USPTO Patent Applications - AI & Computing (G06N)

SYSTEM, EDGE DEVICE, AND CONTROL METHOD THEREOF

USPTO published patent application US20260111770A1, filed October 22, 2025 by inventor KEI KITAGATA, covering systems and methods for distributed edge device reasoning. The technology enables edge devices to determine whether a local reasoning unit, a second device, or a cloud-based application service should handle a given reasoning task, routing query data directly between devices when appropriate. Technology companies developing edge computing, distributed AI inference, or IoT systems should monitor this application to assess potential licensing implications or freedom-to-operate considerations.

Methods For Using Quantum Computers - Quantinuum Ltd

Quantinuum Ltd filed US patent application US20260111775A1, published April 23, 2026, covering methods for enhancing quantum state preparation circuits to encode multivariate probability distributions for quantum Monte Carlo integration. The application, filed July 26, 2023 (application no. 18998150), names Steven Herbert and Michael Spranger as inventors and carries CPC classifications G06N 10/20, G06N 10/40, and G06N 10/60. The invention adds dimensions for sums, products, maxima/minima and includes a flag qubit for conditional control in QMCI computations.

Layout Optimization of Error Correction in Measurement-Based Quantum Computing Patent

USPTO published patent application US20260111781A1, filed October 23, 2024 (Application No. 18924047), titled 'Layout Optimization of Error Correction in Measurement-Based Quantum Computing.' The application covers a method, apparatus, and product for converting a logical representation of a quantum program into a physical representation executable on a measurement-based quantum computer, involving segmentation of quantum sources into disjoint segments generating consecutive layers of qubit clusters separated by time delays, and selection of non-uniform interleaving depths across segments. Named inventors are Amir Naveh, Shmuel Ur, Ron Cohen, Eyal Cornfeld, Yehuda Naveh, and Nati Erez. The patent is classified under CPC G06N 10/40 and G06N 10/20.

Qubit State Monitoring Patent for Quantum Computers Using rf-SQUIDs

USPTO published patent application US20260111778A1 titled 'Qubit State Monitoring in Quantum Computers Utilizing Radio Frequency Superconducting Quantum Interference Devices with Classical Computer Readout and Processing,' filed October 17, 2024. The application covers real-time monitoring of qubit states in quantum computers using a system based on rf-SQUIDs and direct current signal detection for superconducting quantum circuits. Inventors are Tejinder Singh and Navjot Kaur Khaira.

Quantum Serverless Suitability Scoring Method

USPTO published patent application US20260111774A1 on April 23, 2026, filed October 18, 2024, for a method of dynamically evaluating serverless functions for quantum computing environments. The invention by inventors Stephen Coady and Leigh Griffin computes a suitability score for proposed quantum serverless functions based on multiple scoring criteria specific to a particular quantum computing environment, then performs an execution action based on that score. CPC classification G06N 10/20 places this in quantum computing models.

Action Synchronization via Entangled Particle Pairs, Apr 23

USPTO published patent application US20260111772A1 titled 'Action Synchronization via Entangled Particle Pairs' filed October 17, 2024 by inventors John Stern Alexander, Steven Lee Fisher-Stawinski, and Charles Palmer. The application covers a quantum computing method for synchronizing actions between recipients at different locations using prepared streams of entangled particle pairs, where measurement results from entangled particles are matched against a predetermined playbook to trigger paired actions. The application is classified under CPC G06N 10/00 (quantum computing).

Fusion-Based Qubit Syndrome Values for Quantum Computing

Psiquantum Corp. has published US patent application US20260111777A1 for a quantum computing method involving fusion-based qubit syndrome values, filed February 6, 2024 and published April 23, 2026. The invention covers receiving multiple independent quantum systems each containing entangled subsystems, performing joint measurements across subsystems from different systems, and determining syndrome graph values from the measurement outcomes. The application names Mercedes Gimeno-Segovia, Terence Rudolph, and Naomi Nickerson as inventors.

Virtual Quantum Gates Obscure Proprietary Circuit Information

USPTO published patent application US20260111773A1 on April 23, 2026, filed October 17, 2024 under Application No. 18919322. Inventors Derek Wang, Ali Javadiabhari, and Alireza Seif Tabrizi disclose methods and systems for obscuring proprietary information encoded in quantum circuits using virtual quantum gates that are not executed on quantum hardware, protecting against power side-channel attacks. The invention transforms target quantum circuits into power-attack resistant forms by encoding proprietary information—including circuit structures, measurement outcomes, initial product states, and parameters—in virtual gates tracked classically and requiring no power.

AI Uncertainty Estimation for Autonomous Object Detection

USPTO published patent application US20260111771A1 on April 23, 2026, covering systems and methods for uncertainty estimation in object detection for autonomous and semi-autonomous systems. The application, filed March 14, 2025, names six inventors including Nikita Durasov, Jiwoong Choi, Rafid Reza Mahmood, Marc Law, James Robert Lucas, and Jose Manuel Alvarez Lopez. The technology uses data from cameras and/or LiDAR sensors to generate object presence probabilities along with corresponding uncertainty estimates, enabling identification of novel scenes, detection of bounding box errors, and flagging of missed object detections.

Cryogenic Magnetless Circulator for Hybrid Classical-Qubit Readout

USPTO published patent application US20260111779A1 for a cryogenic magnet-less circulator device designed for integration with quantum computers to facilitate qubit measurements. The invention uses spatiotemporal modulation of angular momentum for microwave photons controlled via a modulated control signal. The circulator topology employs tank circuits arranged in a star topology with different sets of radio frequency superconducting quantum interference devices (rf-SQUIDs) per tank circuit, which resonate out-of-phase based on the modulated control signal. The application was filed on October 17, 2024, with inventors Navjot Kaur Khaira and Tejinder Singh.

Quantum Gate Device with Superconducting Qubit Technology

USPTO published patent application US20260111776A1, filed September 13, 2023 as Application No. 19111598, covering a quantum gate device using superconducting qubit technology. The invention includes at least two data qubits and a coupler qubit, with a pulse irradiator that applies cross-Rabi transition and residual ZZ interaction elimination drive pulses to reduce quantum decoherence errors. Inventors are Atsushi Noguchi, Shotaro Shirai, Yuta Okubo, and Yasunobu Nakamura.

rf-SQUID Technology for Quantum Computer Qubit State Monitoring

USPTO published patent application US20260111780A1 on April 23, 2026, filed October 17, 2024 with application number 18919218. The application, invented by Tejinder Singh and Navjotjot Kaur Khaira, covers distributed radio-frequency superconducting quantum interference devices (rf-SQUIDs) for real-time monitoring of qubit states in quantum computers. The system uses a superconducting control wire carrying RF signals to detect phase shifts without adding load or noise to superconducting quantum circuit wires, enhancing detection of qubit entanglement, decoherence, and state changes.

AI Expert System Using LLM Personas

The USPTO has published patent application US20260099718A1, filed October 3, 2024 (Application No. 18905232), covering a system and method for providing task-specific expert insights using large language model (LLM) personas. The method involves preprocessing relevant datasets, receiving user input data, prompting an LLM component with an expert persona and expert persona instruction, and presenting the resulting LLM output to the user. The application is classified under CPC G06N 3/091 (AI computing) and names five inventors.

AI HTML Denoising System and Method for Website Application Interactions

USPTO published patent application US20260099717A1 by inventors David Colwell, Marius Viktor, Mark Bugno, Venkata Bommireddipalli, and Michael Keeley. The application covers systems and methods for denoising HTML components of website applications using a trained generative AI service comprising at least one multimodal machine learning model and one large language model, to identify website elements and execute tasks automatically. Filing date was October 4, 2024, with application number 18907025.

Patent - Detecting Structure Data from Aerial Imagery

The USPTO has published patent application US20260099884A1, filed by Insurance Services Office, Inc. on December 2, 2025, covering systems and methods for detecting, extracting, and categorizing structure data from aerial imagery following major weather events. The invention uses machine learning to process digital images and weather mapping data, cross-reference property attributes, and assign risk ratings to structures within a user-defined region of interest. The system generates and delivers data packages to users for risk assessment purposes.

Transaction Data Processing Systems And Methods

The USPTO published patent application US20260099885A1 on April 9, 2026, filed December 10, 2025, by nine inventors led by Delia Rusu. The application discloses an AI-based transaction data processing system using numerical representation generation models trained on historical transaction records and transaction attribute prediction models trained on previously reconciled financial records.

Temporally Amortized Supersampling in Graphics Processing

Intel Corporation filed patent application US20260099895A1 with the USPTO for temporally amortized supersampling in graphics processing. The invention covers a GPU-implemented neural network for spatiotemporal upscaling filter kernel prediction, including an input processing stage, neural network stage, and output filtering stage. Inventors are Dmitry Kozlov, Andreas Weinmann, and Mikhail Petrushkov, with filing date October 9, 2024.

Autonomous Spreadsheet Creation from Natural Language Queries

The USPTO published patent application US20260099668A1 titled 'Autonomous Spreadsheet Creation.' The application covers a method for receiving a natural language query requesting a document with a tabular structure, identifying attribute categories, converting the query into a data access query for external data sources, retrieving data items corresponding to those categories, and populating spreadsheet cells with the retrieved data items. The application was filed on December 11, 2025, and names Andrew J. Lavery, Earl J. Wagner, and Matthew Albright as inventors.

Virtual Figure Generation from Pressure Sensor Weight Device

The USPTO published patent application US20260100011A1, filed August 22, 2025, for a system and method combining a weight support device with a pressure sensor grid and a connected computer to capture pressure data, generate a virtual figure representing the user, and display animated movement on a screen. The kinematic model adjusts body-segment positions in the virtual figure based on detected movement patterns.

Generative AI Claim Processing System, LLM Integration, Patent Application

The USPTO published patent application US20260099883A1, filed April 9, 2025, covering a generative AI-based system for claims data processing and evaluation. The system generates prompt data from parsed standard operating procedures and rules, uses a large language model to produce first and second rule sets, extracts non-adjudicated claims data with corresponding edit codes, and validates claims against the second rule set to generate resolution recommendations. The invention appears directed toward automating insurance or healthcare claims adjudication workflows.

Superconducting Qubit Reset with Microwave Photon Counter

USPTO published patent application US20260099741A1 by inventors Robert Francis McDermott, III and Alexander M. Opremcak disclosing systems and methods for deterministic reset of superconducting qubit and cavity modes with a microwave photon counter. The invention comprises multiple qubit-microwave photon counter pairs coupled by qubit-qubit coupling, each containing a qubit circuit, microwave photon counter circuit, and resonant cavity. The filing date is October 22, 2024.

Quantum Computing Tunable Coupler with Non-Galvanic Coupling

USPTO published patent application US20260099740A1 for a tunable coupler with non-galvanic coupling interface for quantum computing applications. The invention by inventors Heinsoo and Ockeloen-Korppi includes a first constant coupling element forming a non-galvanic coupling interface to at least a first qubit. The application was filed on September 26, 2024 under application number 18898030.

Machine Learning Model for Flow Cytometry Observation Space Classification

The USPTO published patent application US20260099739A1 filed by Yoshimi Noguchi and Naoko Senda. The application covers a computer system that stores measurement data from flow cytometry and a machine learning model configured to receive feature values of regions in an observation space and output class probabilities. The system maps measurement data into an observation space, divides it into regions based on distribution features, calculates feature values for each region, and sets gates based on class probability outputs from the model.

AI Education Platform with Graph-Based Q&A System

USPTO published patent application US20260100142A1 filed by Nicholas X. Wang on October 4, 2024. The application covers an AI-driven education platform using graph-based question-answering with an Intuitive Level Metric (ILM) optimization system. The invention generates causal graphs from multiple-choice questions and uses machine learning to produce answers alongside step-by-step visual instructions. CPC classifications include G09B 7/06 (electrically-operated education appliances) and G06N 20/00 (machine learning).

Image Generation Using One or More Neural Networks

USPTO has published patent application US20260099963A1 by inventor Ming-Yu Liu for neural network-based image generation techniques. The application covers training neural networks to generate and modify digital images with user-selectable features, filed December 11, 2025 and published April 9, 2026.

4D Generative Models From in the Wild Videos

USPTO published patent application US20260099987A1 disclosing a method for generating 4D generative models from 2D video frames using 3D Gaussians and neural networks. The system processes sequential image frames to construct deformed 3D Gaussians with motion features, enabling creation of dynamic 4D scene representations from real-world video inputs. The application, filed October 7, 2025 under application number 19352319, names Koki Nagano, Kaiwen Jiang, Shalini De Mello, and Xueting Li as inventors.

AI Text Analysis System to Accelerate Academic Research

USPTO published patent application US20260099682A1, filed November 20, 2024, for an AI-based text analysis system using a BERT model fine-tuned for academic and industry literature. The system includes multilingual query support, named entity recognition for datasets and methodologies, and integration with a Student Information System providing insights into student extracurricular decision-making activities. The application is classified under CPC codes G06F 40/40 and G06N 3/0455/3/096 covering AI and computing technologies.

Automated Pipeline for Training Language Models

USPTO has published patent application US20260099674A1 assigned to U.S. Bank National Association, describing a method and system for generating training datasets for language models using an automated pipeline. The system receives input samples, performs rephrasing operations via a generative LLM to produce semantically equivalent versions with different phrasing, labels entity references, and aggregates them into an expanded labeled dataset for natural language processing model training.

Memory-Efficient Draft Machine Learning Model

USPTO published patent application US20260099673A1 titled 'Memory-Efficient Draft Machine Learning Model' on April 9, 2026. The application covers systems and methods for model training using linear layers, decoding layers, down-projection layers, and token predictors to generate ML output tokens with reduced memory requirements. The invention addresses computational efficiency in neural network architectures by processing embeddings through dimension-reducing projection operations.

Methods and Arrangements for Coupling Quantum Mechanical System to Quantum Mechanical Environment

The USPTO published patent application US20260099751A1 titled 'Methods and Arrangements for Coupling a Quantum Mechanical System to a Quantum Mechanical Environment.' The application covers a quantum information processing system comprising a quantum mechanical system capable of exhibiting transitions between eigenstates, coupled to a quantum control environment with modes from a continuous spectrum. Inventors Akseli Mäkinen and Mikko Möttönet filed the application on August 31, 2023.

Quantum Computing Scalability Evaluation Patent Application

USPTO published patent application US20260099742A1 by inventor Paul Victor Klimov, disclosing systems and methods for evaluating quantum computing scalability. The application covers obtaining benchmark performance data for quantum system architectures, deriving scaling parameters through quantum scaling models, and determining scaling metrics at processor sizes beyond tested ranges to inform control actions for operational quantum systems.

Quantum Circuit Data Encoding Using Complex Numbers

The USPTO published patent application US20260099744A1 titled 'Complex Data Encoding for Quantum Circuits' on April 9, 2026. The application covers methods for encoding data into quantum circuits by chunking, tokenizing, and representing coefficients as complex numbers, allowing multiple values and positional information to be encoded in a single encoding input. Inventors are Brendan Burns Healy and Yasmin Mansy, with CPC classification G06N 10/20.

Quantum Fourier Transform Partitioning in Data Center

The USPTO published patent application US20260099748A1 disclosing a method for distributed quantum Fourier transform (QFT) with reduced EPR pairs across multiple quantum processors in a datacenter. The invention by inventors Eneet Kaur and Reza Nejabati enables QFT computation to be partitioned across two or more quantum processors while using fewer entanglement resources and properly positioning qubits at completion without requiring swapping operations.

Quantum Entanglement Management for Quantum Informatics System

USPTO published patent application US20260099749A1 for a quantum entanglement management system for quantum informatics processing. The system includes an entanglement scheduler that identifies operations requiring entanglement of quantum systems, evaluates potential entanglement attempts, and schedules attempts based on probability of success. The application was filed September 19, 2023, by inventors Stephanie Simmons and Camille Chartrand under CPC classifications G06N 10/40 and G06N 10/20.

Quantum Computing Arrangement and Quantum Computer with Permanent Magnet Design

USPTO published patent application US20260099750A1 for a quantum computing arrangement featuring a permanent magnet system that establishes position-dependent magnetic fields along an axis. The invention includes space for trapping at least two quantum particles and soft magnetic material configured to enhance the magnetic field. The patent names seven inventors and was filed September 26, 2023.

Quantum Processing Units Rack Interconnect via Near Infrared Optical Links

The USPTO published patent application US20260099755A1 on April 9, 2026, covering architectures for quantum data centers. The invention describes quantum processing units arranged within a rack interconnected via near infrared optical links through a top-of-rack switch, with connection to external quantum network switches via telecommunication wavelength links.

Fine-Tuning a Target Generative Neural Network Using an Improvement Generative Neural Network

Six inventors (Wang, Fan, Zhao, Ramachandran, Yang, Jain) filed USPTO patent application US20260099906A1 for methods to train generative neural networks. The technique uses an improvement neural network to process outputs from a target network and generate preferred training examples iteratively, reducing reliance on static datasets and external human annotation. The application was published April 9, 2026, following an October 3, 2025 filing date.

Variable Compression, De-Resolution, and Restoration of Medical Images Using GAN-Based AI

The USPTO published patent application US20260099920A1 filed by inventor Sean M. Kelly on September 25, 2023. The application covers an AI image processing platform that uses GAN models to compress, down-resolve, up-resolve, and restore medical images based on diagnostic and therapeutic relevance. The platform classifies image regions by cellular morphologies and diagnostic relevance and conveys processed images to client devices via communication networks. CPC classifications include G06T 7/0012, G06N 3/0475, and multiple G16H healthcare informatics categories.

Machine Learning Device, Estimation System, Training Method, and Recording Medium

USPTO published patent application US20260099723A1 for a machine learning device that trains encoding models for sensor data using adversarial estimation techniques. The application, filed by NEC Corporation (Application No. 19414668) on December 10, 2025, covers devices and methods for training first and second encoding models with an adversarial estimation model to improve estimation accuracy.

Synthetic Corruption of Machine Learning Output

USPTO published patent application US20260099721A1 for a system and method enabling a corrupter module to synthesize qualified corrupt data for training machine learning safeguard models. The system receives first output data from a large language model, identifies entity-to-concept mappings in a domain ontology, and generates training data by replacing entities with alternative entities mapped to different ontological concepts that comply with predefined corruption rules. The trained safeguard model is configured to detect errors in subsequent language model outputs.

NEC Multi-Modal Machine Learning Encoding Device Patent Application

The USPTO published patent application US20260099722A1 filed by NEC Corporation on December 10, 2025, and made available on April 9, 2026. The application covers machine learning devices and methods that train encoding models for multiple sensor data types and use adversarial training to improve estimation accuracy. CPC classifications G06N 3/094 and G06N 3/045 indicate AI and neural network innovations. This is a publication of a patent application, not a granted patent.

Oracle Patent Application US20260099533A1 Topic Maps for Raw Content

USPTO published patent application US20260099533A1 assigned to Oracle International Corporation on April 9, 2026. The application covers techniques for receiving, storing, and analyzing raw text content using topic maps, including normalizing input text, identifying topics via tags, and generating topic maps referencing content items. Filing date was November 24, 2025; application number 19398208.

Visa Entity Linking Using Subgraph Matching

The USPTO published patent application US20260099545A1 filed by Visa International Service Association for a system enabling entity linking through graph neural networks. The invention extracts attributes from unknown entities, generates graph representations, and matches them against known entity graphs using machine learning. The application, filed September 25, 2023 under application number 19115996, was published on April 9, 2026 with inventors Sheng Wang, Xutong Wang, Jie Yuan, Peng Wu, and Dan Wang.

Dimensionality Reduction of Neural Networks Intermediate Feature Maps Using Two-Dimensional Principal Component Analysis

USPTO published patent application US20260099700A1 on April 9, 2026, titled 'Dimensionality Reduction of Neural Networks Intermediate Feature Maps Using Two-Dimensional Principal Component Analysis.' The application discloses a method for reconstructing input matrices by decoding mean, principal components, and row projection matrices from a bitstream and using two-dimensional PCA for neural network feature map dimensionality reduction. Inventors include Afrabandpey, Aminlou, Rezazadegan Tavakoli, Zhang, and Hannuksela.

Deep Neural Network Multi-Terminal Logic Gates Patent

USPTO published patent application US20260099702A1 for a deep neural network circuit with multiple layers formed of multi-terminal logic gates. Inventor Fabio Lorenzo Traversa filed the application on December 11, 2025, under application number 19416277. The circuit includes logic gates arranged into layers with logical connectors between adjacent layers, where each connector has one of multiple states and can be trained to implement functions.

Private Function Evaluation Using Machine Learning

USPTO published patent application US20260099704A1 for private function evaluation using machine learning. The invention covers systems and methods for training neural networks to approximate functions, generating shallow neural networks, and using secret sharing to enable private function evaluation without exposing inputs or function parameters.

Big Data Intelligent Selection Design Method for Rockburst-Prevention Hydraulic Supports in Rockburst Roadways

The USPTO published patent application US20260099708A1 for a big data intelligent selection design method for rockburst-prevention hydraulic supports used in rockburst roadways. The invention uses machine learning and statistical analysis to identify key factors affecting hydraulic support performance based on geological conditions, rock mechanical properties, mine face layout, and historical usage and maintenance records. The system aims to improve selection accuracy and efficiency of rockburst-prevention hydraulic supports for mining safety applications.

Techniques for Improved User Experience Prediction

The USPTO published patent application US20260099694A1 titled 'Techniques for Improved User Experience Prediction' filed on October 8, 2024. The application discloses a computer-implemented method that applies machine learning models to sequences of web pages and associated metrics data to generate user experience values using neural network embeddings and hidden layers.

Electronic Apparatus for Providing Recommendation Information and Operating Method Thereof

The USPTO published patent application US20260099696A1 filed by Visa on April 9, 2026, covering an AI recommendation system. The application (originally filed October 30, 2024, Application No. 18931310) describes an electronic apparatus using a trained artificial intelligence model with self-attention layers and inductive bias processing to derive recommendation information from sequential data patterns. Inventors include Noseong Park, Yehjin Shin, Jeongwhan Choi, and Hyowon Wi.

Bridging LLMs of Differing Sizes to Reduce Latency

USPTO published patent application US20260099528A1 by inventor Brett Barros for methods of reducing LLM latency by using a smaller LLM to generate immediate responses while a larger LLM produces refined content starting from the smaller model's output. The larger model generates a refined portion succeeding the initial content, which can then be rendered to the user. Alternative implementations use default text strings or predefined templates selected via natural language understanding of the user query.