Which of the following is not a part of a convolutional neural network?

In a convolutional neural network (CNN), the architecture is typically composed of layers specifically designed for processing grid-like data, such as images. These layers include the convolutional layer, which is responsible for applying filters to input data to capture spatial hierarchies, and the pooling layer, which reduces the spatial dimensions of the data to decrease computational load and retain essential features while eliminating lesser details. A fully-connected layer is also present in many CNN architectures, particularly at the end of the network, where it helps in deriving the final output from the features learned throughout the convolution and pooling processes.

Conversely, a bidirectional hidden layer is not a standard component of CNNs. Bidirectional layers are primarily found in recurrent neural networks (RNNs), where they allow the network to process data in both forward and backward directions, thereby capturing patterns across time series or sequences. This design is beneficial for tasks where the context from the past and future observations is critical, such as in natural language processing. Therefore, since a bidirectional hidden layer is not integrated into the framework of convolutional neural networks, it is correctly identified as the option that does not belong in this context.