| CPC G06N 3/047 (2023.01) [G06N 3/045 (2023.01); G06N 3/088 (2013.01)] | 20 Claims |
|
1. A method comprising:
receiving a request to generate an output example of a particular type having a sequence of T N-bit samples in a sample order, wherein N and T are respective integers greater than one;
accessing dependency data, wherein the dependency data:
partitions the N*T bits in the output example into a plurality of blocks of bits, each block of bits comprising a respective plurality of bits from the output example,
assigns, for each sample in the sequence, a different portion of the N bits in the sample to each of a plurality of different blocks of bits, and
for each of a plurality of the blocks, defines a respective dependency for the block that identifies one or more other blocks in the plurality of blocks on which values of the bits in the block depend; and
generating the output example by, at each of a plurality of generation time steps:
identifying one or more current blocks for the generation time step, wherein each current block is a block for which the values of the bits in all of the other blocks identified in the dependency for the block have already been generated; and
generating, using an auto-regressive neural network, the values of the bits in the one or more current blocks for the generation time step by, for each current block, performing a forward pass through a plurality of layers of the auto-regressive neural network conditioned on the already generated values of the bits in the other blocks identified in the dependency for the current block to generate the values of the bits in the current block.
|