Verseon Develops Superior Approach in Combining AI Models for Greater Accuracy
Peer-reviewed paper describes novel ensembling technique that offers 150% improvement over current state of the art
FREMONT, Calif., Oct. 9, 2024 /PRNewswire/ — Verseon International Corporation is pleased to announce the peer-reviewed results for its fundamental innovation in combining machine-learning models. Combining models is known as “ensembling” in the AI field. Verseon has incorporated its novel class-weighted ensemble system (CWES) into its patent-pending VersAI™ technology.
In the face of complex datasets, relying on a single AI model poses significant disadvantages that greatly diminish AI’s predictive capabilities. Over the past 30 years, ensemble learning has emerged as a principal method to deal with datasets that are imbalanced, that contain a large number of noisy, scattered data points, or whose number of features is large relative to the number of observations within the data.
However, combining individual models into a single coherent overarching paradigm still remains a significant challenge in the AI field. Among the most prevalent approaches for ensembling are simple voting, weighted-majority voting (WMVE), and the current state of the art known as class-specific soft voting (CSSV).
In the paper Ensemble Learning with Highly Variable Class-Based Performance, Verseon’s team describes their newly developed technique and contrasts it with simple voting, WMVE, and CSSV. Unlike other approaches, Verseon’s CWES decides how to combine a set of models based on the predictions of each model. CWES excels at analyzing complex data because it dynamically assigns importance to the predictions of each model based on analysis of its performance on similar training data. Benchmarking reported in the paper shows that CWES outperforms other ensembling techniques on a wide range of datasets.
“Given its significant performance improvement over the current state of the art, our approach marks a significant step forward in AI ensembling,” said Verseon’s Head of Machine Learning Ed Ratner.
CWES has broad applicability to the vast majority of real-world scenarios, for which datasets are typically complex, sparse, or both. Because CWES makes such datasets more tractable, it greatly expands the utility of AI modeling in a broad range of scenarios. The field of small-molecule drug discovery is one such scenario.
CEO Adityo Prakash said, “Our new ensembling approach demonstrates Verseon’s dedication to making groundbreaking science and technology advancements whenever currently available tools prove inadequate for the tasks at hand. This commitment leads us to innovations that have widespread application in multiple fields. And it’s how we continue to find promising drug candidates for the breakthrough treatments of tomorrow.”
About Verseon
Verseon International Corporation (www.verseon.com) is a clinical-stage, technology-driven pharmaceutical company transforming the delay, prevention, and treatment of disease. Using its Deep Quantum Modeling + AI platform, Verseon is rolling out a steady stream of life-changing medicines. Each of the company’s drug programs features multiple novel candidates with unique therapeutic properties. None of these candidates can be found by other current methods. Verseon’s fast-growing pipeline addresses major human diseases in the areas of cardiometabolic disorders and cancers. The company’s supporters and advisors include multiple Nobel laureates, former heads of R&D of major pharmaceutical companies, and various key opinion leaders in medicine.
Media Contact
Walter Jones
(510) 225-9000
View original content:https://www.prnewswire.com/news-releases/verseon-develops-superior-approach-in-combining-ai-models-for-greater-accuracy-302270921.html
SOURCE Verseon International Corporation
