DOCSEARCHbyAlgolia
Submit Docs

xbZhong's Technical Knowledge Base

website·Indexed 29 days ago

https://xbzhong.cn/

A comprehensive collection of technical notes and study materials covering Deep Learning (PyTorch, CNN, LLM, Word2Vec), Backend Development (JavaSE, SpringBoot, SpringCloud, Redis, RabbitMQ, MySQL, Django), Go programming, Data Structures and Algorithms (STL, Balanced Trees, Recursion), and academic course notes on AI chips, embedded systems, and brain-computer interaction.

Tokens
179.9K
Snippets
589
Records
1.1K
Agent score
50%

What's inside xbZhong

  1. Overview of Brain-Computer Interface (BCI) System Architecture

    A standard BCI system generally consists of three main parts: signal acquisition, signal processing (decoding), and command output. The system can be categorized as either an 'online' system (real-time interaction) or an 'offline' system (post-processing of recorded data). A 'closed-loop' BCI system provides feedback to the user, offering a core advantage over 'open-loop' systems by allowing the user to adjust their brain activity based on the system's output.

  2. Overview of Brain-Computer Interface (BCI) Systems

    A Brain-Computer Interface (BCI) system is divided into three primary components: signal acquisition, signal processing, and command output. BCI applications generally fall into five categories: replacement, recovery, enhancement, supplementation, and improvement. Systems are further categorized as synchronous or asynchronous. The stability and practicality of a BCI depend on whether a key brain region can adaptively optimize the action signals it emits. The two adaptive controllers in a BCI system are the central nervous system and the BCI system itself.

  3. Overview of Brain-Computer Interface (BCI) Principles

    A Brain-Computer Interface (BCI) is an information interaction and control pathway established directly between the human brain and an external electronic device, bypassing the peripheral nervous system and muscle tissue. The core objective is to capture neural activity signals, process and decode them into control commands, and enable direct manipulation of external devices.

  4. Overview of Word2Vec Architectures

    Word2Vec is used to compute word vector matrices to obtain word embeddings. It utilizes a sliding window to maintain the relationship between a center word and its surrounding context words. The word embedding vector is calculated as $v = W \times one\text{-}hot(v)$, where $W$ is the input weight matrix.

  5. Overview of model compression optimization goals

    Model compression aims to optimize three primary areas: reducing the amount of memory-intensive data, improving the time required to retrieve model parameters, and accelerating the overall model inference time.

  6. Overview of Artificial Intelligence Schools of Thought

    Artificial Intelligence is generally categorized into three primary schools of thought, each with a different approach to simulating intelligence:

    1. Symbolism (符合主义): Based on mathematical logic. It views human cognition as the manipulation of symbols. The goal is functional simulation—constructing systems that simulate the functions of the brain.
    2. Connectionism (连接主义): Based on bionics. It views cognition as the activity of interconnected neurons. The goal is structural simulation—constructing neural networks that mimic the brain's physical structure.
    3. Behaviorism (行为主义): Based on cybernetics. It posits that intelligence depends on perception and behavior and the ability to adapt to complex environments. The goal is behavioral simulation—constructing systems with evolutionary capabilities.

  7. Overview of Dubbo RPC Framework

    Dubbo is an RPC-based service development framework designed for microservices. It provides tools for programming paradigms, high-performance RPC communication, and service monitoring and governance. It is well-adapted for Spring Boot microservices. The core philosophy is 'localization of remote calls,' allowing developers to call remote services as if they were local methods using TCP long connections via Netty.

  8. Understand Dubbo Cluster Fault Tolerance Architecture

    Dubbo handles cluster failures through a specific call chain: Cluster -> Directory -> Router -> LoadBalance -> Invoker.

    • Invoker: An abstraction of a callable service on a provider, encapsulating the address and interface.
    • Directory: A collection of multiple Invokers.
    • Cluster: Wraps multiple Invokers from a Directory into a single Invoker and implements the fault tolerance logic (e.g., retrying another provider on failure). Default mode is failover.
    • Router: Filters the list of Invokers based on routing rules to select a subset.
    • LoadBalance: Selects a single specific Invoker from the subset for the current call.

  9. Develop and use an Instruction Set Simulator (ISS)

    An ISS is a processor model written in a high-level language used during the architecture design phase. It should allow flexible configuration of parameters (e.g., Memory size, Buffer size, MAC count, DDR bandwidth) and provide profiling capabilities to detect MAC utilization or buffer starvation.

    There are two primary types of ISS models:

    • Behavior Model: Simulates processor behavior without timing concepts. Used for rapid modeling and as a reference for RTL verification.
    • Cycle Accurate Model: Includes timing concepts. Used as a reference for software development to reflect real-world execution performance.

  10. Identify CNN and RNN Characteristics

    Different network architectures are used based on data types:

    • CNN (Convolutional Neural Networks): Characterized by local connectivity, weight sharing, and spatial/temporal sub-sampling.
    • RNN (Recurrent Neural Networks): Used for sequential data. Common patterns include Sequence-to-Category, Synchronous Sequence-to-Sequence, and Asynchronous Sequence-to-Sequence.

  11. Understand invasive BCI and Local Field Potentials (LFP)

    Invasive BCIs involve implanted electrodes, offering higher spatial resolution than non-invasive methods.

    • Intracortical Microelectrodes: Capable of recording single-neuron firing activity.
    • Local Field Potentials (LFP): Reflect the synaptic activity of a local population of neurons.

  12. Understand Docker Networking Architecture

    Docker networking is implemented using a combination of Network Namespaces, veth pairs, Linux Bridges, and iptables.

    Core Components:

    • Network Namespace: Isolates the network stack for each container.
    • veth pair: Virtual Ethernet cables that connect the container's namespace to the host's namespace.
    • Linux Bridge: A virtual switch that routes traffic between containers on the same host.
    • iptables/NAT: Handles external communication.
      • SNAT (MASQUERADE): Used when a container accesses the external network (changes source address).
      • DNAT: Used when external traffic is routed to a specific container (changes destination address).

    Available Network Modes:

    • bridge: Default mode; provides an independent namespace.
    • host: Shares the host's network stack (high performance, but risks port conflicts).
    • none: Complete network isolation.
    • container: Shares the network namespace of another specific container.