How do VAEs resolve the unconstrained latent space issue?

In a standard feedforward neural network layer, which mathematical operation correctly represents the transformation of an input vector x to the pre-activation output z?

During the backpropagation phase in deep learning, what mathematical principle is primarily utilized to compute the gradients of the loss function with respect to the weights in early layers?

Which training methodology involves updating all or a subset of model parameters on a smaller, task-specific dataset after the model has completed broad pre-training?

In the Self-Attention mechanism of a Transformer, how is the attention weight matrix scaled before applying the Softmax function?

In deep neural networks, what primary role do "biases" serve alongside weight matrices?

Which Transformer architectural variant omits the explicit cross-attention mechanism and is optimized primarily for auto-regressive text generation tasks?

What is a defining paradigm shift when moving from Traditional AI (Discriminative/Task-specific) to Foundation Models?

What is the derivative of the Rectified Linear Unit (ReLU) activation function for any input x > 0?

In the reverse diffusion process, what is the objective of the U-Net?