THE TORONTO-DOMINION BANK patent applications published on November 30th, 2023

Summary of the patent applications from THE TORONTO-DOMINION BANK on November 30th, 2023

The Toronto-Dominion Bank has recently filed several patents related to computer models, augmented reality, ambient commerce systems, personalized notifications, data storage, training models without sharing private data, and probability density modeling. These patents demonstrate the bank's focus on innovation and technology in various areas.

In summary, the bank's recent patent filings include:

- A computer model that can account for data samples in a high-dimensional space as lying on different manifolds. The model treats the data set as a union of manifolds and groups data samples expected to belong to the same underlying manifold. It also includes information on how frequently to sample from each sub-model to accurately represent the entire data set.

- A system that can send a stored-value card to a recipient's mobile device for display in augmented reality. The system generates a three-dimensional object representing the stored-value card based on parameters and sends it to the recipient's mobile device for augmented reality display.

- An ambient commerce system that detects unauthenticated entities at ambient commerce premises using sensors and performs commerce operations for authenticated entities based on indications from a trusted system.

- A method for providing personalized notifications to users while browsing web pages associated with specific merchants. The method generates personalized notifications based on historical transactions data and resource allocation, and provides them to the user's computing device for display in a browser application.

- A system and method for storing data generated during the execution of a process workflow. The system receives and stores messages exchanged during the workflow in a database based on a database schema, and retrieves specific message properties in response to read requests.

- A system for training models without sharing private data sets. The system updates client weights based on how well sampled models represent the private data set, calculates gradients for each sampled model, and weights them according to the client weight for that model.

- A method for modeling probability densities of data on a high-dimensional manifold using an implicitly-defined manifold and an energy function. The method trains an energy function to learn a probability density for the manifold and filters it through the defined manifold for more effective modeling.

- A model evaluation system that assesses how privacy-aware training processes impact the training gradients for different groups. The system adjusts per-sample gradients based on reference and clipping bounds, calculates a scaling factor for each gradient, and determines the relative privacy cost between groups.

Notable applications:

• Computer model for data samples on different manifolds.
• Sending stored-value cards for augmented reality display.
• Ambient commerce system for authenticated entities.
• Personalized notifications based on historical transactions.
• Data storage during process workflows.
• Training models without sharing private data.
• Probability density modeling on high-dimensional manifolds.
• Model evaluation system for privacy-aware training processes.

Patent applications for THE TORONTO-DOMINION BANK on November 30th, 2023

IDENTIFYING AND MITIGATING DISPARATE GROUP IMPACT IN DIFFERENTIAL-PRIVACY MACHINE-LEARNED MODELS (18202435)

Main Inventor

Jesse Cole Cresswell

Brief explanation

- The patent application describes a model evaluation system that assesses how privacy-aware training processes impact the training gradients for different groups.

- The system uses a modified differential-privacy training process that adjusts per-sample gradients based on reference and clipping bounds. - A scaling factor is determined for each per-sample gradient based on the higher value between the reference bound and the magnitude of the gradient. - The per-sample gradients are then adjusted using a ratio of the clipping bound to the scaling factor. - The system also calculates a relative privacy cost between groups by comparing the group gradient direction to an unadjusted batch gradient and the adjusted batch gradient. - The innovation aims to improve privacy-aware training processes by evaluating their impact on training gradients and determining the relative privacy cost for different groups.

Abstract

A model evaluation system evaluates the extent to which privacy-aware training processes affect the direction of training gradients for groups. A modified differential-privacy (“DP”) training process provides per-sample gradient adjustments with parameters that may be adaptively modified for different data batches. Per-sample gradients are modified with respect to a reference bound and a clipping bound. A scaling factor may be determined for each per-sample gradient based on the higher of the reference bound or a magnitude of the per-sample gradient. Per-sample gradients may then be adjusted based on a ratio of the clipping bound to the scaling factor. A relative privacy cost between groups may be determined as excess training risk based on a difference in group gradient direction relative to an unadjusted batch gradient and the adjusted batch gradient according to the privacy-aware training.

IDENTIFYING AND MITIGATING DISPARATE GROUP IMPACT IN DIFFERENTIAL-PRIVACY MACHINE-LEARNED MODELS (18202440)

Main Inventor

Jesse Cole Cresswell

Brief explanation

- The patent application is for a model evaluation system that assesses how privacy-aware training processes impact the training gradients for different groups.

- The system uses a modified differential-privacy training process that adjusts per-sample gradients based on reference and clipping bounds. - The scaling factor for each per-sample gradient is determined by comparing the reference bound and the magnitude of the gradient. - The per-sample gradients are then adjusted based on the ratio of the clipping bound to the scaling factor. - The system also calculates a relative privacy cost between groups by comparing the group gradient direction to an unadjusted batch gradient and the adjusted batch gradient. - The privacy-aware training process helps maintain privacy while training models.

Abstract

A model evaluation system evaluates the extent to which privacy-aware training processes affect the direction of training gradients for groups. A modified differential-privacy (“DP”) training process provides per-sample gradient adjustments with parameters that may be adaptively modified for different data batches. Per-sample gradients are modified with respect to a reference bound and a clipping bound. A scaling factor may be determined for each per-sample gradient based on the higher of the reference bound or a magnitude of the per-sample gradient. Per-sample gradients may then be adjusted based on a ratio of the clipping bound to the scaling factor. A relative privacy cost between groups may be determined as excess training risk based on a difference in group gradient direction relative to an unadjusted batch gradient and the adjusted batch gradient according to the privacy-aware training.

LEARNED DENSITY ESTIMATION WITH IMPLICIT MANIFOLDS (18202450)

Main Inventor

Jesse Cole Cresswell

Brief explanation

The patent application describes a method for modeling probability densities of data on a high-dimensional manifold using an implicitly-defined manifold and an energy function.

• Probability density modeling for data on a high-dimensional manifold is performed using an implicitly-defined manifold.
• The manifold is defined as the zero set of a manifold-defining function.
• An energy function is trained to learn a probability density for the manifold.
• The energy function is filtered through the defined manifold for training and in application.
• The combined energy function and manifold-defining function create an "energy-based implicit manifold" for more effective modeling.
• The method avoids distortions caused by modeling the manifold in a lower-dimensional space.

Abstract

Probability density modeling, such as for generative modeling, for data on a manifold of a high-dimensional space is performed with an implicitly-defined manifold such that points belonging to the manifold is the zero set of a manifold-defining function. An energy function is trained to learn an energy function that, evaluated on the manifold, describes a probability density for the manifold. As such, the relevant portions of the energy function are “filtered through” the defined manifold for training and in application. The combined energy function and manifold-defining function provide an “energy-based implicit manifold” that can more effectively model probability densities of a manifold in the high-dimensional space. As the manifold-defining function and the energy function are defined across the high-dimensional space, they may more effectively learn geometries and avoid distortions due to change in dimension that occur for models that model the manifold in a lower-dimensional space.

DISTRIBUTED MODEL TRAINING WITH COLLABORATION WEIGHTS FOR PRIVATE DATA SETS (18202459)

Main Inventor

Jesse Cole Cresswell

Brief explanation

The patent application describes a system for training models without sharing private data sets.

• Private data sets learn client weights for computer models during training.
• Inference for a specific data set is determined by a combination of model parameters based on client weights.
• Client weights are updated based on how well sampled models represent the private data set.
• Gradients are calculated for each sampled model and can be weighted according to the client weight for that model.
• This increases the contribution of a private data set to the model parameters that are more relevant to that data set.

Abstract

Model training systems collaborate on model training without revealing respective private data sets. Each private data set learns a set of client weights for a set of computer models that are also learned during training. Inference for a particular private data set is determined as a mixture of the computer model parameters according to the client weights. During training, at each iteration, the client weights are updated in one step based on how well sampled models represent the private data set. In another step, gradients are determined for each sampled model and may be weighed according to the client weight for that model, relatively increasing the gradient contribution of a private data set for model parameters that correspond more highly to that private data set.

System and Method for Persisting Data Generated in Executing A Process Workflow (18447908)

Main Inventor

Joseph Vincent SCARFUTTI

Brief explanation

The patent application describes a system and method for storing data generated during the execution of a process workflow. 

• The method involves receiving messages exchanged during the process workflow through a communications module and a message broker.
• A writer service is used to break down each received message into multiple properties based on a database schema.
• The writer service then stores the message in a database according to the database schema using the communications module.
• A separate reader service is used to access the database and retrieve the properties of a specific message in response to a read request received through the communications module.

Abstract

A system and method are provided for persisting data generated in executing a process workflow. The method is executed by a device having a communications module and includes receiving via the communications module messages exchanged in executing the process workflow by a message broker. The method also includes using a writer service to disassemble each received message into multiple properties according to a database schema and persist the received message in a database according to the database schema via the communications module. The method also includes using a reader service to access the database and assemble the multiple properties of a first persisted message, in response to a read request received via the communications module, wherein the reader service is separate from the writer service.

METHODS FOR PROVIDING CONTEXTUAL NOTIFICATIONS IN A WEB BROWSER SESSION (17825641)

Main Inventor

Rachel CURWEN

Brief explanation

The patent application describes a computer-implemented method for providing personalized notifications to users while browsing web pages associated with specific merchants. Here are the key points:

• The method involves receiving a request to retrieve a web page from a web server through a browser application on a computing device.
• If the web page is associated with a specific merchant, the method proceeds to obtain historical transactions data of a resource account associated with the computing device.
• Based on the historical transactions data and a predetermined allocation of resources, the method determines the availability of resources for the user.
• When the user interacts with the web page, the method identifies products associated with the merchant based on the user's actions.
• Using the historical transactions data and resource allocation, the method generates personalized notifications related to the identified products.
• Finally, the generated notifications are provided to the computing device for display in the browser application, allowing the user to view and potentially act upon them.

Abstract

A computer-implemented method is disclosed. The method includes: receiving, via a browser application on a computing device, a request to retrieve a first web page from a web server; determining that the first web page is associated with a first merchant; in response to determining that the first web page is associated with the first merchant: obtaining historical transactions data of a resource account associated with the computing device; determining a first allocation of resources associated with the resource account; detecting a user action in the browser application for interacting with the web page; in response to detecting the user action in the browser application: identifying at least one product associated with the merchant based on the detected user action; generating a notification associated with the identified at least one product based on the historical transactions data of the resource account and the first allocation of resources; and providing, to the computing device for display in the browser application, the generated notification.

DISTRIBUTED AUTHENTICATION IN AMBIENT COMMERCE (17824165)

Main Inventor

Milos DUNJIC

Brief explanation

The abstract describes an ambient commerce system that includes a sensor, communication module, processor, and memory.

• The system detects an unauthenticated entity at an ambient commerce premises using sensors.
• It receives an indication of authentication from a trusted system for the unauthenticated entity.
• The system determines if the entity is authenticated based on the indication and additional authentication parameters.
• If the entity is authenticated, the system performs an ambient commerce operation for the entity.

Abstract

According to an aspect there is provided an ambient commerce system. The ambient commerce system may include a sensor at an ambient commerce premises, a communication module, and a processor coupled to the sensor and the communication module. The ambient commerce system further includes a memory coupled to the processor. The memory stores processor-executable instructions which, when executed, cause the processor to: detect, based on an output of one or more of the sensors, an unauthenticated entity at an ambient commerce premises; receive, from a first independent trusted system and via the communication module, an indication that authentication has been performed by the first independent trusted system for the unauthenticated entity using a first authentication parameter; determine that the entity is an authenticated entity based on the indication that the authentication has been performed and at least a second authentication parameter; and perform an ambient commerce operation for the authenticated entity.

SYSTEM AND METHOD FOR PROVIDING A THREE-DIMENSIONAL OBJECT REPRESENTING A STORED-VALUE CARD FOR DISPLAY IN AUGMENTED REALITY (18449313)

Main Inventor

Adrian Chung-Hey MA

Brief explanation

The abstract describes a computer server system that can send a stored-value card to a recipient's mobile device for display in augmented reality.

• The system includes a communications module, a processor, and a memory.
• The processor is configured to receive a request from a device to send a stored-value card to a recipient.
• The request includes parameters of the stored-value card.
• The processor generates the stored-value card and a three-dimensional object representing it based on the parameters.
• The system then sends a signal to the recipient's mobile device, including the three-dimensional object for display in augmented reality.

Abstract

A computer server system comprises a communications module; a processor coupled with the communications module; and a memory coupled to the processor and storing processor-executable instructions which, when executed by the processor, configure the processor to receive, via the communications module and from a requesting device, a signal that includes a request to send a stored-value card to a recipient, the request identifying one or more parameters of the stored-value card; generate the stored-value card and a three-dimensional object representing the stored-value card according to the one or more parameters; and send, via the communications module and to a mobile device of the recipient, a signal that includes the three-dimensional object representing the stored-value card for display in augmented reality.

MODELING DISJOINT MANIFOLDS (18202455)

Main Inventor

Jesse Cole Cresswell

Brief explanation

- The patent application describes a computer model that can account for data samples in a high-dimensional space as lying on different manifolds.

- Instead of representing the entire data set as a single manifold, the model treats it as a union of manifolds. - The model groups data samples that are expected to belong to the same underlying manifold. - For generative models, a sub-model is trained for each group using their respective data samples. - Each sub-model can account for the manifold of its corresponding group. - The overall generative model includes information on how frequently to sample from each sub-model to accurately represent the entire data set. - The grouping of data samples can also be used to improve classification accuracy in multi-class classification models. - Group data samples are weighed based on the estimated latent dimensionality of the group.

Abstract

A computer model is trained to account for data samples in a high-dimensional space as lying on different manifolds, rather than a single manifold to represent the data set, accounting for the data set as a whole as a union of manifolds. Different data samples that may be expected to belong to the same underlying manifold are determined by grouping the data. For generative models, a generative model may be trained that includes a sub-model for each group trained on that group's data samples, such that each sub-model can account for the manifold of that group. The overall generative model includes information describing the frequency to sample from each sub-model to correctly represent the data set as a whole in sampling. Multi-class classification models may also use the grouping to improve classification accuracy by weighing group data samples according to the estimated latent dimensionality of the group.