Methods and systems for reducing latency in automated assistant interactions

A system comprising:

• at least one processor;
• and
• memory storing instructions that, when executed by the at least one processor, cause the at least one processor to be operable to:
• obtain a plurality of training instances to be utilized in training a latency prediction model,
• each of the plurality of training instances including corresponding training instance input and corresponding training instance output,
• the corresponding training instance input including corresponding data for a given assistant command, and
• the corresponding training instance output including corresponding ground truth latency associated with fulfillment of the given assistant command included in the corresponding training instance input;
• train, based on the plurality of training instances, the latency prediction model, wherein the instructions to train the latency prediction model based on a given training instance, from among the plurality of training instances, comprise instructions to:
• process, using the latency prediction model, the corresponding data for the given assistant command, that is included in the corresponding training instance input for the given training instance, to generate corresponding predicted latency;
• generate, based on a comparison of the corresponding predicted latency and the corresponding ground truth latency associated with fulfillment of the given assistant command, that is included in the corresponding training instance output for the given training instance, one or more losses;
• and
• update, based on one or more of the losses, the latency prediction model;
• subsequent to updating the latency prediction model:
• cause the latency prediction model to be deployed at a plurality of client devices of respective users;
• and
• subsequent to causing the latency prediction model to be deployed:
• process, using the latency prediction model, data for a given additional assistant command that is received at a given client device, from among the plurality of client devices, to determine a predicted latency associated with fulfillment of the given additional assistant command that is received at the given client device;
• determine an actual latency associated with the fulfillment of the given additional assistant command that is received at the given client device;
• generate, based on an additional comparison of the predicted latency associated with the fulfillment of the given additional assistant command that is received at the given client device and the actual latency associated with the fulfillment of the given additional assistant command that is received at the given client device, one or more additional losses;
• and
• update, based on the one or more additional losses, the latency prediction model.

A non-transitory computer-readable media storing computer-readable instructions that when executed by at least one processor, cause the at least one processor to perform operations, the operations comprising:

• obtaining a plurality of training instances to be utilized in training a latency prediction model,
• each of the plurality of training instances including corresponding training instance input and corresponding training instance output,
• the corresponding training instance input including corresponding data for a given assistant command, and
• the corresponding training instance output including corresponding ground truth latency associated with fulfillment of the given assistant command included in the corresponding training instance input;
• training, based on the plurality of training instances, the latency prediction model, wherein training the latency prediction model based on a given training instance, from among the plurality of training instances, comprises:
• processing, using the latency prediction model, the corresponding data for the given assistant command, that is included in the corresponding training instance input for the given training instance, to generate corresponding predicted latency;
• generating, based on a comparison of the corresponding predicted latency and the corresponding ground truth latency associated with fulfillment of the given assistant command, that is included in the corresponding training instance output for the given training instance, one or more losses;
• and
• updating, based on one or more of the losses, the latency prediction model;
• subsequent to updating the latency prediction model:
• causing the latency prediction model to be deployed at a plurality of client devices of respective users;
• and
• subsequent to causing the latency prediction model to be deployed:
• processing, using the latency prediction model, data for a given additional assistant command that is received at a given client device, from among the plurality of client devices, to determine a predicted latency associated with fulfillment of the given additional assistant command that is received at the given client device;
• determining an actual latency associated with the fulfillment of the given additional assistant command that is received at the given client device;
• generating, based on an additional comparison of the predicted latency associated with the fulfillment of the given additional assistant command that is received at the given client device and the actual latency associated with the fulfillment of the given additional assistant command that is received at the given client device, one or more additional losses;
• and
• updating, based on the one or more additional losses, the latency prediction model.