Brief

Researchers have developed Eureka, an LLM-driven framework for automated feature engineering in AI. Eureka uses an Expert Agent to create feature design plans, a Feature Factory to generate Python code for these features, and a Self-Evolving Alignment Engine to refine the code. This approach treats feature creation as an agentic code generation problem, allowing learned patterns to transfer across different domains. In evaluations, Eureka outperformed existing methods on public benchmarks and achieved significant improvements in cloud GPU resource demand prediction at Alibaba Cloud. AI

Researchers have introduced a new optimization paradigm called graph-grounded optimization, which leverages property knowledge graphs as the primary input modality. This approach contrasts with existing systems that rely on natural language or static tables. The framework was implemented using the open-source samyama-graph database and evaluated across seven real-world problems, including drug repurposing and supply chain rerouting. AI