Khóa học nâng cao 6 tháng bao gồm các chủ đề AI tiên tiến như Transformers, Generative AI, Self-Supervised Learning, Reinforcement Learning và AI Reasoning.
Tháng 1-3: Advanced Architectures & Self-Supervised Learning (Kiến Trúc AI Tiên Tiến & Học Tự Giám Sát)
✅ Goal (Mục tiêu): Master modern deep learning techniques and architectures
✅ Hands-on Focus (Thực hành): Fine-tuning and developing cutting-edge AI models
🔥 Week 1-2: Advanced Transformers & Attention Mechanisms (Transformer nâng cao & Cơ Chế Chú Ý)
- Deep dive into Self-Attention & Multi-Head Attention (Chú ý Đa Đầu)
- Vision Transformers (ViT) & how they compare to CNNs
- Transformer variations: Long-Short Transformer, Linear Transformers
- Efficient Transformers: FlashAttention, Performer, Linformer
📝 Project: Implement Vision Transformer (ViT) & compare it with ResNet
🧑🎨 Week 3-4: Generative AI & Diffusion Models (AI Tạo Sinh & Mô Hình Khuếch Tán)
- Stable Diffusion & DALL·E (Tạo Ảnh Bằng AI)
- GANs (Generative Adversarial Networks) & StyleGAN
- Autoencoders & Variational Autoencoders (VAEs)
- Text-to-Image & Image-to-Text Models
📝 Project: Train a Stable Diffusion Model to generate artistic images
🔄 Week 5-6: Self-Supervised Learning (Học Tự Giám Sát - SSL)
- Contrastive Learning (SimCLR, MoCo, BYOL, DINO)
- SSL for NLP: BERT, RoBERTa, T5
- SSL for Vision: MAE, MoCo, DINO
📝 Project: Implement a contrastive learning model for image classification
🌐 Week 7-8: Graph Neural Networks (Mạng Nơ-Ron Đồ Thị - GNNs)
- Introduction to Graph Representation Learning
- Graph Convolutional Networks (GCNs) & Graph Attention Networks (GATs)
- Applications in social networks, fraud detection, recommendation systems
📝 Project: Build a GNN for social network link prediction
Tháng 4-6: Large Language Models, Reinforcement Learning, Fine-Tuning & AI Reasoning (Mô hình Ngôn Ngữ Lớn, Học Tăng Cường, Tinh Chỉnh Mô Hình & Lý Luận AI)
🧠 Week 9-12: In-Depth Study of Large Language & Multimodal Models (Mô Hình Ngôn Ngữ Lớn & Mô Hình Đa Phương Thức)
- Overview of Large Language Models (LLMs): GPT-3, T5, BERT, LLaMA
- Introduction to Multimodal Models: CLIP, Flamingo, VisualGPT
- Understanding the architecture: Attention mechanisms, positional encoding, transformers
- Applications of LLMs: Text generation, question answering, summarization
- Applications of LMMs: Text-to-image, image captioning, video generation
📝 Project: Implement a text-to-image model using CLIP or DALL·E
🤖 Week 13-14: Reinforcement Learning & AI Agents (Học Tăng Cường & Tác Tử AI)
- Fundamentals of Reinforcement Learning (Học Tăng Cường)
- Deep Q-Learning (DQN), PPO, A3C
- RL for game AI, robotics, finance
- Advanced RL Topics: Proximal Policy Optimization (PPO), Actor-Critic Methods, Deep Deterministic Policy Gradient (DDPG)
- RL for aligning Large Language Models (e.g., RLHF - Reinforcement Learning from Human Feedback)
📝 Project: Train an AI to play Atari games with RL
⚙️ Week 15-16: Fine-Tuning & Adaptation of Foundation Models (Tinh Chỉnh Mô Hình Nền Tảng - FM Tuning)
- LoRA (Low-Rank Adaptation) & QLoRA
- Adapter Layers, Prefix Tuning, Parameter Efficient Fine-Tuning (PEFT)
- Fine-tuning Llama, GPT, BERT, & Stable Diffusion
📝 Project: Fine-tune GPT-3.5 for domain-specific text generation
🔀 Week 17-18: Mixture of Experts (Mô Hình Pha Trộn Chuyên Gia - MoE)
- MoE architecture & routing algorithms
- Sparse vs. Dense MoE models
- Scaling large models efficiently
📝 Project: Implement a MoE transformer for NLP tasks
⚖️ Week 19-20: AI Fairness & Explainability (Công Bằng AI & Giải Thích Mô Hình)
- Bias in AI models & debiasing techniques
- SHAP, LIME, Integrated Gradients for explainability
- Ethical considerations in AI
📝 Project: Analyze fairness & bias in a deep learning model
🚀 Week 21-24: AI Reasoning & Real-World AI Deployments (Lý luận AI & Triển khai AI thực tế)
- Chain-of-Thought (CoT) Prompting & Self-Consistency
- Tool-using agents: ReAct, Toolformer
- Hybrid AI: Symbolic + Deep Learning Systems
- AI in Finance, Medicine, and Autonomous Systems
- Large-Scale Model Serving with Triton, vLLM
- Model Compression (Distillation, Quantization, Pruning
📝 Project: Deploy a fine-tuned foundation model for a real-world application
Time Commitment (Thời Gian Học)
📅 6-Tháng duration (6 tháng học)
⏳ 15 hours/week (10-15 giờ/tuần, bao gồm 5 giờ học và 10 giờ làm project)