Generating dual sequence inferences using a neural network model
A computer-implemented method for dual sequence inference using a neural network model includes generating a codependent representation based on a first input representation of a first sequence and a second input representation of a second sequence using an encoder of the neural network model and generating an inference based on the codependent representation using a decoder of the neural network model. The neural network model includes a plurality of model parameters learned according to a machine learning process. The encoder includes a plurality of coattention layers arranged sequentially, each coattention layer being configured to receive a pair of layer input representations and generate one or more summary representations, and an output layer configured to receive the one or more summary representations from a last layer among the plurality of coattention layers and generate the codependent representation.
1. A computer-implemented method for training a neural network model, the method comprising:
generating, using the neural network model, a series of inferences from a first training sequence and a second training sequence;
generating a mixed learning objective from the series of inferences, wherein the generating further comprises:
determining, using a supervised learning objective, a first loss or a first reward for an inference in the series of inferences independently of other inferences in the series of inferences;
determining, using a reinforcement learning objective, a second loss or a second reward over the series of inferences by:
determining a baseline score using a scoring function, wherein the baseline score is based on baseline start and end positions of an answer span and a ground truth start and end positions of the answer span;
determining a reinforcement learning reward function based on the scoring function and the baseline score, wherein the reinforcement learning reward function is based on the ground truth start and end positions of the answer span and a start and an end positions of an inference answer span; and
determining the second loss or the second reward based on the reinforcement learning reward function; and
combining the supervised learning objective and the reinforcement learning objective into a mixed objective using the first and second loss or the first and second reward; and
updating parameters of the neural network model based on a loss or a reward in the mixed learning objective.
2. The computer-implemented method of claim 1 , wherein the updating further comprises:
determining a gradient of the mixed learning objective with respect to the parameters; and
updating the parameters based on the gradient.
3. The computer-implemented method of claim 1 , further comprising:
determining the loss or the reward of the mixed objective based on the series of inferences and a ground truth associated with the first training sequence or the second training sequence.
4. The computer-implemented method of claim 1 , wherein a differentiable objective function determines the first loss or the first reward.
5. The computer-implemented method of claim 4 , wherein the differentiable objective function is based on the start and the end positions of the inference answer span in an inference to a question in the first training sequence or the second training sequence and the start and the end positions of the ground truth answer span.
6. The computer-implemented method of claim 1 , wherein the first training sequence corresponds to a document, the second training sequence corresponds to a question, and the inference corresponds to a span of text in the document that answers the question.
7. The computer-implemented method of claim 1 , wherein the supervised learning objective is a binary evaluation objective and the reinforcement learning objective is a non-binary objective.
8. A system for training a neural network model, the system comprising:
a memory configured to store the neural network model; and
a processor configured to:
generate, using the neural network model, a series of inferences from a first training sequence and a second training sequence;
generate a mixed learning objective from the series of inferences, wherein to generate the mixed learning objective the processor is further configured to:
determine, using a supervised learning objective, a first loss or a first reward for an inference in the series of inferences independently of other inferences in the series of inferences;
determine, using a reinforcement learning objective, a second loss or a second reward over the series of inferences by:
determine a baseline score using a scoring function, wherein the baseline score is based on baseline start and end positions of an answer span and a ground truth start and end positions of the answer span;
determine a reinforcement learning reward function based on the scoring function and the baseline score, wherein the reinforcement learning reward function is based on the ground truth start and end positions of the answer span and a start and an end positions of an inference answer span; and
determine the second loss or the second reward based on the reinforcement learning reward function; and
combine the supervised learning objective and the reinforcement learning objective into a mixed objective using the first and second loss or the first and second reward; and
update parameters of the neural network model based on the mixed learning objective.
9. The system of claim 8 , wherein to update the parameters, the processor is further configured to:
determine a gradient of the mixed learning objective with respect to the parameters; and
update the parameters based on the gradient.
10. The system of claim 8 , wherein the processor is further configured to:
determine a loss and reward of the mixed objective based on the series of inferences and a ground truth associated with the first training sequence or the second training sequence.
11. The system of claim 8 , wherein a differentiable objective function determines the first loss or the first reward.
12. The system of claim 11 , wherein the differentiable objective function is based on the start and the end positions of the inference answer span in an inference to a question in the first training sequence or the second training sequence and start and end positions of the ground truth answer span.
13. The system of claim 8 , wherein the first training sequence corresponds to a document, the second training sequence corresponds to a question, and the inference corresponds to a span of text in the document that answers the question.
14. The system of claim 8 , wherein the supervised learning objective is a binary evaluation objective and the reinforcement learning objective is a non-binary objective.
15. A non-transitory computer readable medium having instructions stored thereon, that when executed by a processor cause the processor to perform operations for training a neural network model, the operations comprising:
generating, using the neural network model, a series of inferences from a training sequence;
generating a mixed learning objective from the series of inferences, wherein generating the mixed learning objective further comprises:
determining, using a supervised learning objective, a first loss or a first reward for an inference in the series of inferences independently of other inferences in the series of inferences;
determining, using a reinforcement learning objective, a second loss or a second reward over the series of inferences by:
determining a baseline score using a scoring function, wherein the baseline score is based on baseline start and end positions of an answer span and a ground truth start and end positions of the answer span;
determining a reinforcement learning reward function based on the scoring function and the baseline score, wherein the reinforcement learning reward function is based on the ground truth start and end positions of the answer span and a start and end positions of an inference answer span; and
determining the second loss or the second reward based on the reinforcement learning reward function; and
combining the supervised learning objective and the reinforcement learning objective into a mixed objective using the first and second loss or the first and second reward; and
updating parameters of the neural network model based on the mixed learning objective.
16. The non-transitory computer readable medium of claim 15 , further comprising:
determining a gradient of the mixed learning objective with respect to the parameters; and
updating the parameters based on the gradient.
17. The non-transitory computer readable medium of claim 15 , wherein the updating further comprises:
determining a gradient of the mixed learning objective with respect to the parameters; and
updating the parameters based on the gradient.
18. The non-transitory computer readable medium of claim 15 , further comprising:
determining the loss or the reward of the mixed objective based on the series of inferences and a ground truth associated with the training sequence.
19. The non-transitory computer readable medium of claim 15 , wherein the training sequence corresponds to a document or to a question, and the inference corresponds to a span of text in the document that answers the question.
20. The non-transitory computer readable medium of claim 15 , wherein the supervised learning objective is a binary evaluation objective and the reinforcement learning objective is a non-binary objective.