Asynchronous Coagent Recommendation at the Edge for Faster Responses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional recommendation systems in edge computing environments suffer from inefficiencies, inflexibilities, and inaccuracies due to synchronous algorithms that require geographically distant edges to synchronize, leading to slow communication and resource consumption, and often fail to accurately represent client device preferences.
Innovation Solution
An edge-based recommender system utilizing an asynchronous coagent network with local coagents that learn and act cooperatively, generating recommendations using local policy gradients and refining them as delayed scores are received, allowing for efficient, flexible, and accurate digital item recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If synchronous algorithms are used to generate recommendations across distributed edges, then coordination and consistency between edges is improved, but communication time and computing resource consumption increase significantly
Solution Approach 1:
The patent divides the recommendation generation process into independent segments executed by local coagents at different edges. Each coagent independently generates recommendations using local data and models, eliminating the need for synchronous coordination. This segmentation allows parallel execution across geographically distributed edges, reducing communication time while maintaining recommendation quality through local autonomy.
2Measurement precision
If edges must wait for signals from geographically distant edges before providing recommendations, then comprehensive data consideration is improved, but response speed deteriorates
Solution Approach 1:
The patent implements preliminary action by pre-training coagent models and pre-positioning recommendation capabilities at each edge before client requests arrive. Local coagents are prepared with local data and models in advance, enabling immediate recommendation generation upon receiving a client request without waiting for distant edges. This preliminary preparation maintains response speed while local models are trained comprehensively using available data.
3Stability of the object's composition
If rigid synchronization between edges is enforced, then system coordination is improved, but flexibility and robustness to network issues deteriorate
Solution Approach 1:
The patent implements dynamics by allowing the recommendation system to adapt its behavior based on local conditions and network status. Each coagent dynamically generates recommendations based on local data availability and client context without rigid synchronization constraints. The system can flexibly handle network issues, edge failures, or varying data availability at different edges, maintaining operational stability through decentralized autonomous operation rather than enforced synchronization.
4Productivity
If a bag-of-algorithms approach is used to combine distributed learning algorithms, then system scalability is improved, but recommendation accuracy deteriorates due to lack of theoretical grounding
Solution Approach 1:
The patent applies parameter changes by transforming the approach from combining heterogeneous algorithms to using consistent coagent models with learnable parameters at each edge. Each coagent uses the same underlying model architecture and learning framework, with parameters adapted locally to regional characteristics. This standardized parameter-based approach maintains theoretical grounding and recommendation accuracy while preserving scalability, as new edges simply instantiate additional coagents with the same proven framework.
Data Source
AI summary
The present disclosure relates to systems, methods, and non-transitory computer-readable media that generate digital item recommendations for client devices utilizing coagent recommendation models of a distributed asynchronous coagent network. Indeed, in one or more embodiments, the disclosed systems operate on an edge computing device of a distributed asynchronous coagent network. In some cases, the disclosed systems utilize recommendation scores generated at the edge computing device via local coagents and additional recommendation scores received from other coagents of other edge computing devices to generate a digital item recommendation. In some cases, the disclosed systems progressively refines the recommendation as delayed scores from the other coagents are received. Further, in some embodiments, the disclosed systems update parameters of the local coagents using local policy gradients determined from responses to the generated recommendations.