Cuestionario vinculado a una guía · fácil Nivel

Prueba de redes neuronales

Pesos de prueba, capas ocultas, ReLU, actualizaciones de gradientes, interpretación de resultados y por qué las redes más grandes aún necesitan pruebas cuidadosas en nuevos ejemplos.

Rutas de guía relacionadasRedes neuronales
Pregunta 1 de 8

In the toy neuron, input 0.8 is multiplied by 0.6 before a bias is added. What is the 0.6?

Vista previa de cada pregunta8 preguntas

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

  1. In the toy neuron, input 0.8 is multiplied by 0.6 before a bias is added. What is the 0.6?
    • The bias added to every weighted sum
    • The weight applied to the first input
    • A held-out target label
    • The final output after activation
  2. What do hidden layers do between input and output?
    • Transform inputs into intermediate representations used by later layers
    • Store the original training labels unchanged
    • Select the final answer without computation
    • Replace the need for an output layer
  3. What role does ReLU play when applied to a value?
    • It computes the loss against a target
    • It updates weights using a learning rate
    • It assigns the correct class label
    • It returns zero for a negative input and passes a positive one through
  4. What does backpropagation provide to the training process?
    • The final parameter values after optimization
    • Gradients describing how parameters affect loss
    • A guarantee of the globally best solution
    • A separate test set for measuring generalization
  5. How should the brain analogy for neural networks be understood?
    • Artificial neurons are mathematical abstractions, not biological neurons
    • Modern networks reproduce brain cells in full detail
    • A successful network proves human-like understanding
    • Each network layer corresponds exactly to a brain region
  6. In the guide’s toy neuron, input 1.5 produces a weighted sum of -0.02. What does ReLU output?
    • 1.5 because ReLU preserves the input
    • -0.02 because ReLU preserves every sum
    • 0.02 because ReLU removes the negative sign
    • 0 because ReLU maps negative values to zero
  7. What happens when linear transformations are stacked without nonlinear activations?
    • Their composition is still linear
    • The stack automatically becomes nonlinear because it is deeper
    • Backpropagation turns the stack into a classifier
    • Each layer becomes an independent probability
  8. Why should a numeric output and category scores be interpreted differently?
    • A numeric output is always a confidence percentage
    • Category scores are guaranteed probabilities
    • Output meaning depends on the task and the chosen output layer
    • Every output layer returns a class label