Advanced Machine Learning & AI

1. Giới thiệu khóa học

Khóa học Advanced ML & AI đưa bạn từ research đến production, tập trung vào:

• ✅ Advanced techniques: AutoML, Meta-learning, Few-shot learning
• ✅ Production ML: MLOps, model deployment, monitoring
• ✅ Real-world project: E-commerce recommendation system
• ✅ Best practices: A/B testing, feature stores, CI/CD

2. Roadmap khóa học

Module 1: Advanced ML Techniques (Tuần 1-3)

Bài 2: Hyperparameter Tuning nâng cao

• Grid Search, Random Search, Bayesian Optimization
• Optuna, Hyperopt
• Neural Architecture Search (NAS)

Bài 3: AutoML

• Auto-sklearn, TPOT, H2O AutoML
• AutoKeras, NAS
• When to use AutoML

Bài 4: Advanced Ensemble Methods

• Stacking, Blending
• Weighted ensembles
• Bayesian Model Averaging

Bài 5: Transfer Learning & Fine-tuning

• Pre-trained models (BERT, GPT, ResNet)
• Domain adaptation
• Few-shot learning

Module 2: Recommendation Systems (Tuần 4-6)

Bài 6: Recommendation Systems Overview

• Collaborative Filtering (User-based, Item-based)
• Content-based filtering
• Hybrid approaches

Bài 7: Matrix Factorization

• SVD, ALS
• Implicit feedback
• Non-negative Matrix Factorization

Bài 8: Deep Learning for RecSys

• Neural Collaborative Filtering
• Deep Matrix Factorization
• Embeddings

Bài 9: Advanced RecSys

• Two-tower models
• Sequential recommendations (RNN, Transformers)
• Context-aware recommendations

Module 3: MLOps & Production (Tuần 7-9)

Bài 10: MLOps Fundamentals

• ML lifecycle
• Experiment tracking (MLflow, Weights & Biases)
• Model registry

Bài 11: Model Deployment

• REST APIs (FastAPI, Flask)
• Model serving (TensorFlow Serving, TorchServe)
• Batch vs Real-time inference

Bài 12: Feature Engineering & Feature Store

• Feature pipelines
• Feature store (Feast, Tecton)
• Online vs Offline features

Bài 13: Model Monitoring & Maintenance

• Data drift detection
• Model performance monitoring
• Retraining strategies

Module 4: Advanced Topics & Project (Tuần 10-12)

Bài 14: A/B Testing & Experimentation

• A/B test design
• Statistical significance
• Multi-armed bandits

Bài 15: Fairness & Bias in ML

• Bias detection
• Fairness metrics
• Mitigation strategies

Bài 16: Final Project - E-commerce Recommendation System

3. Dự án Thực hành

🎯 E-commerce Recommendation System

Mô tả: Xây dựng hệ thống gợi ý sản phẩm hoàn chỉnh cho website thương mại điện tử.

Components:

11. Data Pipeline
2   ├── User interactions (clicks, purchases, ratings)
3   ├── Product catalog & metadata
4   └── Feature engineering pipeline
5 
62. Recommendation Models
7   ├── Collaborative Filtering (ALS)
8   ├── Content-based (TF-IDF + Cosine Similarity)
9   ├── Deep Learning (Two-tower model)
10   └── Ensemble (Weighted combination)
11 
123. Deployment
13   ├── REST API (FastAPI)
14   ├── Real-time inference (<100ms)
15   ├── Batch predictions (daily)
16   └── A/B testing framework
17 
184. Monitoring
19   ├── Click-through rate (CTR)
20   ├── Conversion rate
21   ├── Model drift detection
22   └── Dashboard (Grafana)

Tech Stack:

• Data: PostgreSQL, Redis, Feast
• Training: PyTorch, Scikit-learn, Implicit
• MLOps: MLflow, DVC, Airflow
• Deployment: FastAPI, Docker, Kubernetes
• Monitoring: Prometheus, Grafana, Evidently

Metrics:

• Precision@K, Recall@K, NDCG@K
• Click-through rate (CTR)
• Conversion rate
• Revenue per user

4. Công cụ & Frameworks

4.1 AutoML

4.2 MLOps

1# Experiment tracking
2import mlflow
3mlflow.start_run()
4mlflow.log_params({"lr": 0.01, "epochs": 100})
5mlflow.log_metric("accuracy", 0.95)
6mlflow.sklearn.log_model(model, "model")
7
8# Feature store
9from feast import FeatureStore
10store = FeatureStore(repo_path=".")
11features = store.get_online_features(
12    features=["user_features:age", "product_features:category"],
13    entity_rows=[{"user_id": 123, "product_id": 456}]
14)
15
16# Model monitoring
17from evidently import ColumnMapping
18from evidently.dashboard import Dashboard
19dashboard = Dashboard(tabs=[DataDriftTab()])
20dashboard.calculate(reference_data, production_data)

5. Yêu cầu đầu vào

5.1 Kiến thức

• ✅ Machine Learning fundamentals (regression, classification, trees)
• ✅ Deep Learning basics (neural networks, CNNs, RNNs)
• ✅ Python advanced (OOP, decorators, context managers)
• ✅ SQL, Pandas, NumPy

5.2 Setup

1# ML frameworks
2pip install scikit-learn xgboost lightgbm
3 
4# Deep Learning
5pip install torch torchvision transformers
6 
7# AutoML
8pip install auto-sklearn tpot h2o
9 
10# MLOps
11pip install mlflow wandb dvc feast
12 
13# Deployment
14pip install fastapi uvicorn docker
15 
16# Monitoring
17pip install evidently prometheus-client

6. Lộ trình học tập

Tuần 1-3: Advanced Techniques

• Mục tiêu: Master AutoML, ensembles, transfer learning
• Deliverable: Kaggle competition submission

Tuần 4-6: Recommendation Systems

• Mục tiêu: Build RecSys from scratch
• Deliverable: Movie recommendation prototype

Tuần 7-9: MLOps

• Mục tiêu: Deploy ML model to production
• Deliverable: Deployed API with monitoring

Tuần 10-12: Final Project

• Mục tiêu: Complete e-commerce RecSys
• Deliverable: Full-stack recommendation system

7. Đánh giá

Pass criteria:

• Total score ≥ 70%
• Final project demo
• Code review passed

8. Career Paths

Sau khóa học, bạn có thể:

• ML Engineer: Deploy & maintain ML systems
• MLOps Engineer: Build ML infrastructure
• Research Scientist: Advanced algorithms
• Data Scientist (Senior): End-to-end ML projects

9. Tài liệu tham khảo

Books

• "Designing Machine Learning Systems" (Chip Huyen) - Production ML
• "Machine Learning Engineering" (Andriy Burkov) - MLOps
• "Recommender Systems Handbook" (Ricci et al.) - RecSys bible

Courses

• Full Stack Deep Learning
• Made With ML
• Stanford CS329S: ML Systems Design

Papers

• Netflix RecSys: "The Netflix Recommender System"
• Google: "Rules of Machine Learning"
• Uber: "Michelangelo: Uber's ML Platform"

Bắt đầu hành trình nâng cao! 🚀

Next: Bài 2 - Hyperparameter Tuning Nâng cao