Skimming data sequences using recurrent neural networks

A system comprising one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the one or more computers to perform operations comprising:

• receiving a request to generate a system output for an input data sequence, the input data sequence comprising a plurality of tokens arranged in an input order;
• sequentially processing each of the plurality of tokens in the input order to generate a final internal state of a recurrent neural network, comprising, for each of the plurality of tokens:
• determining whether the token has previously been designated as a token to be skipped during the processing of a previous token;
• when the token has not been designated as a token to be skipped:
• processing the token using the recurrent neural network to update a current internal state of the recurrent neural network, and
• when the token is not last in the input order, processing the updated internal state of the recurrent neural network through one or more neural network layers to determine whether to designate one or more next tokens in the input order as tokens to be skipped;
• wherein one or more tokens of the plurality of tokens are not designated as tokens to be skipped;
• and
• generating the system output from the final internal state of the recurrent neural network.

One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

• receiving a request to generate a system output for an input data sequence, the input data sequence comprising a plurality of tokens arranged in an input order;
• sequentially processing each of the plurality of tokens in the input order to generate a final internal state of a recurrent neural network, comprising, for each of the plurality of tokens:
• determining whether the token has previously been designated as a token to be skipped during the processing of a previous token;
• when the token has not been designated as a token to be skipped:
• processing the token using the recurrent neural network to update a current internal state of the recurrent neural network, and
• when the token is not last in the input order, processing the updated internal state of the recurrent neural network through one or more neural network layers to determine whether to designate one or more next tokens in the input order as tokens to be skipped;
• wherein one or more tokens of the plurality of tokens are not designated as tokens to be skipped;
• and
• generating the system output from the final internal state of the recurrent neural network.