AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Landmark-Guided Policy Optimization for Multi-Objective Language Model Selection

Marcio Monteiro, Weichen Li, Puyu Wang, Marius Kloft, Sophie Fellenz

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression

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 changedWhereMeasuredEffect
use learned RL allocation policy with landmark fine-tuning instead of MO-ASHALAMPS · 12 held-out datasetstime to 99% of optimal hypervolume: best on 9 of 12 datasets vs 3 for MO-ASHAImproved
use LAMPS instead of exhaustive searchLLM selection for fine-tuning · Amazon (Cell Phone)cost (USD): $5,141.67 -> $449.58Improved
use LAMPS instead of exhaustive searchLLM selection for fine-tuning · 12 held-out datasetssearch time: 73.6% reductionImproved
use LAMPS instead of MO-ASHA for stabilityLAMPS · 12 held-out datasetssearch time stabilityImproved
LAMPS fails to reach optimal hypervolume promptly in some casesLAMPS · 12 held-out datasetstime to optimality: near-optimal quickly in 9 of 12 casesMixed
use sparse reward |X\Xt|/delta_t instead of hypervolume minus lambda|X| (ablation)PPO policy · policy trainingtraining stabilityImproved
add invalid action masking for completed models (ablation)RL policy · policy trainingexploration efficiencyImproved
meta-learn on 11 datasets before deploymentLAMPS · leave-one-out over 12 datasetspayback in future tasks: 5,268 GPU-hours amortized after 13.9 tasksImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy