← AI Researcher
AI Researcher
Lý thuyết sâu
Vietnamese onlyThe learning content is available in Vietnamese only. The surrounding interface follows your language.Backprop as chain rule / autograd Vanishing and exploding gradients L1/L2, dropout, early stopping vs capacity
7h3 lessonsAI Researcher
Format
Guided module
Duration
7h
Lessons
3
What you will practice
- Backprop as chain rule / autograd
- Vanishing and exploding gradients
- L1/L2, dropout, early stopping vs capacity
Lesson order
3 lessons1
Lan truyền ngược — truy ngược lỗi qua từng lớp
Mạng đoán sai — lỗi đến từ lớp nào, weight nào cần sửa bao nhiêu? Backprop dùng quy tắc chuỗi để truy ngược lỗi từ đầu ra về từng trọng số, chỉ trong một lần duyệt.
lesson140 min
2
Gradient triệt tiêu & bùng nổ
Hai vấn đề khiến việc huấn luyện mạng sâu trở nên khó khăn khi gradient quá nhỏ hoặc quá lớn.
lesson140 min
3
Chính quy hóa
Các kỹ thuật chống overfitting bằng cách thêm ràng buộc vào quá trình huấn luyện.
lesson140 min