Multitask learning as question answering

1. A method for training a question answering system, the method comprising:

receiving a plurality of training samples, each of the training samples including a natural language context, a natural language question, and a natural language ground truth answer, wherein the plurality of training samples contains different training samples each corresponding to training the question answering system for a different task type from a plurality of task types;

presenting the plurality of training samples to a neural model to generate an answer;

determining an error between the generated answer and the natural language ground truth answer for each training sample presented; and

adjusting parameters of the neural model based on the error;

wherein the plurality of training samples are presented to the neural model by:

initially selecting a first set of training samples from the plurality of training samples according to a first training strategy that first selects samples corresponding to a first task type and subsequently selects samples corresponding to a second task type resulting in a first ordering of the first set of training samples that covers each of the plurality of task types; and

switching to selecting a second set of training samples from the plurality of training samples according to a second training strategy that mixes samples corresponding to the different task types, resulting in a second ordering of the second set of training samples that covers each of the plurality of task types.

2. The method of claim 1 , wherein each of the plurality of task types is a language translation task type, a classification task type, or a question answering task type.

3. The method of claim 1 , wherein the first training strategy is a sequential training strategy where each of the first set of training samples for the first task type are selected before selecting training samples of the second task type.

4. The method of claim 3 , wherein the sequential training strategy includes reselecting training samples for the first task type after selecting training samples for each of the plurality of task types.

5. The method of claim 1 , wherein the second training strategy is a joint training strategy where each of the second set of training samples are selected so that consecutively selected training samples are selected from different ones of the plurality of task types.

6. The method of claim 1 , wherein the second training strategy is a joint training strategy where each of the second set of training samples are selected so that consecutively selected small groups of training samples are selected from different ones of the plurality of task types.

7. The method of claim 1 , wherein the first training strategy is a modified sequential training strategy where the first set of training samples are selected according to a sequential training strategy with periodic intervals where the training samples are selected according to a joint training strategy.

8. The method of claim 1 , further comprising switching to selecting the second set of training samples using the second training strategy after each of the first set of training samples for each of the plurality of task types is presented to the neural model a predetermined number of times.

9. The method of claim 1 , further comprising switching to selecting the second set of training samples using the second training strategy based on monitoring of performance metrics associated with each of the plurality of task types.

10. The method of claim 1 , wherein the neural model comprises:

an input layer for encoding first words from the context and second words from the question;

a self-attention based transformer comprising an encoder and a decoder;

a bi-directional long-term short-term memory (biLSTM) for further encoding an output of the encoder;

a long-term short-term memory (LSTM) for generating a context-adjusted hidden state from an output of the decoder and a hidden state;

an attention network for generating attention weights based on an output of the encoder and the context-adjusted hidden state;

a vocabulary layer for generating a distribution over third words in a vocabulary based on the attention weights;

a context layer for generating a distribution over the first words from the context based on the attention weights; and

a switch for:

generating a weighting between the distribution over the third words from the vocabulary and the distribution over the first words from the context;

generating a composite distribution based on the weighting of the distribution over the third words from the vocabulary and the distribution over the first words from the context; and

selecting a word for inclusion in an answer using the composite distribution.

11. A non-transitory machine-readable medium comprising a plurality of machine-readable instructions which when executed by one or more processors associated with a computing device are adapted to cause the one or more processors to perform a method comprising.

receiving a plurality of training samples, each of the training samples including a natural language context, a natural language question, and a natural language ground truth answer, the training samples including training samples from a plurality of task types;

presenting the training samples to a neural model to generate an answer;

determining an error between the generated answer and the natural language ground truth answer for each training sample presented; and

adjusting parameters of the neural model based on the error;

wherein the plurality of training samples are presented to the neural model by:

initially selecting a first set of training samples from the plurality of training samples according to a first training strategy that first selects samples corresponding to a first task type and subsequently selects samples corresponding to a second task type resulting in a first ordering of the first set of training samples that covers each of the plurality of task types; and

switching to selecting a second set of training samples from the plurality of training samples according to a second training strategy that mixes samples corresponding to the different task types, resulting in a second ordering of the second set of training samples that covers each of the plurality of task types.

12. The non-transitory machine-readable medium of claim 11 , wherein the first training strategy is a sequential training strategy where each of the first set of training samples for the first task type are selected before selecting training samples of the second task type.

13. The non-transitory machine-readable medium of claim 11 , wherein the second training strategy is a joint training strategy where each of the second set of training samples are selected so that consecutively selected training samples are selected from different ones of the plurality of task types.

14. The non-transitory machine-readable medium of claim 11 , wherein the second training strategy is a joint training strategy where each of the second set of training samples are selected so that consecutively selected small groups of training samples are selected from different ones of the plurality of task types.

15. The non-transitory machine-readable medium of claim 11 , further comprising switching to selecting the second set of training samples using the second training strategy after each of the training samples for each of the plurality of task types is presented to the neural model a predetermined number of times.

16. A system for deep learning, the system comprising:

a multi-layer neural network;

wherein the system is configured to:

receive a plurality of training samples, each of the training samples including a natural language context, a natural language question, and a natural language ground truth answer, the training samples including training samples from a plurality of task types;

present the training samples to a neural model to generate an answer;

determine an error between the generated answer and the natural language ground truth answer for each training sample presented; and

adjust parameters of the neural model based on the error;

wherein the plurality of training samples are presented to the neural model by:

initially selecting a first set of training samples from the plurality of training samples according to a first training strategy that first selects samples corresponding to a first task type and subsequently selects samples corresponding to a second task type resulting in a first ordering of the first set of training samples that covers each of the plurality of task types; and

switching to selecting a second set of training samples from the plurality of training samples according to a second training strategy that mixes samples corresponding to the different task types, resulting in a second ordering of the second set of training samples that covers each of the plurality of task types.

17. The system of claim 16 , wherein the first training strategy is a sequential training strategy where each of the first set of training samples for the first task type are selected before selecting training samples of the second task type.

18. The system of claim 16 , wherein the second training strategy is a joint training strategy where each of the second set of training samples are selected so that consecutively selected training samples are selected from different ones of the plurality of task types.

19. The system of claim 16 , wherein the second training strategy is a joint training strategy where each of the second set of training samples are selected so that consecutively selected small groups of training samples are selected from different ones of the plurality of task types.

20. The system of claim 16 , wherein the system is further configured to select the second set of training samples using the second training strategy after each of the first set of training samples for each of the plurality of task types is presented to the neural model a predetermined number of times.