Software

  1. Hyperparameter Optimization Framework for ABCI.
    Created during the internship at AIST.
  2. Multi-Objective Tree-Structured Parzen Estimator.
    Created during the internship at AIST jointly with Yoshihiko Ozaki. It is available in Optuna.
  3. Warm-Start CMA-ES.
    Created during the internship at AIST jointly with Masahiro Nomura. It is available in Optuna.
  4. Auto-PyTorch Project.
    The project at the Machine Learning Lab headed by Prof. Frank Hutter supported by BOSCH GmbH. GitHub repository.
  5. Various Extensions of Tree-Structured Parzen Estimator.
    Tree-structured Parzen estimator is a well-known hyperparameter optimizer and I extended this method to meta-learning, multi-objective, and constrained settings in this repository.
  6. Reproduction of Meta-Learning Bayesian Optimization Methods.
    I reproduced a weighted combination based meta-learning for Gaussian-process based Bayesian optimization: (1) TSTR and (2) RGPE. This implementation is exclusively for the fair comparisons in papers and I used BoTorch for the implementation to allow users to apply multi-objective optimization.
  7. Fast Implementation for Finding Pareto Solutions and Non-Domination Sort.
    I implemented a quick method to find Pareto solutions and to obtain non-domination rank. This package can be installed via pip.
  8. Visualization for Bi-Objective Optimization Comparisons with an Uncertainty Band.
    I implemented the empirical attainment function or summary attainment surface to visualize the expected performance of multi-objective optimization methods with an uncertainty band. This package can be installed via pip.
  9. A Solver of MineSweeper Using Depth First Search Based Probability Computation.
    I implemented a solver of MineSweeper that achieves the success rate of 45% in the hardest setting. This solver uses the depth first search to compute the probabilities of each cell having a mine and I speeded up the code using NumPy and caching. There might be room for improvement in the speed, but my implementation achieves the theoretical performance bound.
  10. A Simulator for Multi-Fidelity or Parallel Optimization Using Tabular or Surrogate Benchmarks.
    I implemented a simulator for multi-fidelity optimization to reduce experiments' time of parallel optimization. This simulator internally force each worker to wait so that the obtained results will come with the the exact order as we let each worker sleep for the exact runtime specified in tabular or surrogate benchmarks.
  11. MineSweeper implemented by TypeScript.
    I used React to implement this game. Enjoy it from here !!

Study

  1. Master thesis: Significant Runtime Reduction for Asynchronous Multi-Fidelity Optimization on Zero-Cost Benchmarks .
    Master thesis at the University of Freiburg completed on September 2023.
  2. The Machine Learning course in the University of Freiburg
    My notebook is based on the course offered by J. Prof. Josif Grabocka in the Winter Semester 2020/2021. Note that all the mistakes I could aware of in the course are fixed in this paper.
  3. The Deep Learning course in the University of Freiburg
    My notebook is based on the course offered by Prof. Frank Hutter in the Winter Semester 2020/2021.
  4. The Reinforcement Learning course in the University of Freiburg
    My notebook is based on the course offered by Prof. Joschka Boedecker in the Winter Semester 2020/2021.
  5. The Simulation in Computer Graphics course in the University of Freiburg
    My notebook is based on the course offered by Prof. Matthias Teschner in the Winter Semester 2020/2021.
  6. The High Performance Computing with Python course in the University of Freiburg
    This submission was written on the Summer semester 2021 and the implementation is available here.
  7. The AutoML course in the University of Freiburg
    My notebook is based on the course offered by Prof. Frank Hutter in the Summer Semester 2021.
  8. The Statistical Pattern Recognition course in the University of Freiburg
    My notebook is based on the course offered by Prof. Thomas Brox in the Summer Semester 2021.

Back to Home