# US 2021/0174562 A1 · System and methods for feature relevance visualization optimization and filtering for explainability in ai-based alert detection and processing systems

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

- Number: US 2021/0174562 A1
- Document ID: US20210174562A1
- Status: Published application
- Kind: A1
- Application number: 16/706,518
- Filed: 2019-12-06
- Published: 2021-06-10
- Earliest claimed: 2019-12-06
- Assignees: Bank of America Corporation
- Applicants: Bank of America Corporation
- Inventors: Eren Kursun
- Web page: https://patentgrep.com/patent/US20210174562A1

## CPC

- G06T 11/26: Drawing of charts or graphs
- G06N 3/00: Computing arrangements based on biological models
- G06N 3/02: Neural networks
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/09: Supervised learning
- G06N 5/045: Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
- G06N 20/00: Machine learning
- G06T 11/00: Two-dimensional [2D] image generation
- G06T 11/10: Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77)
- G06T 11/40: Filling planar surfaces by adding surface attributes, e.g. adding colours or textures
- G06T 19/00: Manipulating three-dimensional [3D] models or images for computer graphics
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A system for feature relevance visualization optimization is provided. The system comprises a controller configured for modifying placement of features in a relevance visualization. The controller is further configured to: receive interaction data comprising one or more features positioned in the relevance visualization, wherein the one or more features are defined and measurable properties of the interaction data; construct a logical grouping of the one or more features based on a type of each of the one or more features, wherein similar features are collocated in the relevance visualization; construct a machine learning-based grouping of the one or more features based on relevance calculations for the one or more features; combine the logical grouping and the machine learning-based grouping to generate a combined feature placement, wherein the one or more features are repositioned in the relevance visualization; and output the relevance visualization having the combined feature placement.

## Claims

### Claim 1

A system for feature relevance visualization optimization, the system comprising:
a controller configured for modifying placement of features in a relevance visualization, the controller comprising a memory device with computer-readable program code stored thereon, a communication device connected to a network, and a processing device, wherein the processing device is configured to execute the computer-readable program code to:
receive interaction data comprising one or more features positioned in the relevance visualization, wherein the one or more features are defined and measurable properties of the interaction data;
construct a logical grouping of the one or more features based on a type of each of the one or more features, wherein similar features are collocated in the relevance visualization;
construct a machine learning-based grouping of the one or more features based on relevance calculations for the one or more features;
combine the logical grouping and the machine learning-based grouping to generate a combined feature placement, wherein the one or more features are repositioned in the relevance visualization; and
output the relevance visualization having the combined feature placement.

### Claim 2 (depends on 1)

The system of claim 1, wherein the processing device is further configured to execute the computer-readable program code to iteratively refine the combined feature placement based on changes in the interaction data.

### Claim 3 (depends on 1)

The system of claim 1, wherein the processing device is further configured to execute the computer-readable program code to receive analyst input on a placement of the one or more features in the relevance visualization, wherein the combined feature placement is at least partially based on the analyst input.

### Claim 4 (depends on 1)

The system of claim 1, wherein the machine learning-based grouping is further based on at least one of historical interaction data, streaming interaction data, analyst input, and misappropriation data.

### Claim 5 (depends on 1)

The system of claim 1, wherein repositioning the one or more features in the relevance visualization further comprises regrouping the one or more features in the combined feature placement.

### Claim 6 (depends on 1)

The system of claim 1, wherein repositioning the one or more features in the relevance visualization further comprises modifying a placement of one or more pixels in an image of the relevance visualization.

### Claim 7 (depends on 6)

The system of claim 6, wherein the processing device is further configured to execute the computer-readable program code to modify a spectrum of colors of the one or more pixels in the image of the relevance visualization based on the relevance calculations, wherein the relevance visualization is a feature relevance heat map.

### Claim 8 (depends on 6)

The system of claim 6, wherein the processing device is further configured to execute the computer-readable program code to generate a highlighted region around a portion of the one or more features in the image of the relevance visualization based on the relevance calculations.

### Claim 9 (depends on 1)

The system of claim 1, wherein the relevance visualization is a two-dimensional or three-dimensional data plot.

### Claim 10 (depends on 1)

The system of claim 1, wherein the relevance calculations and the relevance visualization are initially generated by a neural network configured to receive and process the interaction data.

### Claim 11 (depends on 10)

The system of claim 10, wherein the neural network is a convolutional neural network.

### Claim 12 (depends on 1)

The system of claim 1 further comprising a machine learning model configured for user type-specific pattern recognition based on the interaction data associated with a specific user type, wherein the relevance visualization is modified based on a user type-specific pattern.

### Claim 13 (depends on 12)

The system of claim 12, wherein modifying the relevance visualization based on the user type-specific pattern comprises applying a compensating filter to the relevance visualization based on the user type-specific pattern.

### Claim 14

A system for machine learning-based neural network output correction, the system comprising:
a controller configured for calculating and applying a compensating block to a neural network-derived output, the controller comprising a memory device with computer-readable program code stored thereon, a communication device connected to a network, and a processing device, wherein the processing device is configured to execute the computer-readable program code to:
calculate, using a neural network comprising a plurality of non-linear nodal layers, a feed-forward scoring of an input of interaction data comprising one or more features;
calculate, using layer-wise relevance propagation, a layer-wise relevance for the one or more features of the feed-forward scoring, wherein an output of the layer-wise relevance propagation is a relevance visualization of the one or more features;
identify, using a machine learning model, one or more side effects of the neural network from the layer-wise relevance propagation, wherein the one or more side effects are indicative of a neural network architecture type;
based on the neural network architecture type, calculate a compensating block for the output of the layer-wise relevance propagation, wherein the compensating block removes the one or more side effects from the layer-wise relevance propagation; and
apply the compensating block to the relevance visualization.

### Claim 15 (depends on 14)

The system of claim 14, wherein the compensating block comprises batch normalization of the layer-wise relevance propagation.

### Claim 16 (depends on 14)

The system of claim 14, wherein the one or more side effects comprise errors associated with at least one of an activation, a weighting, or an assumption of the neural network.

### Claim 17 (depends on 14)

The system of claim 14, wherein the relevance visualization is a displayed bitmap, and wherein applying the compensating block to the relevance visualization further comprises applying the compensating block as a filter to the displayed bitmap.

### Claim 18 (depends on 14)

The system of claim 14, wherein a calculation of the layer-wise relevance propagation comprises a Padé approximation.

### Claim 19

A computer-implemented method for machine learning-based neural network output correction, the computer-implemented method comprising:
calculating, using a neural network comprising a plurality of non-linear nodal layers, a feed-forward scoring of an input of interaction data comprising one or more features, the feed-forward scoring identifying potential misappropriation in the interaction data;
calculating, using layer-wise relevance propagation, a layer-wise relevance for the one or more features of the feed-forward scoring, the layer-wise relevance propagation comprising a Padé approximation, wherein the layer-wise relevance propagation is batch normalized;
based on the layer-wise relevance propagation, generating a relevance visualization map for the potential misappropriation in the interaction data; and
displaying the relevance visualization map.

### Claim 20 (depends on 19)

The computer-implemented method of claim 19 further comprising:
identifying, using a machine learning model, one or more side effects of the neural network from the layer-wise relevance propagation, wherein the one or more side effects comprise errors associated with at least one of an activation, a weighting, or an assumption of the neural network, and wherein the one or more side effects are indicative of a neural network architecture type;
based on the neural network architecture type, calculating a compensating block for an output of the layer-wise relevance propagation, wherein the compensating block removes the one or more side effects from the layer-wise relevance propagation; and
applying the compensating block to the relevance visualization map.
