Please written by computer source An LSTM with a single hidden layer has 2 neurons in all its gates. The forget gate weight matrix (Wf), the input gate weight matrix (Wi) and the output gate weight matrix (Wo) are all given. Wf=Wi=Wo=[[-0..8, 0, -0.3,-0.75, 0.26, 0.09];[0.05, -0.1, -0.9, 0, 0.85, 0.07]].Where the first two columns are the weights for the CEC memory (i.e Wfc/Wic/Woc) the following two columns are weights for hidden memory (i.e Wfh/Wih/Who) and the last two columns are weights for the input (i.e Wfx/Wix/Wox). The input pattern detection weights Wc are given by The input pattern detection weights Wc are given by Wc = [[0.8, 0.9, 0.3, 1];[0.3, 0.45, 0.7, 0.5]]. Where the first two columns are the weights for the hidden vector (Wch) and the following two are the weights for the input (Wcx). Let us define the “memory duration” of the network as the minimum number of time steps N such that for an isolated input at time t (with no previous or subsequent inputs) the length of the cell memory activation vector at time t+N almost certainly falls to less than 1/100th of its value at time t. This is effectively the amount of time in which the influence of the input at time t all but vanishes from the network.The input to the network at time t x(t) is [-1 1]^T where T is transpose. The CEC memory and the hidden memory are both zero vectors initially. What is the memory duration of the above network (choose the closest answer). You may want to simulate the network to determine this value (an analytical solution is difficult to get). Hint: You may want to check the Recurrent Networks:Stability analysis and LSTMs lecture slide 88 if not sure about what the weights are.
Please written by computer source
An LSTM with a single hidden layer has 2 neurons in all its gates. The forget gate weight matrix (Wf), the input gate weight matrix (Wi) and the output gate weight matrix (Wo) are all given. Wf=Wi=Wo=[[-0..8, 0, -0.3,-0.75, 0.26, 0.09];[0.05, -0.1, -0.9, 0, 0.85, 0.07]].Where the first two columns are the weights for the CEC memory (i.e Wfc/Wic/Woc) the following two columns are weights for hidden memory (i.e Wfh/Wih/Who) and the last two columns are weights for the input (i.e Wfx/Wix/Wox). The input pattern detection weights Wc are given by The input pattern detection weights Wc are given by Wc = [[0.8, 0.9, 0.3, 1];[0.3, 0.45, 0.7, 0.5]]. Where the first two columns are the weights for the hidden vector (Wch) and the following two are the weights for the input (Wcx). Let us define the “memory duration” of the network as the minimum number of time steps N such that for an isolated input at time t (with no previous or subsequent inputs) the length of the cell memory activation vector at time t+N almost certainly falls to less than 1/100th of its value at time t. This is effectively the amount of time in which the influence of the input at time t all but vanishes from the network.The input to the network at time t x(t) is [-1 1]^T where T is transpose. The CEC memory and the hidden memory are both zero
Step by step
Solved in 2 steps
