Loss scaling for neural networks

A system comprising:

• one or more computers comprising one or more processors to:
• identify a distribution of different geometric features of one or more paths within a plurality of images;
• determine a relative frequency of the different geometric features based on the distribution;
• calculate loss values based, at least in part, on one or more outputs generated by one or more neural networks using the plurality of images;
• compute scaled loss values by scaling at least one of the loss values based, at least in part, on the relative frequency of the different geometric features;
• and
• update the one or more neural networks to predict one or more additional paths of one or more vehicles based, at least in part, on the scaled loss values.

A machine-readable medium having stored thereon a set of instructions which, if performed by one or more processors, cause the one or more processors to at least:

• identify a distribution of different geometric features of one or more paths within a plurality of images;
• determine a relative frequency of the different geometric features based on the distribution;
• calculate loss values based, at least in part, on one or more outputs generated by one or more neural networks using the plurality of images;
• compute scaled loss values by scaling at least one of the loss values based, at least in part, on the relative frequency of the different geometric features;
• and
• update the one or more neural networks to predict one or more additional paths of one or more vehicles based, at least in part, on the scaled loss values.