ATR

ATR(input_size, hidden_size;
    return_state = false, kwargs...)

Addition-subtraction twin-gated recurrent cell ^[Zhang2018]. See ATRCell for a layer that processes a single sequence.

Arguments

• input_size => hidden_size: input and inner dimension of the layer.

Keyword arguments

• init_kernel: initializer for the input to hidden weights. Default is glorot_uniform.
• init_recurrent_kernel: initializer for the hidden to hidden weights. Default is glorot_uniform.
• bias: include a bias or not. Default is true

Equations

\[\begin{aligned} \mathbf{p}(t) &= \mathbf{W}_{ih} \mathbf{x}(t) + \mathbf{b}, \\ \mathbf{q}(t) &= \mathbf{W}_{hh} \mathbf{h}(t-1), \\ \mathbf{i}(t) &= \sigma\left( \mathbf{p}(t) + \mathbf{q}(t) \right), \\ \mathbf{f}(t) &= \sigma\left( \mathbf{p}(t) - \mathbf{q}(t) \right), \\ \mathbf{h}(t) &= \mathbf{i}(t) \circ \mathbf{p}(t) + \mathbf{f}(t) \circ \mathbf{h}(t-1). \end{aligned} \]

Forward

atr(inp, state)
atr(inp)

Arguments

• inp: The input to the atr. It should be a vector of size input_size x len or a matrix of size input_size x len x batch_size.
• state: The hidden state of the ATR. If given, it is a vector of size hidden_size or a matrix of size hidden_size x batch_size. If not provided, it is assumed to be a vector of zeros, initialized by Flux.initialstates.

Returns

• New hidden states new_states as an array of size hidden_size x len x batch_size. When return_state = true it returns a tuple of the hidden stats new_states and the last state of the iteration.