An activation function is necessary to introduce non-linearity, enabling the neural network to learn complex patterns.

Activation functions are only used for output layers.

Activation functions help in increasing the learning rate.

They are necessary to reduce the number of neurons in the network.

Regression tasks do not require a loss function, while classification tasks always do.

Both regression and classification use the same loss function, typically MSE.

The choice of loss function differs as regression uses continuous value loss functions (e.g., MSE), while classification uses discrete class loss functions (e.g., Cross-Entropy).

Regression uses categorical loss functions like MSE, while classification uses continuous loss functions.

Backpropagation updates the weights of a neural network to minimize the error during training.

Backpropagation increases the error during training.

Backpropagation determines the architecture of the neural network.

Backpropagation is used to initialize the neural network weights.

Hidden layers have no impact on the model's performance.

Hidden layers enhance the model's ability to learn complex representations and improve performance.

Hidden layers reduce the model's capacity to learn.

Hidden layers only serve to increase computation time.

Swish

Softmax

ReLU, Sigmoid, Tanh, Leaky ReLU

Maxout

sigmoid

softmax

tanh

relu

The optimizer increases the loss function to improve accuracy.

The optimizer is only used for initializing the neural network.

The optimizer determines the architecture of the neural network.

The optimizer adjusts the weights of the neural network to minimize the loss function, crucial for effective learning.

The learning rate determines the number of iterations in the optimization process.

The learning rate has no impact on the convergence of the algorithm.

The learning rate affects the speed and stability of the optimization process.

The learning rate only affects the final outcome of the optimization.

To reduce the dimensionality of the input data.

To increase the learning rate during training.

To apply regularization to the model parameters.

To convert logits into probabilities for multi-class classification.

Overfitting can be solved by reducing the size of the training dataset.

Overfitting occurs when the model has too few layers.

Overfitting is beneficial for improving model accuracy on unseen data.

Overfitting in a multilayer perceptron occurs when the model learns noise in the training data, leading to poor generalization. It can be mitigated through regularization, dropout, early stopping, and increasing training data.

Weight initialization is significant as it affects convergence speed and helps avoid vanishing or exploding gradients.

Weight initialization is primarily used for regularization techniques.

Weight initialization is only important for convolutional layers.

Weight initialization has no impact on model performance.

Dropout layers increase the number of neurons to enhance learning.

Dropout layers replace the activation function with a linear function.

Dropout layers are used only during testing to improve accuracy.

Dropout layers improve model generalization by randomly deactivating neurons during training, which prevents overfitting.

ReLU is faster and mitigates vanishing gradients but can cause dying neurons; sigmoid is suitable for binary outputs but suffers from vanishing gradients.

ReLU is always better than sigmoid for all tasks.

Sigmoid is faster and prevents dying neurons effectively.

ReLU is suitable for multi-class outputs without issues.

Use mean squared error for multi-class classification tasks.

Use mean squared error for regression tasks with continuous outputs.

Use cross-entropy for regression tasks with continuous outputs.

Use mean squared error for binary classification tasks.