Verseon is pleased to announce the publication of its peer-reviewed paper “X-ELM: A Fast Explainability Approach for Extreme Learning Machines” in Advances in Computational Intelligence. The company’s X-ELM approach provides an accurate, real-time solution to an increasingly important problem in AI modeling – how to explain AI decision making.
Being able to interpret an AI model’s decisions is vital in many mission-critical real-world AI applications. The most advanced alternative to date, SHAP, is highly computationally intensive and takes days to yield results, rendering SHAP-based explainable AI impractical in day-to-day applications. By contrast, Verseon’s novel X-ELM delivers real-time, state-of-the-art results while requiring far fewer computational resources than other methods.
Verseon’s X-ELM brings real-time explainability to a range of AI algorithms called Extreme Learning Machines (ELM). Verseon’s Extreme AutoML, a proprietary ELM-based technology, has demonstrably better accuracy than Google AutoML for a wide range of problems for which big datasets are not available. Now, Verseon’s X-ELM explainability brings a new dimension of utility to Verseon’s Extreme AutoML.
Knowing how an AI model arrives at its conclusions allows human decision makers to evaluate the validity and applicability of the model’s output – and to evaluate more effectively which variables matter in any given scenario. “Being able to see in real time how our AI generates results dramatically speeds up the process of iterative model refinement and evolution, making model results more actionable,” said Verseon’s Head of Machine Learning, Edward Ratner.
In their mission to improve human health, Verseon’s scientists consistently make breakthroughs in multiple disciplines that have profound implications in the field of drug discovery and beyond. The company’s breakthroughs in physics- and AI-driven molecule engineering allow Verseon to design drugs that cannot be found by any other current means, including others’ AI-based methods.
SOURCE: PRNewswire
