JavaScript is required

Trong học máy, để giảm thiểu hiện tượng overfitting, ta có thể thực hiện các biện pháp nào sau đây?

A.

Sử dụng kích thước batch lớn

B.

Sử dụng learning rate nhỏ

C.

Thêm dữ liệu huấn luyện

D.

Tăng số lượng epoch

Trả lời:

Đáp án đúng: C


Overfitting occurs when a model learns the training data too well, resulting in good performance on the training data but poor performance on new data. To reduce overfitting, we can apply the following measures: - **Add more training data:** The more data, the more opportunity the model has to learn more general features rather than just memorizing the existing training data. - **Use regularization techniques:** Such as L1, L2 regularization, dropout, and early stopping. - **Reduce model complexity:** By reducing the number of parameters, layers, or neurons in the neural network. - **Use a large batch size:** Using a larger batch size can help reduce noise during training, which can help the model generalize better and reduce overfitting. Analysis of the options: - **A. Use a large batch size:** Correct. A large batch size tends to reduce overfitting compared to a small batch size. - **B. Use a small learning rate:** A small learning rate mainly affects the learning speed and does not directly reduce overfitting. It can help the model converge better but does not address the root cause of overfitting. - **C. Add more training data:** Correct. Adding more training data helps the model learn more general features and reduces overfitting. - **D. Increase the number of epochs:** Increasing the number of epochs can increase overfitting if there are no other control measures, as the model will have more opportunities to memorize the training data. Since both A and C are correct, but the question only asks for one answer. In this case, C is often considered one of the most basic and effective approaches to reducing overfitting. However, in the context of the options given, using a large batch size is also a valid method for minimizing overfitting. Therefore, both A and C can be considered correct. However, based on experience and practical implementation, adding more training data (C) is often a more effective solution than just adjusting the batch size (A). Therefore, option C is somewhat "stronger" in solving overfitting. Therefore, both A and C could be correct answers. However, C is the more appropriate choice.

Câu hỏi liên quan