#United States #Cognizant #Teaneck, NJ #AI Lab #LLMs

Groundbreaking Advances in AI from Cognizant

Cognizant, a leader in IT services, has recently unveiled groundbreaking research from its AI Lab that introduces a new method for fine-tuning large language models (LLMs). With the announcement made on October 28, 2025, this innovation not only optimizes efficiency but also significantly reduces training costs when compared to conventional techniques. At the same time, Cognizant has been acknowledged for its innovation with two newly granted U.S. patents, bringing its total to an impressive 61.

The Efficiency-Focused Method

The latest method developed by Cognizant leverages evolution strategies (ES) to fine-tune LLMs, marking a departure from traditional reinforcement learning (RL) approaches. Babak Hodjat, Chief AI Officer at Cognizant, expressed excitement over this breakthrough, highlighting how it has the potential to disrupt the industry. The ES method not only requires less training data than RL but also offers enhanced accuracy in the AI’s output, substantially improving the overall quality of work.

In a research paper titled "Evolution Strategies at Scale LLM Fine-Tuning Beyond Reinforcement Learning," Cognizant's AI Lab demonstrates the successful application of ES techniques to LLMs that contain billions of parameters. This transformative approach is not only more efficient but also presents a stabilized and scalable solution for AI training processes. As the complexity of LLMs increases, the ES-based method brings forth a new level of reliability and adaptability in post-training adjustments, which have traditionally been a chaotic and complex process.

Significant Enhancements and Future Goals

Since the release of its initial ES fine-tuning code, Cognizant has achieved remarkable progress, enhancing its codebase and achieving a tenfold increase in operational speed. The lab's next objective is to apply this refined technique to fine-tune the largest existing LLMs across various complex tasks, further advancing the capabilities of AI in real-world applications.

Expanding AI Innovations: A Closer Look at New Patents

In addition to the groundbreaking fine-tuning method, Cognizant's AI Lab has secured two critical U.S. patents that reinforce its standing as a leader in AI innovation.

1. U.S. Patent No. 12,424,335 - This patent covers systems and methodologies for AI-enhanced decision-making specifically targeting epidemiological modeling. By utilizing neural networks to project trends—including those related to epidemics like COVID-19—this patent combines various LSTM models to create a unified predictor, aiming to improve forecasting accuracy even with limited data.

2. U.S. Patent No. 12,406,188 - This patent describes a system that utilizes evolutionary strategies for data augmentation and selection. By employing a population-based search, this method automates the discovery of optimal data augmentation operations, thus enhancing robustness and performance even with small datasets.

Dr. Risto Miikkulainen, a notable figure in the Cognizant AI Lab, emphasized that while deep learning offers transformative potential across many sectors, this potential is often unfulfilled due to the difficulty of obtaining extensive datasets. These recent innovations allow models to be effectively trained with smaller datasets, enhancing the versatility and reach of deep learning technologies.

Conclusion: Driving AI Forward

Cognizant continues to push the frontiers of AI technology, embodying its commitment to harnessing advanced artificial intelligence for a modernized future. Through the efforts of its skilled researchers and the transformative applications of its findings, Cognizant's AI Lab isn’t just redefining the technical landscape; it's reshaping how AI can enhance everyday life. To learn more about Cognizant and its AI initiatives, visit Cognizant's official website.