Search

US-20260127585-A1 - ARTIFICIAL INTELLIGENCE MODEL AND DATASET SECURITY FOR TRANSACTIONS

US20260127585A1US 20260127585 A1US20260127585 A1US 20260127585A1US-20260127585-A1

Abstract

AI data and datasets that are represented as NFTs and carry all applicable data for a dataset's provenance, authenticity, and ownership. NFTs are used to validate datasets useful in AI training and can also be used to identify datasets that include faulty, biased, or otherwise erroneous data to improve predictive usefulness and reliability in decision making from the AI models.

Inventors

  • Rakesh Ramde
  • Amanjyot Singh JOHAR

Assignees

  • Datacurve, Inc.

Dates

Publication Date
20260507
Application Date
20251110

Claims (11)

  1. 1 .- 3 . (canceled)
  2. 4 . A computer-implemented method in a system, on a data communication network, for providing immutable information for renting digital assets in 3-D environments, the method comprising: creating an entity private key/wallet address pair associated with a specific user for a unique digital asset on a specific blockchain; creating a smart contract for the specific blockchain and associated with the entity private key/wallet address pair, wherein the smart contract executes in an Ethereum Virtual Machine (EVM) to govern the unique digital asset of the specific user; providing, by the smart contract, for rental of the unique digital asset by providing a public key corresponding to the private key/wallet to a renter in exchange for a payment; and providing, by the smart contract, for import of the unique digital asset to a 3-D environment powered by a gaming engine.
  3. 5 . The method of claim 1 , wherein the unique digital asset comprises a copy of a digital asset.
  4. 6 . The method of claim 1 , wherein the 3-D environment comprises a metaverse environment.
  5. 7 . The method of claim 1 , wherein the smart contract generates an NFT to bind the unique digital asset.
  6. 8 . The method of claim 1 , wherein the specific blockchain comprises at least one of Ethereum, Polygon, Avalanche, Optimism, Solana, and Ripple.
  7. 9 . The method of claim 1 , wherein the payment is made via cryptocurrency.
  8. 10 . The method of claim 1 , wherein an AI agent reports on interactions with the unique digital asset on the specific blockchain.
  9. 11 . The method of claim 1 , further comprising registering biometric data for the specific user to use in accessing the private key/wallet address pair.
  10. 12 . A non-transitory computer-readable medium in a rental verification system, on a data communication network, storing code that when executed, performs a method for providing immutable information for renting digital assets in 3-D environments, the method comprising: creating an entity private key/wallet address pair associated with a specific user for a unique digital asset on a specific blockchain; creating a smart contract for the specific blockchain and associated with the entity private key/wallet address pair, wherein the smart contract executes in an Ethereum Virtual Machine (EVM) to govern the unique digital asset of the specific user; providing, by the smart contract, for rental of the unique digital asset by providing a public key corresponding to the private key/wallet to a renter in exchange for a payment; and providing, by the smart contract, for import of the unique digital asset to a 3-D environment powered by a gaming engine.
  11. 13 . A rental verification system, on a data communication network, for providing immutable information for renting digital assets in 3-D environments, the rental verification system comprising: a processor; a network interface communicatively coupled to the processor and to a data communication network; and a memory, communicatively coupled to the processor and storing: a first module create an entity private key/wallet address pair associated with a specific user for a unique digital asset on a specific blockchain; a second module to create a smart contract for the specific blockchain and associated with the entity private key/wallet address pair, wherein the smart contract executes in an Ethereum Virtual Machine (EVM) to govern the unique digital asset of the specific user; a third module to provide, by the smart contract, for rental of the unique digital asset by providing a public key corresponding to the private key/wallet to a renter in exchange for a payment; and a fourth module to provide, by the smart contract, for import of the unique digital asset to a 3-D environment powered by a gaming engine.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority as a continuation under 35 USC 120 (a), to US App. No. Ser. No. 18/669,109, filed May 20, 2024, by Ramde et al., ARTIFICIAL INTELLIGENCE MODEL AND DATASET SECURITY FOR TRANSACTIONS, which claims priority under 35 USC 119 (e) to 63/467,726, entitled BIOMETRIC AND ARTIFICIAL INTELLIGENCE DATASET SECURITY FOR TRANSACTIONS, and filed May 19, 2023, by Ramde et al., the contents of which are hereby incorporated in its entirety. FIELD OF THE INVENTION The invention relates generally to computer networks, and more specifically, to protect and secure artificial intelligence models and datasets used to train the models for inference via cryptographic techniques. BACKGROUND Artificial Intelligence models and datasets used to train them are known. It is also known that such AI models and datasets require careful curation, preprocessing, and validation to ensure their reliability, robustness, and business and ethical alignment with intended applications. With the advent of Large Language Models (LLMs), the field of natural language processing has witnessed a significant leap forward. These powerful models, trained on vast amounts of text data, possess remarkable capabilities in understanding and generating human-like text. LLMs leverage deep learning architectures, such as transformers, to capture intricate patterns and relationships within language, enabling them to perform tasks like text generation, summarization, translation, and question answering with unprecedented accuracy. However, the development of LLMs poses challenges related to computational resources, data privacy, and potential biases inherited from the training data. For the purposes of this invention, models refers to predictive AI models, or generative AI models such as LLMs. AI training datasets are collections of data that are used to train and test artificial intelligence models. These datasets typically consist of a large amount of structured or unstructured data, such as images, text, audio, or video. AI datasets are critical for the development of artificial intelligence models, as they provide the raw material needed to teach the models to recognize patterns, make predictions, and ultimately ensure that the models can make their own decisions without human interventions. The quality and size of the dataset can have a significant impact on the performance of the resulting AI model. There are many publicly available AI datasets, such as the MNIST or ImageNet datasets for image recognition or the COCO dataset for object detection. In addition, many organizations and companies create their own proprietary datasets for specific applications, such as healthcare or finance. It is important to note that AI datasets can also raise concerns, particularly around issues such as bias, privacy, objectionable content, sexist, discriminatory, hallucinations, accuracy and security. Therefore, it is important to approach the creation and use of AI datasets with care and to ensure that they are developed and used in a responsible manner. As such it is important to keep them secure to ensure their integrity and protect sensitive information they may contain. If the datasets comprise faulty, biased, or otherwise erroneous data, the predictions and the decision making from the AI models are bound to be biased, discriminatory, objectionable or outright erroneous. This can also be a major security issue if bad actors can deliberately alter datasets or the model training pipelines and processes for their own nefarious purposes. Different methods exist for adding security to datasets. Differential privacy is a technique that adds random noise to a dataset to protect the privacy of individuals in a dataset and is deployed, for example, with healthcare datasets. Federated learning and multiparty computation are techniques where multiple entities or devices collaborate to train an AI model without sharing the underlying data. Homomorphic encryption allows for the processing of encrypted data without the need to decrypt it first. There are various other similar techniques that while allowing for protecting or shielding some data, do not overall contribute meaningfully to the transparency, provenance, or utility of the overall data sets. In addition, generative models can be used to create synthetic training datasets that capture the statistical properties of real data without directly exposing sensitive information, reducing apparent privacy risks. However, when mixed with real data, it is almost impossible to know where the original training data came from and whether it may have been altered from its original state at all. None of the techniques described above are useful when it comes to fundamentally securing the provenance, lineage, and ownership of the dataset or the model itself. These are fundamental to the robustness of the data contained in the datasets. In an enterprise environment, there are