Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Online Multi-LLM Selection via Contextual Bandits Under Unstructured Context Evolution
Manhin Poon, Xiangxiang Dai, Xutong Liu, Fang Kong, John C.S. Lui, Jinhang Zuo
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 |
|---|---|---|---|
| use multi-step contextual bandit selection (Greedy LinUCB) instead of single-step routers | multi-LLM routing · Average of 4 benchmarks | accuracy (%): MixLLM 64.25 -> Greedy LinUCB 71.99 | Improved |
| add positionally-aware knapsack heuristic (PAKH) instead of budget-aware greedy (ablation) | B.A. LinUCB · Average of 4 benchmarks | accuracy (%): 69.17 -> 74.84 | Improved |
| add PAKH instead of Greedy LinUCB on average steps (ablation) | LinUCB variants · Average of 4 benchmarks | average steps: 2.52 -> 1.58 | Improved |
| add a budget constraint to Greedy LinUCB (ablation) | Greedy LinUCB · Average of 4 benchmarks | accuracy (%) and steps: accuracy 71.99 -> 69.17; steps 2.52 -> 2.04; cost 1.51E-03 -> 2.63E-04 | Mixed |
| use B.A. LinUCB instead of MixLLM / MetaLLM | multi-LLM routing · Average of 4 benchmarks | accuracy (%) and cost: MixLLM 64.25 -> 69.17; cost 2.27E-03 -> 2.63E-04 | Improved |
| use PAKH instead of B.A. LinUCB at tight budgets (below $0.001) (ablation) | budget-constrained selection · budget sensitivity | accuracy | Hurt |
| increase per-query budget (ablation) | PAKH vs B.A. LinUCB · budget sensitivity | accuracy | Mixed |
| use adaptive LLM choice per step instead of retrying the initial router choice | MetaLLM, MixLLM · AIME | accuracy (%): MetaLLM 20.83 -> Greedy LinUCB 54.16 | Improved |
Closest papers by meaning
- EVOLvE: Evaluating and Optimizing LLMs For In-Context Exploration (ICML 2025 · similarity 0.91)
- A Multi-Agent Conversational Bandit Approach to Online Evaluation and Selection of User-Aligned LLM Responses (AAAI 2026 · similarity 0.89)
- On Speeding Up Language Model Evaluation (ICLR 2025 · similarity 0.89)
- Can large language models explore in-context? (NeurIPS 2024 · similarity 0.88)
- Do LLM Agents Have Regret? A Case Study in Online Learning and Games (ICLR 2025 · similarity 0.88)
- GraphRouter: A Graph-based Router for LLM Selections (ICLR 2025 · similarity 0.88)
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.
