Neural-symbolic transaction execution snapshot and rollback method for long-range agent collaboration

By employing neural symbolic transaction execution and causal dependency rollback methods, the problem of unifying the alignment between neural reasoning state and symbolic fact state in multi-agent collaboration is solved, achieving accurate state recovery and system stability, and improving the robustness and availability of intelligent systems.

CN122240245APending Publication Date: 2026-06-19HANGZHOU TUBU ER TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HANGZHOU TUBU ER TECHNOLOGY CO LTD
Filing Date
2026-03-19
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing distributed transaction and workflow orchestration technologies struggle to achieve unified alignment and consistency verification between neural inference states and symbolic fact states in multi-agent collaboration. They lack a cross-agent version alignment link based on a global logical clock, leading to unstable commit and recovery boundaries. Furthermore, the lack of causal dependency registration and dirty data tracking and marking can easily result in uncontrolled downstream propagation and excessively large rollback ranges.

Method used

The method employs neural symbolic transaction execution and causal dependency rollback. By acquiring tensor snapshots and logical fact sets of each agent, a global logical clock is established to form a full state view. Speculative external side effects are executed within the shadow session to trigger write-time differential copying, perform consistency checks and causal dependency registration, and dynamically rollback to determine the minimum necessary steps, thus achieving precise recovery.

Benefits of technology

It enables refined management of reasoning, decision-making, and execution in long-range intelligent agent collaboration, reduces the risk of state inconsistency and implicit dependency propagation, improves system robustness and recoverability, avoids system oscillation and performance loss, and enhances the security and engineering availability of large-scale intelligent systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122240245A_ABST
    Figure CN122240245A_ABST
Patent Text Reader

Abstract

This invention discloses a method for snapshotting and rolling back neural symbolic transactions for long-range agent collaboration, comprising the following steps: acquiring the states of multiple agents and establishing a unified temporal view; enabling an isolated shadow transaction runtime environment on the unified view; performing differential recording on key changes during shadow runtime to form a snapshot chain; determining the rationality of state changes through neural symbolic consistency verification; if the verification passes, submitting the result and establishing a traceable collaborative state; if the verification fails, predicting the minimum rollback range and performing cascading rollback recovery. This invention employs neural symbolic transaction execution and causal dependency rollback methods to achieve traceable submission and minimal rollback in long-range agent collaboration, possessing advantages such as strong consistency, accurate recovery, and high system stability.
Need to check novelty before this filing date? Find Prior Art