# US 2026/0073222 A1 · Ai-driven adaptive meta-learning framework for self-evolving neural architecture optimization

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

- Number: US 2026/0073222 A1
- Document ID: US20260073222A1
- Status: Published application
- Kind: A1
- Application number: 19/389,108
- Filed: 2025-11-14
- Published: 2026-03-12
- Earliest claimed: 2025-11-14
- Inventors: Ramesh Inala; Ravi Shankar Garapati; Avinash Reddy Aitha; Venkata Bhardwaj Komaragiri; Vijaya Rama Raju Gottimukkala; Mahesh Recharla; Pallav kumar Kaulwar; Sneha Singireddy; Sateesh Kumar Rongali; Keerthi Amistapuram; Dwaraka Nath Kummari; Goutham Kumar Sheelam; Durga Bramarambika Sailaja Varri
- Web page: https://patentgrep.com/patent/US20260073222A1

## CPC

- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/006: based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/045: Combinations of networks
- G06N 3/086: using evolutionary algorithms, e.g. genetic algorithms or genetic programming
- G06N 3/092: Reinforcement learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The invention provides an AI-driven adaptive meta-learning framework for self-evolving neural architecture optimization. The system employs a meta-controller network trained via reinforcement learning to autonomously generate and refine neural architectures. An adaptive reward engine and a self-evolution module enable the system to evolve its optimization policies dynamically across multiple tasks and environments. The invention reduces human dependency in model design, enhances generalization, and facilitates continuous AI evolution through reinforced meta-learning principles.

## Claims

### Claim 1

An adaptive self-evolving neural architecture optimization system comprising:

### Claim 2 (depends on 1)

The system as claimed in claim 1, wherein the meta-controller network employs a reinforcement learning policy gradient to optimize architecture generation strategy.

### Claim 3 (depends on 1)

The system as claimed in claim 1, wherein the adaptive reward engine utilizes Pareto-based multi-objective optimization balancing accuracy and efficiency.

### Claim 4 (depends on 1)

The system as claimed in claim 1, wherein the self-evolution module performs neural crossover and mutation operations on meta-controller weights.

### Claim 5 (depends on 1)

The system as claimed in claim 1, wherein the framework supports continual learning by reusing meta-knowledge from prior tasks.

### Claim 6 (depends on 1)

The system as claimed in claim 1, wherein the reinforcement signal comprises both short-term and long-term returns.

### Claim 7 (depends on 1)

The system as claimed in claim 1, wherein the invention is implemented through a distributed computing environment or edge-AI platform.

### Claim 8 (depends on 1)

The system as claimed in claim 1, wherein the framework autonomously modifies architecture search policies based on environmental feedback.

### Claim 9

A method for adaptive self-evolving neural architecture optimization comprising:

### Claim 10 (depends on 9)

The method as claimed in claim 9, wherein the architecture search and meta-learning occur concurrently to enable real-time adaptation.
