Cognizant, a technology and consulting company, has announced a major advancement from its AI Lab, introducing a new approach for fine-tuning large language models using evolution strategies (ES). The research, published under the title “Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning,” demonstrates how ES can reduce training data requirements and improve task alignment compared to traditional reinforcement learning methods.
Babak Hodjat, Chief AI Officer at Cognizant’s AI Lab, said, “Our latest research… has the potential to disrupt the industry. Our approach not only uses less training data than reinforcement learning; it also makes the process more accurate.” The team also optimized its ES implementation, achieving a tenfold increase in tuning speed as it prepares to scale the method to larger models across varied use cases.
Alongside the research breakthrough, the lab received two new U.S. patents, bringing its total to 61. One patent focuses on improved epidemiological trend prediction using deep learning models with real-world constraints, while the other covers a system for automated data augmentation that enhances model robustness when data is limited.
Risto Miikkulainen, VP of Research and Computer Science Professor, noted, “These innovations enable effective model training with fewer examples, expanding the applicability of deep learning.” Cognizant’s AI Lab continues to develop Decision AI, which supports its Neuro AI platform used by enterprises to improve organizational decision-making and performance.