New paper: We recast automated scientific model discovery with LLMs as Bayesian inference! LLMs write code and carry domain knowledge, great for proposing models. The key idea: discovery is inference, not just generation. What distribution of models explains the data? 🧵 arxiv.…
AI Weekly's analysis
→
- The authors introduce ModelSMC, an algorithm based on Sequential Monte Carlo sampling that treats candidate scientific models as weighted particles.
- They recast LLM-based model discovery as sampling from an unknown distribution over mechanistic models capable of explaining observational data.
- Reported experiments on unnamed real-world scientific systems claim interpretable mechanisms and improved posterior predictive checks.
Read full analysis →
