CPC G06N 3/08 (2013.01) [G10L 15/063 (2013.01); G10L 15/16 (2013.01); G10L 15/22 (2013.01); G10L 25/30 (2013.01); G10L 25/51 (2013.01); G10L 2015/088 (2013.01)] | 20 Claims |
1. A method for training an event classifier comprising:
receiving a training dataset including a plurality of input samples having segmented labeled data;
computing a classifier output for each input sample of the training dataset in a forward pass through the classifier;
tuning a many-or-one detection (MOOD) cost function using a hyperparameter in accordance with at least one event classifier goal;
updating weights and biases of the classifier through a backward pass, including determining whether an input frame is in a Region of Target (ROT), and estimating the update of the weights and the biases of the classifier;
wherein the classifier is trained using the tuned MOOD cost function to cause the classifier to spike at least one time during a duration of the event.
|