The system automates model validation using visual exploration agents.
Hypothesis -> experiments -> analysis -> conclusions. LLMs are great at writing code and conducting experiments.
But there’s a weakness in their ability to propose statistical models and evaluate their fit.
Enter VESTA: Visual Exploration with Statistical Tool Agents.
No Digg Deeper questions have been answered for this story yet.
Finding good statistical models is central to many scientific exploration. Check out VESTA that makes AI good at this, and new benchmark DAWN!
On names: NASA mission Dawn 🛰️ explored Vesta ☄️ @CosmicAI_Inst
Hypothesis -> experiments -> analysis -> conclusions. LLMs are great at writing code and conducting experiments.
But there’s a weakness in their ability to propose statistical models and evaluate their fit.
Enter VESTA: Visual Exploration with Statistical Tool Agents.
We test three variants of VESTA: no tools, dynamically created tools, and expert-written tools. VESTA significantly outperforms baselines, with the biggest gains on Hard and Astro. Dynamically generated tools land just behind expert tools.
VESTA explores data and refines statistical models by creating visual tools that target specific statistical hypotheses. These tool outputs then inform the next proposed model, producing a more accurate fit to the data.
We introduce DAWN: Dataset for Automated Workflows and Numerical Modeling. DAWN covers two tasks – time series modeling and distribution fitting – each with three splits: Easy, Hard, and Astro. 🤗 william-rudman/DAWN.
Check out our pre-print here: https://arxiv.org/abs/2606.00384 💻 Code: https://github.com/wrudman/VESTA 🤗HF Dataset: https://huggingface.co/datasets/william-rudman/DAWN
