AI Paper Atlas

Research map / Large language models / Alignment and preferences

Multi-Objective Preference Optimization: Improving Human Alignment of Generative Models

Akhil Agnihotri, Rahul Jain, Deepak Ramachandran, Zheng Wen

ICML 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 0 citations (OpenAlex)

Read the paper

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 the Lagrangian formulation (MOPO-Lag) instead of the log-barrier formulation (MOPO-LB) (ablation)MOPO · Helpful Assistant helpful-harmless, Zephyr-7b-betanormalized reward r1: 0.51 -> 0.53Improved
constrain secondary objectives instead of scalarizing with MODPOMODPO · Three-objective Helpful Assistant, Zephyr-7b-betanormalized reward (helpful): 0.04 -> 0.39Improved
use multi-objective MOPO instead of single-objective RLHFsingle-reward RLHF · Three-objective Helpful Assistant, Zephyr-7b-betaharmless reward: -0.23 -> 0.17Improved
train DPO only on pairs where both objectives agree (DPO on DJ) versus MOPODPO on DJ · Helpful Assistant helpful-harmless, Zephyr-7b-betanormalized reward r1: 0.53 -> 0.34Hurt
vary constraint relaxation parameter beta from 0.9 to 0.999 (ablation)MOPO · Llama-3.1-8B, Helpful Assistant and Reddit Summaryaverage rewardNo clear effect
vary lagged-reference interval t0 (1, 100, 1000, never) (ablation)MOPO · Llama-3.1-8B, Helpful Assistant and Reddit Summaryaverage rewardNo clear effect
add MOPO alignment to open-source 7B chat modelsZephyr-7b-beta and other SFT models · Helpful Assistant helpfulness vs harmlessnesshelpfulness and harmlessnessImproved
impose length constraints as secondary objectivesMOPO · Verbosity-constrained task, Zephyr-7b-betaresponse length within range; helpfulnessImproved
lagged reference policy updates and adaptive constraint scheduleMOPO · general optimizationoptimization stabilityImproved
MOPO versus PARM on Mistral-7b-v0.2 Instruct helpful-harmlessPARM · Helpful Assistant helpful-harmless, Mistral-7b-v0.2normalized reward r1: PARM 0.48±0.03 vs MOPO-Lag 0.45±0.04Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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