Focus and Scope

Focus: JRHO publishes research on hyperparameter optimization (HPO) in machine learning and deep learning, covering methods, theories, systems, benchmarks, and real-world applications with an emphasis on performance, computational efficiency, and reproducibility.

Scope includes (but is not limited to):

  • HPO Algorithms: Bayesian/SMBO, TPE, SMAC, Hyperband/ASHA, BOHB, PBT, evolutionary/metaheuristics (GA, PSO, CMA-ES), bandits, hybrids.

  • Hypergradient & Bilevel Optimization: implicit differentiation, differentiable HPO.

  • Multi-fidelity & Early-Stopping: surrogates, learning-curve modeling, pruning, low-fidelity proxies.

  • Multi-objective & Constrained HPO: accuracy–latency–memory–energy/cost; fairness, robustness, regulatory constraints.

  • Search-space & Priors Design: continuous/categorical/conditional spaces, transformations, schedules, warm-starts.

  • Meta-learning & Transfer HPO: cross-task/dataset/model warm starts and surrogate transfer.

  • AutoML & Pipeline Tuning: preprocessing, feature selection, ensembling, end-to-end orchestration.

  • NAS & Co-tuning: weight sharing/one-shot, NAS+HPO, low-cost NAS.

  • HPO for Foundation/Large Models: LLM/vision FMs (LoRA/QLoRA, schedulers, batch size, quantization/distillation).

  • Systems & Scalability: distributed/parallel HPO, resource scheduling, checkpointing/caching, fault tolerance, GPU/TPU/edge.

  • Federated & Privacy-Preserving HPO: communication-aware methods, DP constraints.

  • Robustness & Security: OOD/generalization, adversarial-aware HPO, uncertainty & calibration.

  • Evaluation, Benchmarks & Reproducibility: protocols, repeated runs & CIs, compute/energy reporting, statistical testing, open code/data.

  • Applications: CV, NLP, time series, recommender systems, finance, healthcare, robotics, scientific ML.

  • Societal & Environmental Impact: energy efficiency, carbon footprint, resource governance.

Article Types: research articles, short papers/letters, surveys/tutorials, systems & software papers, reproducibility reports, benchmark/challenge reports.

Out of scope: generic ML/DL contributions without an HPO perspective; tuning reports limited to naive grid/random search without rigorous methodology; non-reproducible results.