Most Common Loss Function In Deep Learning
2 min readJul 23, 2024
In deep learning, loss functions are crucial for training models as they measure the difference between the predicted output and target values.
Regression Loss Function
1. Mean Absolute Error (MAE)
- Measures the average absolute difference between actual and predicted values.
- It’s robust to outliers compared to MSE.
- It may not penalize large errors as heavily.
- Also known as L1 Loss.
2. Mean Squared Error (MSE)
- Measures the average squared difference between actual and predicted values.
- It’s widely used for regression tasks.
- It’s good for minimizing overall error but sensitive to outliers.
- Also known as L2 Loss.
3. Huber Loss
- This combines the characteristics of MSE and MAE.
- It’s less sensitive to outliers than MSE.
- δ is the delta parameter, which determines the threshold for switching between the quadratic and linear components of the loss function.
Classification Loss Function
1. Binary Cross-Entropy Loss (Log Loss)
- Used for binary classification tasks.
- It measures the performance of a classification model whose output is a probability value between 0 and 1.
2. Categorical Cross-Entropy Loss
- Used for multi-class classification tasks.
- It generalizes binary cross-entropy to multiple classes.
3. Sparse Categorical Cross-Entropy Loss
- Similar to categorical cross-entropy, it is used when target labels are integers rather than one-hot encoded vectors.