines (Computability Theory), and Computational Complexity (P vs NP).

  9. Overview of CSCI3251 Engineering Practice Course

    CSCI3251 - Engineering Practice is a practical course offered by the Department of Computer Science and Engineering at the Chinese University of Hong Kong, designed to be taken in conjunction with CSCI3250. The course focuses on providing students with hands-on experience in computer and engineering skills, emphasizing practical application, team collaboration, and version control.

  10. Overview of LangChain Core Concepts

    LangChain is an open-source framework for building LLM applications by standardizing LLM capabilities into composable modules. Key components include:

    • Model I/O: Unified interfaces for various LLMs (OpenAI, Anthropic, Gemini, Ollama) and PromptTemplate for structured prompt management.
    • Chains: Pipelines that sequence steps (e.g., Prompt → LLM → OutputParser). The LangChain Expression Language (LCEL) uses the | operator for declarative composition, supporting streaming and parallelism.
    • Memory: Context preservation for conversations using buffers, sliding windows, or vector summaries, with persistence options like Redis or PostgreSQL.
    • Retrieval-Augmented Generation (RAG): A pipeline consisting of DocumentLoader → TextSplitter → Embedding → VectorStore → Retriever, supporting stores like Chroma, Pinecone, FAISS, and pgvector.
    • Agents & Tools: LLMs that autonomously decide which tools to call (e.g., Google Search, Python REPL, SQL) using strategies like ReAct or Tool Calling.
    • LangSmith: An observability platform for tracing LLM calls, managing prompt versions, and running regression tests.
    • LangGraph: A framework for building stateful, multi-actor workflows using a graph structure (nodes as functions, edges as conditional jumps), supporting loops and human-in-the-loop interaction.

  11. Overview of XFeat for Lightweight Image Matching

    XFeat is a lightweight Convolutional Neural Network (CNN) designed for efficient local image feature extraction and matching. It is specifically engineered for resource-constrained platforms (e.g., mobile robots, AR devices, and low-end CPUs) and is hardware-agnostic, meaning it does not require specialized hardware optimizations to achieve real-time performance.

    Key capabilities include:

    • Sparse Matching: Keypoint-based matching suitable for Structure-from-Motion (SfM) and visual localization.
    • Semi-Dense Matching: High-density matching using a novel match refinement module that derives pixel-level offsets from coarse local descriptors, making it effective for relative camera pose estimation in texture-poor scenes.
    • Performance: Up to 5x faster than other lightweight deep learning local feature alternatives while maintaining accuracy comparable to larger models like SuperPoint and DISK.

  12. Overview of Lensless Computational Optical Imaging

    Lensless imaging replaces traditional optical lenses with a mask to capture image information. The core process involves designing a mask, capturing sensor data, and using reconstruction algorithms to invert the captured information back into a real image. Key approaches include:

    • Fresnel Aperture Coded Imaging: Uses precise modeling of a coded mask and deep learning for decoding and reconstruction.
    • Holographic Imaging: Utilizes phase encoding of the coherent diffraction propagation process, employing algorithms that combine physical models with prior-knowledge fine-tuning for reconstruction.
    • Hologram Twin-Image Removal: Techniques to eliminate twin images, including single-measurement compressed holography, multi-measurement physical encoding, and off-axis interference signal filtering.