Bayesian Optimistic Optimisation with Exponentially Decaying Regret
Authors: Hung Tran-The, Sunil Gupta, Santu Rana, Svetha Venkatesh
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We perform experiments on optimisation of various synthetic functions and machine learning hyperparameter tuning tasks and show that our algorithm outperforms baselines. |
| Researcher Affiliation | Academia | 1Applied Artificial Intelligence Institute, Deakin University, Geelong, Australia. |
| Pseudocode | Yes | Algorithm 1 The BOO Algorithm |
| Open Source Code | No | The paper does not contain any statements or links indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | Using MNIST dataset, we train the models with this hyperparameter setting using the 55000 patterns and then test the model on the 10000 patterns. [...] XGBoost classification We demonstrate a classification task using XGBoost (Chen & Guestrin, 2016) on a Skin Segmentation dataset 1. [...] 1https://archive.ics.uci.edu/ml/datasets/skin+segmentation |
| Dataset Splits | Yes | Using MNIST dataset, we train the models with this hyperparameter setting using the 55000 patterns and then test the model on the 10000 patterns. [...] The Skin Segmentation dataset is plit into 15% for training and 85% for testing for a classification problem. |
| Hardware Specification | No | The paper mentions 'All implementations are in Python' but does not specify any hardware details such as GPU models, CPU types, or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions 'All implementations are in Python' and refers to 'the function SGDClassifier in the scikit-learn package' and 'Adam optimizer', but it does not provide specific version numbers for any of these software dependencies. |
| Experiment Setup | Yes | For each test function, we repeat the experiments 15 times. [...] We used Mat ern kernel with ν = 4+(D+1)/2 which satisfies our assumptions, and estimated the kernel hyper-parameters automatically from data using Maximum Likelihood Estimation. [...] For our BOO algorithm, we choose parameters m, a, b and N as per Corollary 1 which suggests using N so that a = O((N 2 )1/D) 2. For Hartmann3 (D = 3) and Schwefel (D = 3) we use partitioning procedure P(8; 2, 3) with N = 200. For Shekel function (D = 4), we use P(16; 2; 4) with N = 800 so that (N 2 )1/D 2. We follow Lemma 6 in our theoretical analysis to set βp = 2log(π2p3/3η), with η = 0.05. [...] The model is trained with the Adam optimizer in 20 epochs with batch size 128. [...] Table 2. Hyperparameters for XGBoost. |