Feedback Control for Multi-Objective Graph Self-Supervision

Can multi-task self-supervised learning on graphs be coordinated without the usual tug-of-war between objectives? Graph self-supervised learning (SSL) offers a growing toolbox of pretext objectives—mutual information, reconstruction, contrastive learning—yet combining them reliably remains a challenge due to objective interference and training instability. Most multi-pretext pipelines use per-update mixing, forcing every parameter update to be a compromise, leading to three failure modes: Disagreement (conflict-induced negative transfer), Drift (nonstationary objective utility), and Drought (hidden starvation of underserved objectives). We argue that coordination is fundamentally a temporal allocation problem: deciding when each objective receives optimization budget, not merely how to weigh them. We introduce ControlG, a control-theoretic framework that recasts multi-objective graph SSL as feedback-controlled temporal allocation by estimating per-objective difficulty and pairwise antagonism, planning target budgets via a Pareto-aware log-hypervolume planner, and scheduling with a Proportional–Integral–Derivative (PID) controller. Across 9 datasets, ControlG consistently outperforms state-of-the-art baselines, while producing an auditable schedule that reveals which objectives drove learning.

Many powerful AI systems learn from data shaped like a network—social networks, the structure of molecules, or webs of citations between scientific papers. To avoid the high cost of human-labeled data, these systems train themselves on several auxiliary tasks at once, each encouraging the model to capture a different kind of pattern. Combining many tasks should yield a richer understanding, but in practice they often work against one another: they push the model in conflicting directions, their usefulness changes over the course of training, and some tasks are crowded out and never properly learned. We take a different approach, borrowing ideas from control engineering—the same principles behind a thermostat or a car's cruise control. Rather than blending all tasks together at every training step, our method, ControlG, trains on one task at a time and continuously decides which task to prioritize next, and for how long, adjusting automatically as it detects which tasks are currently struggling or interfering with one another. Across nine standard network datasets, this approach outperforms today's leading methods. It also produces a clear, interpretable record of which tasks drove the learning and when—making the training process easier to understand and trust.