# US 2023/0274149 A1 · Predictive System Based on a Semantically Trained Artificial Neural Network and ANN

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

- Number: US 2023/0274149 A1
- Document ID: US20230274149A1
- Status: Published application
- Kind: A1
- Application number: 18/143,802
- Filed: 2023-05-05
- Published: 2023-08-31
- Earliest claimed: 2019-04-03
- Assignees: Emotional Perception Ai Limited
- Applicants: Emotional Perception Ai Limited
- Inventors: Joseph Michael William LYSKE; Nadine KROHER; Angelos PIKRAKIS
- Web page: https://patentgrep.com/patent/US20230274149A1

## CPC

- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06F 16/55: Clustering; Classification
- G06F 16/65: Clustering; Classification
- G06F 16/9024: {Graphs; Linked lists (G06F16/9027 takes precedence)}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06F 18/22: Matching criteria, e.g. proximity measures
- G06F 40/30: Semantic analysis
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06V 30/274: {Syntactic or semantic context, e.g. balancing}
- G10H 1/0008: {Associated control or indicating means}
- G10L 15/16: using artificial neural networks
- G06N 3/045: Combinations of networks
- G06N 3/048: Activation functions
- G10H 2210/071: for rhythm pattern analysis or rhythm style recognition
- G10H 2210/081: for automatic key or tonality recognition, e.g. using musical rules or a knowledge base
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G10H: ELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

A property vector representing extractable measurable properties, such as musical properties, of a file is mapped to semantic properties for the file. This is achieved by using artificial neural networks “ANNs” in which weights and biases are trained to align a distance dissimilarity measure in property space for pairwise comparative files back towards a corresponding semantic distance dissimilarity measure in semantic space for those same files. The result is that, once optimised, the ANNs can process any file, parsed with those properties, to identify other files sharing common traits reflective of emotional-perception, thereby rendering a more liable and true-to-life result of similarity/dissimilarity. This contrasts with simply training a neural network to consider extractable measurable properties that, in isolation, do not provide a reliable contextual relationship into the real-world.

## Claims

### Claim 1

A predictive system comprising:
a processor-based artificial neural network (ANN) configured by training the ANN by subjecting the ANN to a multiplicity of pairs of training data files sharing a common modality,
wherein independent vectors representing both human-perceived semantic meaning derived from natural language processing (NLP) of descriptive content for each one of a pair of training data files and derivable measurable qualities extracted from media content of each one of the pair of training data files provide separation distance measure respectively in semantic embedding space and property embedding space, and
wherein the configuration of the ANN through training thereof maps pairwise similarity in property embedding space for each pair of training data files towards corresponding pairwise semantic similarity in semantic embedding space for each pair of training data files thereby configuring the resulting ANN to value measures of semantic similarity over measures of property similarity through said convergence during training of property separation distances towards semantic separation distances,
wherein the predictive system, once trained on said pairs of training data files, is arranged to produce a measured quantification, expressed in property space, of similarity and/or dissimilarity in media content between a critiqued data file presented to the predictive system and processed by the ANN and a reference file processed by the ANN, and
wherein the quantification is based on extracted file properties extracted from each of the critiqued data file and the reference file.

### Claim 2 (depends on 1)

The predictive system of claim 1, further including an output from which the measured quantification is presented.

### Claim 3 (depends on 2)

The predictive system of claim 2, further including:
a database storing a multiplicity of reference files; and
system intelligence arranged to generate, at the output, a reference file recommendation selected from the database,
wherein the reference file recommendation is selected by the system intelligence to reflect semantic similarity to the critiqued data file as determined by quantitatively measured closeness of respective property vectors obtained from respective extracted file properties.

### Claim 4 (depends on 3)

The predictive system of claim 3, wherein human-expressed descriptive semantic content is coded into metadata of each data file.

### Claim 5 (depends on 3)

The predictive system of claim 3, further comprising:
processing intelligence arranged selectively to extract a plurality of identified qualifies from measurable signal properties of the critiqued data file,
wherein, for instances where the critiqued data file contains audio and the properties are musical properties, the measurable signal properties define qualities relating to at least one of rhythm, tonality, timbre and musical texture.

### Claim 6 (depends on 1)

The predictive system of claim 1, further comprising processing intelligence arranged:
for each file in the training set, to identify smallest and largest separation distances relative to other files in the training set;
for each file in the training set, to create a set of pairs in which a first pair has the smallest separation distance and a second pair has the largest separation distance; and
to assign a first value representing semantic closeness to the first pair and assigning a second value representing semantic dissimilarity to the second pair, wherein the second value is different to the first value.

### Claim 7 (depends on 6)

The predictive system of claim 6, wherein for a subset comprising m smallest separation distances and m largest separation distances, the processing intelligence is arranged to assign the first value to the m smallest and the second value to the m largest, where m is a positive integer less than N.

### Claim 8

A processor-based artificial neural network (ANN) comprising:
an output level capable of generating a property vector in property space;
an input layer; and
a plurality of intermediate hidden layers separating the output level from the input layer, wherein the input layer and the hidden layers each contain a plurality of weight-adjustable neurons and where neural pathways are established between the plurality of neurons in the input layer to the output level based on weighting factors applied by neurons in the input layer and the hidden layers;
wherein:
the neural pathways and adjustable weights of the neurons are configured based on a training regime for the ANN that:
subjects the ANN to a multiplicity of pairs of training data files sharing a common modality and where for each pair of training data files there are two independently derived separation distances, namely:
a first independently derived separation distance that expresses a measure of relative distance between a first pair of training data files in semantic embedding space, where the first independently derived separation distance is obtained from natural language processing “NLP” of a semantic description of the nature of the data associated with each one of the first pair of training data files; and
a second independently derived separation distance that expresses a measure of relative distance similarity between the first pair of training data files in property embedding space, where the second independently derived separation distance is a property distance derived from measurable qualities extracted from each one of the first pair of training data files, and
wherein training of the ANN by a backpropagation process uses output vectors generated at the output level of the ANN from processing of said multiplicity of pairs to adjust weighting factors to adapt iteratively the ANN during training to converge distances of generated output vectors, in property embedding space, towards corresponding pairwise semantic distances in semantic space, and
wherein the common modality relates to one of comparable: (i) video data files to video data files; or alternatively (ii) audio data files to audio data files; or alternatively (iii) static image files to static image files; or alternatively (iv) text files to text files.

### Claim 9 (depends on 8)

The processor-based ANN of claim 8, wherein for target data, when in the modality of an audio data file, the properties are musical properties and the measurable signal qualities are measurable properties indicative of one of rhythm, tonality, timbre and musical texture.