MGDA-Decoupled: Geometry-Aware Multi-Objective Optimisation for DPO-based LLM Alignment

arXiv:2604.20685v1 Announce Type: new Abstract: Aligning large language models (LLMs) to desirable human values requires balancing multiple, potentially conflicting objectives such as helpfulness, truthfulness, and harmlessness, which presents a multi-objective optimisation challenge. Most alignment pipelines rely on a fixed scalarisation of these objectives, which can introduce procedural unfairness by systematically under-weighting harder-to-optimise or minority objectives. To promote more equitable trade-offs, we introduce MGDA-Decoupled, a geometry-based multi-objective optimisation algorithm that finds a shared descent direction while explicitly accounting for each objective's convergence dynamics. In contrast to prior methods that depend on reinforcement learning (e.g., GAPO) or explicit reward models (e.g., MODPO), our approach operates entirely within the lightweight Direct Preference Optimisation (DPO) paradigm. Experiments on the UltraFeedback dataset show that geometry-aware methods -- and MGDA-Decoupled in particular -- achieve the highest win rates against golden responses, both overall and per objective.

Related

• GIFT: Group-Relative Implicit Fine-Tuning Integrates GRPO with DPO and UNA
• Meet Dynamic Individual Preferences: Resolving Conflicting Human Value with Paired Fine-Tuning
• Pareto-Optimal Offline Reinforcement Learning via Smooth Tchebysheff Scalarization
• Continual Safety Alignment via Gradient-Based Sample Selection

Source: arXiv cs.LG | 2026-04-23