Research map / Probabilistic and scientific modeling / Bayesian optimization
Thompson Sampling via Fine-Tuning of LLMs
Nicolas Menet, Aleksandar Terzic, Michael Hersche, Andreas Krause, Abbas Rahimi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add model fine-tuning to generation | ToSFiT · Protein Search wall-clock | latency per step: 19% overhead (1.08 s generation, 1.29 s total) | Hurt |
| add optimism-driven fine-tuned sampling policy vs unguided LLM generation | ToSFiT · FAQ Response Refinement, Protein Search, Quantum Circuit Design | best-seen reward | Improved |
| add optimism instead of undirected exploration (actor critics) | ToSFiT · three tasks | exploration stability | Improved |
| use parameter updates instead of in-context learning (FIBO) | ToSFiT · three tasks | reward | Improved |
| increase model size (ablation) | ToSFiT · FAQ Response Refinement vs Protein Search and Quantum Circuit | reward | Mixed |
| increase batch size up to 16 (ablation) | batched ToSFiT · Protein Search | sample vs iteration efficiency: up to 16 proteins in parallel | Mixed |
| add more gradient ascent steps per round (ablation) | ToSFiT · computational efficiency study | sample efficiency | Improved |
| mask number of qubits in the prompt (weak context) (ablation) | ToSFiT · Quantum Circuit Design | reward | Hurt |
| use a large learning rate (ablation) | ToSFiT · fine-tuning learning rate study | reward | Hurt |
| add entropy-like term sqrt(-2 ln pi) sigma to objective | ToSFiT · generation diversity | diversity | Improved |
Closest papers by meaning
- Large Language Models to Enhance Bayesian Optimization (ICLR 2024 · similarity 0.89)
- Inference-Aware Fine-Tuning for Best-of-N Sampling in Large Language Models (ICLR 2025 · similarity 0.88)
- Scalable Power Sampling: Unlocking Efficient, Training-Free Reasoning for LLMs via Distribution Sharpening (ICML 2026 · similarity 0.88)
- On the Limits of Test-Time Compute: Sequential Reward Filtering for Better Inference (ICML 2026 · similarity 0.88)
- Tail Competition Explains Scaling in Best-of-N Sampling for Verifiable Problems (NeurIPS 2026 · similarity 0.87)
- Adaptive Kernel Design for Bayesian Optimization Is a Piece of CAKE with LLMs (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
