System and method for deep memory network
AssigneeSamsung Electronics Co., Ltd.
InventorsYilin Shen, Yue Deng, Avik Ray, Hongxia Jin
Life of this patent
Estimating the term…
Abstract
An electronic device including a deep memory model includes at least one memory and at least one processor coupled to the at least one memory. The at least one processor is configured to receive input data to the deep memory model. The at least one processor is also configured to extract a history state of an external memory coupled to the deep memory model based on the input data. The at least one processor is further configured to update the history state of the external memory based on the input data. In addition, the at least one processor is configured to output a prediction based on the extracted history state of the external memory.
Claims20
A method for managing a deep memory model Coupled to an external memory, the method comprising:
- receiving input data to the deep memory model;
- extracting a history state of the external memory based on the input data, wherein the external memory includes:
- a first set of memory cells for storing content item state data including data based on previously provided content items processed by the deep memory model;
- and
- a second set of memory cells for storing reward state data associated with the content item state data, the reward state data including one or more rewards corresponding to the previously provided content items, wherein the one or more rewards are used by the deep memory model to minimize a regret parameter associated with the previously provided content items;
- updating the history state of the external memory based on the input data by balancing most relevant read locations with least used locations, wherein the balancing includes:
- determining a set of weights least relevant to the input data;
- setting a last read weight vector to provide that the external memory is filled before any overwriting of the external memory;
- and
- defining write weights for writing the input data to the external memory based on the last read weight vector and the set of weights least relevant to the input data;
- and
- outputting a prediction based on the extracted history state of the external memory.
An electronic device including a deep memory model, the electronic device comprising:
- at least one memory;
- and
- at least one processor coupled to the at least one memory, the at least one processor configured to:
- receive input data to the deep memory model;
- extract a history state of an external memory coupled to the deep memory model based on the input data, wherein the external memory includes:
- a first set of memory cells for storing content item state data including data based on previously provided content items processed by the deep memory model;
- and
- a second set of memory cells for storing reward state data associated with the content item state data, the reward state data including one or more rewards corresponding to the previously provided content items, wherein the one or more rewards are used by the deep memory model to minimize a regret parameter associated with the previously provided content items;
- update the history state of the external memory based on the input data by balancing most relevant read locations with least used locations, wherein, to balance the most relevant read locations with the least used locations, the at least one processor is configured to:
- determine a set of weights least relevant to the input data;
- set a last read weight vector to provide that the external memory is filled before any overwriting of the external memory;
- and
- define write weights for writing the input data to the external memory based on the last read weight vector and the set of weights least relevant to the input data;
- and
- output a prediction based on the extracted history state of the external memory.
A non-transitory computer readable medium embodying a computer program, the computer program comprising instructions that when executed cause at least one processor of an electronic device to:
- receive input data to a deep memory model;
- extract a history state of an external memory coupled to the deep memory model based on the input data, wherein the external memory includes:
- a first set of memory cells for storing content item state data including data based on previously provided content items processed by the deep memory model;
- and
- a second set of memory cells for storing reward state data associated with the content item state data, the reward state data including one or more rewards corresponding to the previously provided content items, wherein the one or more rewards are used by the deep memory model to minimize a regret parameter associated with the previously provided content items;
- update the history state of the external memory based on the input data by balancing most relevant read locations with least used locations, wherein the instructions that when executed cause the at least one processor to balance the most relevant read locations with the least used locations comprise instructions that when executed cause the at least one processor to:
- determine a set of weights least relevant to the input data;
- set a last read weight vector to provide that the external memory is filled before any overwriting of the external memory;
- and
- define write weights for writing the input data to the external memory based on the last read weight vector and the set of weights least relevant to the input data;
- and
- output a prediction based on the extracted history state of the external memory.
Dependencies are inferred from the claim text.