# US 2019/0244609 A1 · Adversarial learning and generation of dialogue responses

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0244609 A1
- Document ID: US20190244609A1
- Status: Published application
- Kind: A1
- Application number: 16/180,774
- Filed: 2018-11-05
- Published: 2019-08-08
- Earliest claimed: 2018-02-08
- Assignees: Capital One Services, LLC
- Applicants: Capital One Services, LLC
- Inventors: Oluwatobi OLABIYI; Erik T. Mueller
- Web page: https://patentgrep.com/patent/US20190244609A1

## CPC

- G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue
- G06F 40/35: Discourse or dialogue representation
- G06N 3/006: based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/0475: Generative networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 3/094: Adversarial learning
- G06N 20/00: Machine learning
- G10L 13/027: Concept to speech synthesisers; Generation of natural phrases from machine-based concepts (generation of parameters for speech synthesis out of text G10L13/08)
- G10L 15/26: Speech to text systems (G10L15/08 takes precedence)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Systems and methods for generating responses to user input such as dialogues, and images are discussed. The system may generate, by a response generation module of at least one server, an optimal generated response to the user communication by applying an generative adversarial network. In some embodiments, the generative adversarial network may include a hierarchical recurrent encoder decoder generative adversarial network including a generator and a discriminator component.

## Claims

### Claim 1

An improved method for generating responses to user communications comprising:
receiving, by at least one server communicatively coupled to a user device, a user communication;
applying, by a response generation module of the at least one server, a generative adversarial network to the user communication to generate an optimal generated response to the user communication, the generative adversarial network comprising a generator component and a discriminator component, the discriminator component comprising a convolutional neural network and a recurrent neural network, wherein applying the generative adversarial network comprises:
generating, via the generator component, a plurality of computer generated responses responsive to the user communication based on the received user communication, noise input, and historical data, and
selecting, via the discriminator component, an optimal computer generated response from among the generated plurality of computer generated responses; and
transmitting, by the at least one server, the optimal generated response to the user communication to a user of the user device.

### Claim 2

(canceled)

### Claim 3 (depends on 1)

The method of claim 1, further comprising training the generative adversarial network by:
providing training data to the generator component of the generative adversarial network,
generating, by the generator component, computer responses to the training data,
evaluating, by the discriminator component of the generative adversarial network, the generated computer responses by at least one comparison to the training data,
ranking, by the discriminator component, the generated computer responses, and
adjusting one or more parameters of the generator component based on the ranking of the generated computer responses.

### Claim 4 (depends on 3)

The method of claim 3, wherein the training data further comprises at least one of noise, dialogue history, conversations, and transcriptions.

### Claim 5 (depends on 3)

The method of claim 3, wherein ranking the generated computer responses further comprises using a stochastic gradient descent algorithm to reduce errors between the generated computer responses and a human user response.

### Claim 6 (depends on 1)

The method of claim 1, wherein the generator component further comprises a local attention component.

### Claim 7 (depends on 1)

The method of claim 1, wherein the optimal computer generated response is selected at least in part based on the discriminator component ranking the generated plurality of computer generated responses based on at least one of grammatical correctness, local responsivity, global responsivity, local relativity and global relativity.

### Claim 8 (depends on 1)

The method of claim 1, wherein at least one of the generator component and the discriminator component further comprises an encoder and a decoder.

### Claim 9 (depends on 8)

The method of claim 8, wherein the encoder is configured to encode at least one of a word, a sentence, grammar, and sentence construction.

### Claim 10

An improved method for providing computer generated responses to user communications comprising:
receiving, by a user interface of a user device, a user communication;
transmitting, by the user device, the user communication to at least one server communicatively coupled to the user device;
receiving, by the user device, a computer generated response to the user communication, wherein the computer generated response is generated by applying an generative adversarial network to the user communication, the generative adversarial network comprising a generator component and a discriminator component, the discriminator component comprising a convolutional neural network and a recurrent neural network, wherein applying the generative adversarial network comprises:
generating, via the generator component, a plurality of computer generated responses responsive to the user communication, and
selecting, via the discriminator component, an optimal computer generated response from among the plurality of computer generated responses; and
providing, by the user interface, the computer generated response to a user of the user device.

### Claim 11

(canceled)

### Claim 12 (depends on 10)

The method of claim 10, wherein the discriminator is configured to discriminate at both the word and sentence levels.

### Claim 13 (depends on 10)

The method of claim 10, wherein the computer generated response is of the same communication type as the user communication.

### Claim 14 (depends on 10)

The method of claim 10, wherein at least one of the generator component and the discriminator component further comprises an encoder and a decoder.

### Claim 15 (depends on 14)

The method of claim 14, wherein the encoder is configured to encode at least one of a word, a sentence, grammar, and sentence construction.

### Claim 16

An improved system for generating responses to user communications comprising:
at least one server communicatively coupled to a user device by a network, wherein the at least one server further comprises a non-transitory memory storing computer-readable instructions and at least one processor;
the execution of the computer-readable instructions causing the at least one server to:
receive, from the user device, a user communication;
apply, by a response generation module of the at least one server, an generative adversarial network to the user communication to generate an optimal generated response to the user communication, the generative adversarial network comprising a generator component and a discriminator component, the discriminator component comprising a convolutional neural network and a recurrent neural network, and wherein application of the generative adversarial network further comprises:
generation, via the generator component, of a plurality of computer generated responses responsive to the user communication, and
selection, via the discriminator component, of an optimal computer generated response form among the plurality of computer generated responses; and
transmit the optimal generated response to the user communication to a user of the user device.

### Claim 17

(canceled)

### Claim 18 (depends on 16)

The system of claim 16, wherein the generator component and the discriminator component are trained by the steps of:
providing training data to the generator component of the generative adversarial network,
generating, by the generator component, computer responses to the training data,
evaluating, by the discriminator component of the generative adversarial network, the generated computer responses by at least one comparison to the training data,
ranking, by the discriminator component, the generated computer responses, and
adjusting one or more parameters of the generator component based on the ranking of the generated computer responses.

### Claim 19 (depends on 18)

The system of claim 18, wherein the optimal computer generated response is selected at least in part based on the discriminator component ranking the generated plurality of computer generated responses based on at least one of grammatical correctness, local responsivity, global responsivity, local relativity and global relativity.

### Claim 20 (depends on 16)

The system of claim 16, wherein at least one of the generator component and the discriminator component further comprises an encoder and a decoder.
