Memory-based reinforcement learning method for storing optional information in streaming data and system therefore
Provided is a memory-based reinforcement learning method and system capable of storing optional information in streaming data. A question-answering (QA) method using memory-based reinforcement learning method includes receiving, in an episodic memory reader (EMR), streaming data about an input context that is input from a user; analyzing, in the EMR, the received streaming data and storing preset semantic information used for QA in an external memory; and, in response to an input of a question front the user, determining, in a pretrained QA model, an answer to the input question based on semantic information stored in the external memory.
1. A question-answering (QA) method using memory-based reinforcement learning, the method comprising:
receiving, in an episodic memory reader (EMR), streaming data about an input context that is input from a user;
analyzing, in the EMR, the received streaming data and storing preset semantic information used for QA in an external memory, wherein the storing comprises, in response to an empty memory entry of the external memory, mapping and storing the semantic information to the empty memory entry, and, in response to a full memory entry of the external memory, computing relative importance between the semantic information and data stored in the external memory and replacing specific data stored in a memory entry corresponding to lowest relative importance compared to the semantic information with the semantic information and storing the replaced semantic data; and,
in response to an input of a question from the user, determining, in a pretrained QA model, an answer to the input question based on semantic information stored in the external memory.
2. The method of claim 1 , wherein the storing comprises encoding each data instance comprising the streaming data to a memory vector representation, acquiring relative importance with data stored in the external memory using the encoded memory vector representation, and, in response to the acquired relative importance being higher than specific data stored in the external memory, replacing the specific data with data about the semantic information and storing the replaced data.
3. The method of claim 1 , wherein the storing comprises computing a replacement probability between the semantic information and each memory entry of the external memory based on a preset memory scheduling agent policy and replacing data stored in one of memory entries with the semantic information based on the computed replacement probability and storing the replaced semantic information.
4. The method of claim 1 , wherein the storing comprises storing the semantic information in the external memory through a reinforcement learning model that is trained through reinforcement learning using a preset training dataset.
5. A question-answering (QA) method using memory-based reinforcement learning, the method comprising:
receiving, in an episodic memory reader (EMR), streaming data about an input context that is input from a user;
analyzing, in the EMR, the received streaming data and acquiring preset semantic information used for QA, and replacing a single piece of specific data of data stored in an external memory with the semantic information and storing the replaced semantic information, wherein the storing comprises, in response to an empty memory entry of the external memory, mapping and storing the semantic information to the empty memory entry, and, in response to a full memory entry of the external memory, computing relative importance between the semantic information and data stored in the external memory and replacing specific data stored in a memory entry corresponding to lowest relative importance compared to the semantic information with the semantic information and storing the replaced semantic data; and,
in response to an input of a question from the user, determining, in a pretrained QA model, an answer to the input question based on semantic information stored in the external memory.
6. The method of claim 5 , wherein the storing comprises encoding each data instance comprising the streaming data to a memory vector representation, acquiring relative importance with data stored in the external memory using the encoded memory vector representation, and, in response to the acquired relative importance being higher than the single piece of specific data stored in the external memory, replacing the specific data with data about the semantic information and storing the replaced data.
7. A question-answering (QA) system using memory-based reinforcement learning, the system comprising:
a receiver configured to receive, in an episodic memory reader (EMR),
streaming data about input context that is input from a user;
a storage configured to analyze, in the EMR, the received streaming data and store preset semantic information used for QA in an external memory, wherein the storage is configured to, in response to an empty memory entry of the external memory, map and store the semantic information to the empty memory entry, and to, in response to a full memory entry of the external memory, compute relative importance between the semantic information and data stored in the external memory and replace specific data stored in a memory entry corresponding to lowest relative importance compared to the semantic information with the semantic information and store the replaced semantic data; and
a determiner configured to, in response to an input of a question from the user, determine, in a pretrained QA model, an answer to the input question based
on semantic information stored in the external memory.
8. The system of claim 7 , wherein the storage is configured to encode each data instance comprising the streaming data to a memory vector representation, to acquire relative importance with data stored in the external memory using the encoded memory vector representation, and to, in response to the acquired relative importance being higher than specific data stored in the external memory, replace the specific data with data about the semantic information and store the replaced data.
9. The system of claim 7 , wherein the storage is configured to compute a replacement probability between the semantic information and each memory entry of the external memory based on a preset memory scheduling agent policy and to replace data stored in one of memory entries with the semantic information based on the computed replacement probability and store the replaced semantic information.
10. The system of claim 7 , wherein the storage is configured to store the semantic information in the external memory through a reinforcement learning model that is trained through reinforcement learning using a preset training dataset.