Apparatus and method for providing fashion coordination knowledge based on neural network having explicit memory

Patent metadata

Field	Value
Publication number	US-2020219166-A1
Application number	US-201916711934-A
Country	US
Kind code	A1
Filing date	Dec 12, 2019
Priority date	Jan 8, 2019
Publication date	Jul 9, 2020
Grant date	—

Abstract

A method and apparatus for estimating a user's requirement through a neural network which are capable of reading and writing a working memory and for providing fashion coordination knowledge appropriate for the requirement through the neural network using a long-term memory, by using the neural network using an explicit memory, in order to accurately provide the fashion coordination knowledge. The apparatus includes a language embedding unit for embedding a user's question and a previously created answer to acquire a digitized embedding vector; a fashion coordination knowledge creation unit for creating fashion coordination through the neural network having the explicit memory by using the embedding vector as an input; and a dialog creation unit for creating dialog content for configuring the fashion coordination through the neural network having the explicit memory by using the fashion coordination knowledge and the embedding vector an input.

First claim

What is claimed is:

1 . An apparatus for providing fashion coordination knowledge based on a neural network having an explicit memory, the apparatus comprising:
a language embedding unit configured to embed a user's question and a previously created answer to acquire a digitized embedding vector; a fashion coordination knowledge creation unit configured to create fashion coordination knowledge through the neural network having the explicit memory by using the embedding vector acquired by the language embedding unit as an input; and a dialog creation unit configured to create dialog content for configuring the fashion coordination through the neural network having the explicit memory by using the fashion coordination knowledge acquired from the fashion coordination knowledge creation unit and the embedding vector acquired by the language embedding unit as an input.

2 . The apparatus of claim 1 , wherein the dialog content for configuring the fashion coordination, which is created by the dialog creation unit, includes at least one of a request for information to be added to the fashion coordination and an answer for explaining new fashion coordination.

3 . The apparatus of claim 1 , wherein the fashion coordination knowledge creation unit comprises:
a working memory, which is a place for memorizing previous questions and answers; a long-term memory, which is a place for memorizing a feature of a fashion item; a reading unit configured to calculate a value of the working memory to be read; a writing unit configured to delete and add a value of the working memory; a requirement estimation unit configured to create parameters necessary to access the working memory using the embedding vector acquired by the language embedding unit and configured to estimate a requirement vector using the value of the working memory acquired from the reading unit; and a category-specific fashion item creation unit configured to classify fashion items according to predetermined categories and create a fashion item for each of the categories using the long-term memory and the requirement vector acquired from the requirement estimation unit.

4 . The apparatus of claim 3 , wherein the reading unit calculates a weight for a position of the working memory to be read and linearly combines the value of the working memory through a medium of the weight by using parameters acquired from the requirement estimation unit in order to calculate the value of the working memory to be read.

5 . The apparatus of claim 3 , wherein the writing unit calculates a weight for a position of the working memory to be written and deletes and adds the value of the working memory according to the weight by using parameters acquired from the requirement estimation unit in order to delete and add the value of the working memory.

6 . The apparatus of claim 3 , wherein the predetermined categories determined by the category-specific fashion item creation unit are one or more categories corresponding to fashion item wearing positions.

7 . The apparatus of claim 3 , wherein the category-specific fashion item creation unit comprises:
a fashion item probability calculation unit configured to calculate a fashion item probability appropriate for a requirement by using the long-term memory and the requirement vector acquired from the requirement estimation unit; a fashion coordination evaluation unit configured to perform replacement of the fashion item and evaluate newly configured fashion coordination by using the long-term memory, previously created fashion coordination, and the requirement vector acquired from the requirement estimation unit; and a fashion item determination unit configured to determine the fashion item from the fashion item probability acquired from the fashion item probability calculation unit and a fashion coordination evaluation result acquired from the fashion coordination evaluation unit.

8 . The apparatus of claim 7 , wherein the fashion item is determined by the fashion item determination unit multiplying the fashion item probability and the fashion coordination evaluation result and then finding a maximum value.

9 . The apparatus of claim 3 , further comprising a value estimation unit configured to estimate a value using the neural network having the explicit memory by using the requirement vector acquired from the requirement estimation unit, the fashion coordination acquired from the fashion coordination knowledge creation unit, and answer data acquired from the dialog creation unit.

10 . The apparatus of claim 9 , wherein the neural network of the fashion coordination knowledge creation unit receives and learns the value estimated by the value estimation unit and the fashion coordination knowledge.

11 . The apparatus of claim 9 , wherein the neural network of the value estimation unit performs learning using a difference between the estimated value and training reward data.

12 . The apparatus of claim 9 , wherein the neural network of the dialog creation unit performs learning using the fashion coordination and a dialog creation probability acquired from the dialog creation unit.

13 . A method of providing fashion coordination knowledge based on a neural network having an explicit memory, the method comprising:
embedding a user's question and a previously created answer to acquire a digitized embedding vector; creating fashion coordination knowledge through the neural network having the explicit memory by using the embedding vector as an input; and creating dialog content for configuring fashion coordination through the neural network having the explicit memory by using the embedding vector and the created fashion coordination knowledge as an input.

14 . The method of claim 13 , wherein the creating of the fashion coordination knowledge comprises:
creating parameters necessary to access a working memory, which is a place for memorizing previous questions and answers, using the embedding vector and estimating a requirement vector; and classifying fashion items according to predetermined categories and creating a fashion item for each of the categories using the requirement vector acquired from a requirement estimation unit and a long-term memory, which is a place for memorizing a feature of the fashion item.

15 . The method of claim 14 , wherein the creating of the fashion item for each of the categories comprises:
calculating a probability of a fashion item appropriate for a requirement by using the long-term memory and the requirement vector; performing replacement of the fashion item and evaluating newly configured fashion coordination by using the long-term memory, previously created fashion coordination, and the requirement vector; and determining the fashion item from the fashion item probability and a fashion coordination evaluation result.

16 . The method of claim 13 , further comprising estimating a value using the neural network having the explicit memory by using a requirement vector, the fashion coordination, and the created dialog content.

Assignees

• Electronics & Telecommunications Res Inst

Inventors

• Kim Hyun Woo
• Song Hwa Jeon
• Chung Eui Sok
• Jung Ho Young
• Park Jeon Gue
• Lee Yun Keun

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/092

  Reinforcement learning · CPC title