StackingNet: Collective Inference Across Independent AI Foundation Models

2026-02-14

AI Summary

This work addresses the limitations of current foundation models, which operate in isolation and lack effective coordination mechanisms to harness their complementary strengths for building trustworthy intelligent systems. To overcome this, we propose StackingNet—a meta-ensemble framework that operates without access to internal model parameters or training data. By orchestrating predictions from multiple heterogeneous black-box foundation models during inference, StackingNet transforms model diversity into collaborative advantage through principles of collective intelligence. The approach is broadly applicable across multimodal tasks, including language understanding, visual estimation, and academic scoring, consistently outperforming individual models and conventional ensemble methods. Empirical results demonstrate substantial improvements in accuracy, robustness, fairness, and the ability to rank and prune models based on reliability.

Technology Category

• Machine Learning: Ensemble Methods
• Multiagent Systems: Other Foundations of Multi Agent Systems
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Web search models and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Artificial intelligence built on large foundation models has transformed language understanding, vision and reasoning, yet these systems remain isolated and cannot readily share their capabilities. Integrating the complementary strengths of such independent foundation models is essential for building trustworthy intelligent systems. Despite rapid progress in individual model design, there is no established approach for coordinating such black-box heterogeneous models. Here we show that coordination can be achieved through a meta-ensemble framework termed StackingNet, which draws on principles of collective intelligence to combine model predictions during inference. StackingNet improves accuracy, reduces bias, enables reliability ranking, and identifies or prunes models that degrade performance, all operating without access to internal parameters or training data. Across tasks involving language comprehension, visual estimation, and academic paper rating, StackingNet consistently improves accuracy, robustness, and fairness, compared with individual models and classic ensembles. By turning diversity from a source of inconsistency into collaboration, StackingNet establishes a practical foundation for coordinated artificial intelligence, suggesting that progress may emerge from not only larger single models but also principled cooperation among many specialized ones.

Problem

• foundation models
• model coordination
• collective intelligence
• heterogeneous AI systems
• black-box integration

Innovation

• StackingNet
• collective intelligence
• meta-ensemble
• black-box model coordination
• foundation model integration