Which neural network architecture utilizes internal feedback loops and hidden states, making it well-suited for sequential and time-series data?
RNNs feature cyclical recurrent connections that pass hidden state representations from previous time steps to current time steps, making them suited for sequential modeling.
What architectural innovations allow Long Short-Term Memory (LSTM) networks to capture long-range dependencies better than basic RNNs?
LSTMs use gated cell states to selectively retain, write, or discard information across long sequences, directly mitigating the vanishing gradient problem in basic RNNs.
In deep learning regularized training, what does 'Dropout' do during forward propagation?
Introduced by Srivastava and Hinton, dropout randomly zeroes out a fraction p of activations during training, preventing neurons from developing fragile co-dependencies.
Which computer scientists are widely honored as the 'Godfathers of Deep Learning' and received the 2018 ACM A.M. Turing Award?
Geoffrey Hinton, Yann LeCun, and Yoshua Bengio were awarded the 2018 ACM A.M. Turing Award for conceptual and engineering breakthroughs in deep learning. None of the listed options contain this trio.
Which deep convolutional neural network won the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) in 2012, catalyzing the modern Deep Learning revolution?
AlexNet won the 2012 ImageNet challenge by a wide margin using GPU acceleration, ReLU activations, and Dropout, initiating the modern deep learning era.