Courses

Artificial Intelligence @ NCU (Fall 2024)

這不是典型的人工智慧課程，而是一個快速了解當代人工智慧發展的教程。課程的目標是快速了解最先進的技術，以及他們發展的軌跡，從基本觀念到最前沿的技術。課程內容將涵蓋機器學習、深度學習、語言模型、強化式學習、到對話代理程式等技術，並以事實查核、閱讀理解、問題回答、以及聊天機器人等做為應用主題。課程結束時，學生將深入了解機器如何學習、推理和進行標記，以及如何應用這些能力來解決現實世界的問題。課程也將延伸至對話代理人的設計和開發，包括聊天機器人和任務導向對話系統的應用。

This is not a typical artificial intelligence course but an express train to quickly understand the development of contemporary artificial intelligence.  This accelerated course aims to provide students with a comprehensive understanding of state-of-the-art AI technologies and their development trajectories from foundational concepts to cutting-edge applications. The course content will cover machine learning, deep learning, language modeling, reinforcement learning, and conversational agents, with application topics such as fact-checking, reading comprehension, question answering, and chatbots. This course offers a unique, fast-paced journey through the rapidly evolving field of AI, equipping students with the knowledge and skills needed to engage with and contribute to this transformative technology.  The course also delves into the design and development of conversational agents, such as chatbots and task-oriented dialogue systems, which have seen rapid progress in recent years.

Prerequisites

• Fundamental understanding of probability and statistics

• Python programming skills and colab 

Schedule

1. Lecture 1 Introduction to AI Course

   • AI - A Modern Approach: four editions 1995, 2010, 2016, 2021

   • How do machine learning shape the world

2. Lecture 2 Machine Learning: Classification+Regression

   • When do we need classification and regression?

   • How do humans and machines learn?

   • How can machines learn by embedding numerous features?

3. Lecture 3: Artificial Neural Networks

   • How can machines learn by distilling hidden features?

   • From perceptrons to neural networks

   • Convolutional neural networks

4. Lecture 4: LSTM ＆ Transformer ＆ AutoEncoder

   • How do machines make sequential decisions?

   • Encoder-Decoder

5. Lecture 5: Reinforcement Learning I

   • How do we make sequential decisions?

   • Markov Decision Process

   • Model-free Prediction and Control

6. Lecture 6: Reinforcement Learning II

   • Policy Gradient

   • Q-Learning

   • Actor Critic

7. Lecture 7: Language Model

   • How do machines generate sentences?

   • Transformer, BERT

   • Sequence to Sequence Model

   • Applications: Machine Translation, Question Generation

8. Lecture 8: LLM  + RLHF (chatGPT)

   • Pre-training, Fine-Tuning

   • Instruct GPT: RLHF

   • How do machines learn to respond to our requests?

9. Lectuer 9: Dialog Systems

   • How to construct a Dialog System?

   • Tasked Oriented Dialog Systems: SGD (Schema-Guided Dialog System)

   • Chatbots

10. Lecture 10: Design Thinking Workshop (Project Proposal)

11. Lecture 11: Conversational Agents for Next Generation Human-Computer Interaction

    • From LLM to LAM: How do machines complete tasks for us?

    • Gorrila、CogAgent、RestAPI

12. Lecture 12: Conversation Agents for Education

    • Conversational Agents for Software Development

    • MetaGPT、DevChat、XAgent

教科書/參考書(中文)

- Stuart Russell and Peter Norvig, Artificial Intelligence: A Modern Approach 4th edition, Pearson, 2020. 

- Ian Goodfellow and Yoshua Bengio and Aaron Courville, Deep Learning, An MIT Press book, 2016

- ACL, AAAI papers

Assignment 

1. Stock price prediction

2. AutoEncoder for sequnece encoder

3. Portfolio management with reinforcement learning

4. Sequence labeling

*評量配分比重(中文) 

homework: 15％*3 = 45％

mid-term: 20％

proposal + related work survey: 15％

final report: 10％

involvement: 10％