MAdam preconditions MOO solver directions with preference-conditioned curvature so that Adam's adaptive steps respect the intended metric instead of entangling it with gradient history.
Gradient-based multi-objective deep learning: Algorithms, theories, applications, and beyond
8 Pith papers cite this work, alongside 1 external citations. Polarity classification is still indexing.
8
Pith papers citing it
1
external citations · external index
representative citing papers
By sharing the B matrix across adapters instead of the A matrix, ALoRA and Fed-ALoRA deliver more balanced performance in multi-task and federated LLM fine-tuning.
Maximizing joint conditional mutual information I(Y; C_Z, W, Z | X) decomposes multi-objective LLM alignment into preference-specific DPO terms plus an I(Y;W|X) exploration term that reduces reward-distribution overlap.
SURF derives weight sampling rules from the arc-length CDF of the scalarization path to uniformly traverse the Pareto front in multi-objective optimization.
CAGE uses common-agency games and an EPEC algorithm to compute equilibrium policies that balance multiple conflicting objectives for test-time LLM alignment.
DR-MOO adds distributional robustness to multi-objective optimization and gives single-loop MGDA algorithms reaching epsilon-Pareto-stationary points in O(epsilon^{-4}) samples for nonconvex problems.
MGDA-Decoupled applies geometry-based multi-objective optimization within the DPO framework to find shared descent directions that account for each objective's convergence dynamics, yielding higher win rates on UltraFeedback.
A barrier-enforced multi-objective optimization framework for neural networks generates sharp non-crossing prediction intervals that meet exact target coverage in probabilistic forecasting.
citing papers explorer
-
MAdam: Metric-Aware Multi-Objective Adam
MAdam preconditions MOO solver directions with preference-conditioned curvature so that Adam's adaptive steps respect the intended metric instead of entangling it with gradient history.
-
Rethinking Parameter Sharing for LLM Fine-Tuning with Multiple LoRAs
By sharing the B matrix across adapters instead of the A matrix, ALoRA and Fed-ALoRA deliver more balanced performance in multi-task and federated LLM fine-tuning.
-
Multi-Objective Exploration and Preference Optimization via Mutual Information
Maximizing joint conditional mutual information I(Y; C_Z, W, Z | X) decomposes multi-objective LLM alignment into preference-specific DPO terms plus an I(Y;W|X) exploration term that reduces reward-distribution overlap.
-
SURF: Steering the Scalarization Weight to Uniformly Traverse the Pareto Front
SURF derives weight sampling rules from the arc-length CDF of the scalarization path to uniformly traverse the Pareto front in multi-objective optimization.
-
Common-agency Games for Multi-Objective Test-Time Alignment
CAGE uses common-agency games and an EPEC algorithm to compute equilibrium policies that balance multiple conflicting objectives for test-time LLM alignment.
-
Distributionally Robust Multi-Objective Optimization
DR-MOO adds distributional robustness to multi-objective optimization and gives single-loop MGDA algorithms reaching epsilon-Pareto-stationary points in O(epsilon^{-4}) samples for nonconvex problems.
-
MGDA-Decoupled: Geometry-Aware Multi-Objective Optimisation for DPO-based LLM Alignment
MGDA-Decoupled applies geometry-based multi-objective optimization within the DPO framework to find shared descent directions that account for each objective's convergence dynamics, yielding higher win rates on UltraFeedback.
-
Barrier-enforced multi-objective optimization for direct point and sharp interval forecasting
A barrier-enforced multi-objective optimization framework for neural networks generates sharp non-crossing prediction intervals that meet exact target coverage in probabilistic forecasting.