A foundation model of numerical intelligence with cross-disciplinary generalization

2026-07-30

AI Summary

This work proposes UNICON (Unified In-Context Operator Networks), a novel framework designed to endow intelligent models with cross-disciplinary zero-shot generalization capabilities for addressing emerging challenges in scientific and societal systems. UNICON introduces the concept of “numerical intelligence,” leveraging graph-structured examples as contextual prompts to infer predictive relationships shared across samples and directly apply them to unseen disciplinary tasks without retraining. By integrating graph neural network architectures, diverse training corpora, and a collaborative reasoning mechanism involving language model agents, UNICON achieves performance on par with or surpassing that of domain experts across multiple interdisciplinary benchmarks. The results substantially outperform existing approaches and underscore the critical role of data diversity in enabling effective cross-domain numerical reasoning.

Abstract

Intelligence is commonly understood as the ability to acquire and apply knowledge, adapt to unfamiliar situations and solve new problems. Large language models exhibit this capacity by inferring task-relevant knowledge from textual context and applying it to new tasks. Yet intelligence need not be confined to language. For scientific and social systems, we need models that acquire and apply knowledge from numerical context-an ability we call numerical intelligence. Here we introduce UNified In-Context Operator Networks (UNICON), a foundation model that exhibits numerical intelligence across disciplines. Using graph-based examples from a system as context, UNICON infers the predictive relation shared across them and applies it to queries from the same system. Across scientific and social systems, including those from disciplines absent from training, the same model approaches specialist performance without retraining. Combining UNICON with language-model agents yields further gains, enabling it to surpass state-of-the-art specialists in a discipline unseen in training. We further show that training-corpus diversity improves generalization to unseen disciplines. Together, these results establish UNICON as a foundation model of numerical intelligence and position it as a building block for a broader ecosystem of artificial intelligence.

Problem

• numerical intelligence
• foundation model
• cross-disciplinary generalization
• in-context learning
• artificial intelligence

Innovation

• numerical intelligence
• foundation model
• cross-disciplinary generalization
• in-context learning
• graph-based reasoning