Regularity-Aware Stochastic MGDA with Adaptive Per-Iterate Conflict-Avoidant Update
Organizations: Dept. of Electrical and Computer Engineering, University of Rochester, Rochester, NY 14627
Abstract
Multi-objective learning (MOL) aims to optimize multiple objectives simultaneously. The multi-gradient descent algorithm (MGDA) constructs a common descent or conflict-avoidant (CA) direction by solving a gradient-aggregation subproblem. In stochastic settings, the vanilla stochastic MGDA method, SMG, replaces full-batch gradients with mini-batch estimates, causing bias in the resulting CA direction. Existing analyses control this error using continuity of the CA-direction mapping, but do not distinguish how its sensitivity varies with the geometry of the MGDA subproblem. Motivated by this limitation, we show that the MGDA direction map is Lipschitz continuous under reduced nondegeneracy, whereas its general -Hölder exponent on bounded sets is sharp. This geometry motivates the Multi-objective Regularity-aware (MoRe) method, which retains adaptive MGDA aggregation in the regular regime and switches to a fixed-weight linear scalarization (LS) near degeneracy. The fallback branch skips the MGDA subproblem solve, which can reduce aggregation cost on fallback iterations while avoiding a potentially noise-sensitive update. Theoretically, we establish a per-iterate CA guarantee, and a convergence rate, improving upon the previously established SMG guarantee, where hides logarithmic factors. Experiments demonstrate favorable multi-task performance and convergence behavior consistent with the theoretical rate.
Figures & tables
| Algorithm | Batch size | Conv. | CA Dist. |
| SMG [ 1 ] [ 2 , Thms 7,8] | per-iterate | ||
| CR-MOGM [ 13 , Thm 3] | - | ||
| MoCo [ 10 , Thm 2, Lem 2] | average | ||
| MoDo [ 2 , Thms 3,5] | average | ||
| MoCo+ [ 17 , Thms 1,2] | average | ||
| MoRe, Thms 1 , 2 | per-iterate |
| Method | Art | Clipart | Product | Real-World | |
| CAGrad [ 7 ] | |||||
| Deterministic MGDA | |||||
| Fixed LS | |||||
| SMG [ 1 ] | |||||
| MoCo [ 10 ] | |||||
| MoDo [ 2 ] |
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
| Notation | Descriptions |
| Model parameter or decision variable | |
| , , | Data point, training sample sequence, and mini-batch, respectively, where , is sampled with replacement from . |
| , | Scalar-valued objective functions evaluated on data point , with , or on dataset , , with , respectively |
| , , | Vector-valued objective functions evaluated on data point , with , or on dataset , with or on mini-batch , with , respectively |
| , | Gradient matrices of the empirical objective and the mini-batch objective , respectively |
| A scalar-valued population objective function, |
Multi-Objective Optimization
Multi-objective optimization studies how to find and choose solutions that balance competing goals, constraints, and preferences.
See all 286 papers