PolyU | Các Khái Niệm Trí Tuệ Nhân Tạo

本实验介绍了 Google Colab 环境的配置，包括挂载 Google Drive 以实现数据持久化、使用 pip 管理 Python 库，以及利用 PyTorch 张量和 GPU 加速 AI 计算。此外，课程还涵盖了通过字典处理符号知识及使用交互式组件优化模型参数的方法，为后续的符号 AI 与深度学习任务奠定基础。

This lab introduces Symbolic AI through the classic SHRDLU "Blocks World," focusing on representing world states, implementing symbolic physics, and using recursion for object manipulation. Students will learn to transition from "blind" command execution to intelligent automated planning by developing logic to resolve physical constraints and dependencies.

本节课介绍了五子棋AI的开发基础，涵盖了使用二维列表表示棋盘、利用嵌套循环与方向向量进行胜负判定，以及通过递归实现Minimax算法与Alpha-Beta剪枝来优化AI的决策能力。课程重点在于如何通过启发式评估函数和回溯法，让AI能够模拟并预判未来的棋局走势。

本实验旨在通过引入位置权重启发式函数、基于邻居的空间局部性过滤以及移动排序策略，优化五子棋AI的对抗搜索算法。这些改进有效降低了分支因子并提升了Alpha-Beta剪枝效率，使AI能够在有限时间内进行更深层次的博弈决策。

本节课介绍了COMP5511课程作业报告的撰写规范，重点讲解了如何使用Markdown进行排版以及如何构建摘要、引言、理论背景和方法论等核心章节。学生将学习如何通过客观的学术语言阐述SHRDLU符号AI与Gomoku博弈算法的实现逻辑，从而提升技术文档的专业性与逻辑性。

本課程介紹了從符號式 AI 到機器學習的範式轉移，並透過 MNIST 手寫數字識別任務，學習如何使用 PyTorch 處理張量數據及進行預處理。同時，課程引導學生利用 Scikit-Learn 進行特徵工程（展平與標準化），並訓練支持向量機（SVM）模型來實現圖像分類。

This lab introduces Deep Learning and Convolutional Neural Networks (CNNs) using PyTorch to overcome the spatial limitations of classic machine learning. Students learn to build modular architectures with convolutional, activation, and pooling layers, and explore the iterative training process involving forward passes, backpropagation, and optimization.

This lab explores the scalability challenges of traditional machine learning versus deep learning by transitioning from simple MNIST digits to the complex CIFAR-10 dataset. Students will learn to optimize model performance through deeper CNN architectures, GPU acceleration, batch processing, and PyTorch compilation techniques.

This lab introduces the fundamentals of Natural Language Processing (NLP) by transitioning from image classification to text generation using mini-LLMs like Qwen-0.6B. Students will learn to use the Hugging Face library to load models and tokenizers, understand the role of parameters and hardware acceleration, and explore how instruction-tuned models process prompts to generate human-like responses.

This lab introduces the concept of fine-tuning by teaching you how to adapt a pre-trained Qwen model to adopt a specific "Emoji Persona" using a small, conversationally formatted dataset. You will learn the practical workflow of preparing training data, applying chat templates, and using tokenization techniques to efficiently nudge a model's behavior without the need for full-scale pre-training.

This lab explores the "Brain and Hands" framework, which overcomes the limitations of LLMs by integrating them with external APIs and Python tools to interact with the real world. Students learn to build AI agents that follow structured system instructions, parse commands, and operate in a continuous "Think-Act-Observe" loop to process live data.

This lab introduces the ReAct (Reasoning + Acting) pattern, which improves AI performance by forcing models to "think out loud" and plan before executing actions. Students will learn to implement a scratchpad memory and a Wikipedia tool to enable agents to perform multi-step, fact-based reasoning while overcoming the stateless nature of LLMs.

This mock exam provides a self-paced review and practice opportunity for the PolyU COMP5511 course to help students prepare for the upcoming assessment. It serves as a tool to test your understanding of key course concepts and evaluate your readiness for the final exam.