Quiz#1

Which component is NOT part of the Transformer model architecture?

• Convolutional Layer

• Feedforward Network

• Decoder

• Encoder

In self-attention, which tokens are used for inputs Q, K, and V?

• Different tokens for each

• Only queries are different

• Same tokens for all three

• Only keys and values are the same

What does the Vision Transformer (ViT) primarily use for image classification?

• Convolutional Neural Networks

• Support Vector Machines

• Recurrent Neural Networks

• Pure Transformer architecture

What is the purpose of positional encoding in the Transformer?

• To reduce dimensionality

• To add information about token order

• To enhance computational speed

• To increase model complexity

What is the classification token used for in the Vision Transformer?

• To reduce training time

• To enhance image resolution

• To represent the entire image

• To increase the number of patches

What is the dimensionality of the model in the training example provided?

• 10

• 8

• 6

• 4

Which layer is NOT mentioned as part of the Transformer architecture?

• Pooling Layer

• Multi-Head Attention

• Layer Normalization

• Residual Connections

What is the main advantage of using Vision Transformers over convolutional networks?

• Faster training times

• Less computational resources required

• Higher accuracy

• Better feature extraction

What is the role of the Gaussian Error Linear Unit (GELU) in the Vision Transformer?

• To increase model size

• To normalize inputs

• To apply activation function

• To reduce overfitting