The following about the standard RNN Model, the correct statement is?

Question

Options

• AThere is no one-to-one model structure
• BDo not consider the time direction when backpropagating
• CThere is no many-to-many model structure
• DThere will be a problem of attenuation of long-term transmission and memory information

• A
  5% (2)
• B
  11% (4)
• C
  3% (1)
• D
  81% (30)

Explanation

Standard RNNs suffer from the vanishing gradient problem: as gradients are backpropagated through many time steps, they shrink exponentially, making it nearly impossible for the network to learn dependencies between distant time steps. This means long-term information effectively "fades out" before it can influence earlier parts of the sequence - making D correct.

Why the distractors are wrong:

• A is false - RNNs do support one-to-one structure (a single input mapped to a single output, like a standard feedforward pass).
• B is false - RNNs use Backpropagation Through Time (BPTT), which explicitly unrolls the network across time steps and propagates gradients backward through them.
• C is false - many-to-many is one of the most common RNN structures (e.g., sequence-to-sequence translation, video captioning).

Memory tip: Think of a standard RNN as having a "short memory" - it can handle nearby context well, but "forgets" the distant past. That's exactly why architectures like LSTM and GRU were invented to fix this long-term memory attenuation problem in D.

No community discussion yet for this question.