multi-objective-optimization

ReMedI 采用基于 Metropolis–Hastings 的推理时搜索策略，让冻结的化学语言模型在多目标药物先导化合物优化中提出并筛选分子编辑，再将搜索轨迹自蒸馏回基础模型，在 C-MuMOInstruct 和 SMDD 基准上以约 7.8× 更少的训练数据达到约 2.8× 于 RL 基线 C-MORAL 的严格成功率。

提出了 Cautious ConFIG，一个三行扩展的优化算法，修复了 ConFIG 与 Adam 等自适应优化器结合时破坏无冲突梯度方向的问题，并在多个 PINN 基准上取得一致性能提升。

该论文提出一个面向企业级GTM客户-销售人员分配的多目标权衡决策系统，整合账户价值预测、基于合成控制反事实的关系连续性估值以及多目标优化层，将原本手工的区域划分转变为可治理、可比较的场景化决策过程。

MODS is a modular framework for multi-objective protein optimization using inference-time diffusion steering with Chebyshev scalarization and fitness-surrogate ensembles, achieving competitive hypervolume across six protein design tasks without retraining the generative model.

The paper proposes EA-SURF, an a priori method that adaptively tunes scalarization weights during training so multi-task speech LLMs converge toward user-specified task preferences along the Pareto front, improving normalized hypervolume by 17.8% over fixed loss weighting across ASR, timestamp prediction, and speech translation tasks.

The paper proposes Approximate Pareto Optimality (APO), a method for collaboratively learning lightweight LLM aligner initializations that support few-shot personalization under data deficiency, grouping users by compatible updates and coordinating competing objectives toward Pareto-optimal points with proven suboptimality bounds.

作者扩展了 Cognition 的 SWE-2 奖励函数，通过调节 λₑ（即帕累托前沿在该努力水平处的导数）来主动引导帕累托前沿，而不仅仅是优化它，从而针对期望的分数/成本改进平衡进行目标化。

This paper introduces component-aware feedback for LLM-guided evolutionary search, logging which component edits changed which fitness metrics in an attribution memory, achieving baseline-quality performance in one-third of the search budget and 7.2% higher nDCG@10 on Bright reranking tasks.

MoFlow formulates agentic workflow generation as a multi-objective MDP and solves it with Convex-Hull Monte Carlo Tree Search with set-valued backups, covering the Pareto front so workflows can be retrieved for any preference without retraining. It outperforms six baselines on math, code, and QA benchmarks in terms of hypervolume.

The paper introduces Meta-Harness, a system that frames LLM harness design as a multi-objective search over accuracy, safety, and token cost, achieving superior performance with an agentic proposer using full filesystem access.

Valerian Rey from Simplex Lab will present practical techniques for multi-objective optimization using the TorchJD library at the PyTorch Conference North America 2026.

This paper introduces Priority-Constrained Descent (PCD), a gradient-based optimization framework for hierarchical multi-objective problems that preserves primary descent while allowing minimal distortion for secondary objectives.

This article previews the advancement in optimizing the Pareto frontier, likely focusing on improvements in multi-objective optimization within AI or computational contexts.

The paper proposes a deep architecture that jointly optimizes cost functions and route-ranking models for customizable route planning using differentiable shortest-path search, demonstrating significant improvements over state-of-the-art methods on real-world datasets.

This paper proposes an LLM-enhanced multi-agent reinforcement learning framework to simultaneously optimize electric vehicle charging scheduling, station profitability, and grid stability, using LLMs for feature selection and adaptive weighting, outperforming state-of-the-art methods with reduced training time.

This paper proposes a 'converge-then-diversify' approach for multi-objective Bayesian optimization that decouples convergence and diversity into two stages, showing improved performance under tight evaluation budgets and in high-dimensional problems.

MOAE proposes a Pareto-preserving evolutionary search method to simultaneously optimize multiple objectives like task performance, trajectory quality, and safety for LLM agents without fixed scalarization during search.

A physics-informed graph attention network is proposed as a surrogate for TCAD simulation, enabling faster design space exploration by predicting mesh-node values and incorporating carrier-transport physics.

The paper proposes TRACE, a transition-aware residual control framework for multi-objective materials discovery using LLM agents, which improves hit rates over the state-of-the-art baseline.

This paper compares four methods for training ethical reinforcement learning agents, evaluating them on per-episode violation distributions rather than mean rates, and demonstrates that per-episode guarantees can be achieved without compromising mean performance.