SciDER: Data-Centric AI System for Automated Scientific Discovery
SciDER, a novel end-to-end system focused on data, leverages large language models to streamline the research lifecycle. In contrast to conventional systems, SciDER employs dedicated agents that work together to interpret and analyze unrefined scientific data, formulate hypotheses, and design experiments based on specific data traits while also generating and executing relevant code. This system stands out in data-centric scientific discovery, surpassing general-purpose agents and leading models due to its adaptive memory and feedback loop driven by critics. Available as a modular Python package with a user-friendly web interface, SciDER is designed to enhance autonomous, data-driven research. It was tested on three benchmarks, showcasing its exceptional ability to handle raw scientific experimental data.
Key facts
- SciDER is a data-centric end-to-end system for automated scientific discovery.
- It uses large language models to automate the research lifecycle from ideation to experimentation.
- Specialized agents collaboratively parse and analyze raw scientific data.
- Agents generate hypotheses and experimental designs based on data characteristics.
- The system writes and executes code for experiments.
- SciDER outperforms general-purpose agents and state-of-the-art models.
- It features a self-evolving memory and critic-led feedback loop.
- SciDER is distributed as a modular Python package with a lightweight web interface.
Entities
—