Publications
Work by Icosa and collaborators across industry and academia.
Icosa Computing's research sits at the intersection of physics-based optimization and language models. Our core method, Combinatorial Reasoning, was developed in collaboration with NASA and USRA and is supported by $1.4 million in funding from the U.S. National Science Foundation. The work below spans peer-reviewed papers and industry whitepapers with partners including Fujitsu and NEC, covering how combinatorial optimization can select stronger reasoning paths, how quantum-inspired solvers improve large language models, and how these techniques apply to real workloads in financial trading. Together they form the scientific foundation for Icosa's products: local, domain-expert AI models that reason at frontier level on the work they know.
Icosa Papers
Combinatorial Reasoning: Selecting Reasons in Generative AI Pipelines via Combinatorial Optimization
Collaboration with NASA, USRA, and HPE. Introduces Combinatorial Reasoning: sampling many candidate reasoning paths from an LLM, mapping selection to a QUBO problem, and using physics-inspired solvers to pick the strongest subset before answering.
Improving Large-Language Models with Quantum-Inspired Optimization
Collaboration with Fujitsu. Shows how quantum-inspired optimization hardware can improve LLM output quality, letting smaller models reach accuracy normally associated with much larger hosted systems.
Quantum-Inspired Advantage in Financial Trading
Collaboration with NEC. Applies quantum-inspired optimization to financial markets, demonstrating measurable advantage in portfolio-level trading decisions on real market data.