Assess whether expanded optimization budgets improve deeper architectures

Determine whether expanding the hyperparameter-optimization budgets enables deeper architectures, including the LSTM, TCN, and Transformer models evaluated in the study, to close their performance gap relative to simpler architectures in remaining useful life prediction.

Background

The study evaluates MLP, LSTM, XGBoost, TCN, and Transformer architectures under fixed hyperparameter-optimization budgets. Its results indicate that simpler architectures frequently outperform deeper temporal models, but the authors acknowledge that the fixed computational budget may have disadvantaged models with larger search spaces. The unresolved issue is whether deeper architectures would perform better if given more extensive optimization resources.

References

Future work should expand optimization budgets to determine whether deeper architectures can close the performance gap, incorporate additional objectives for resource-constrained deployment, and extend evaluation to manufacturing, energy, and transportation domains.

— Calibrating Prediction Timeliness Through Multi-Objective Hyperparameter Optimization for Remaining Useful Life Prediction  (2610.01530 - Hakyemez et al., 1 Oct 2026) in Section Conclusion, paragraph beginning “Future work”