A Hybrid Stochastic Gradient Hamiltonian Monte Carlo Method
Authors: Chao Zhang, Zhijian Li, Zebang Shen, Jiahao Xie, Hui Qian10842-10850
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical studies on both simulated and real-world datasets demonstrate the advantage of our method. |
| Researcher Affiliation | Academia | Zhejiang University 2University of Pennsylvania {zczju,lizhijian}@zju.edu.cn, zebang@seas.upenn.edu, {xiejh,qianhui}@zju.edu.cn |
| Pseudocode | Yes | Algorithm 1 Hybrid Stochastic Gradient Hamiltonian Monte Carlo (HSG-HMC) method |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code or a link to a code repository for the methodology described. |
| Open Datasets | Yes | Six publicly available benchmark datasets, a9a, mushrooms, phishing, pima, a3a and ijcnn are used for evaluation 3. The statistics of datasets are listed in Table 3. ...All the methods are tested on the standard MNIST dataset, consisting of 28 28 images (thus 784-dimensional input vectors) from 10 different classes (digits from 0 to 9), with 6 104 training samples and 104 test samples. ... 3https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/ |
| Dataset Splits | No | The paper specifies training and test sizes (e.g., "6 104 training samples and 104 test samples" for MNIST) but does not explicitly mention validation splits or percentages for all datasets. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers. |
| Experiment Setup | Yes | In HSG-HMC, we use the weight parameter strategy ρk+1 = 1 mod(k, η 1 )+1 as indicated in Corollary 2. In all the experiments, we grid search the hyperparameters of each methods. ... Practically, we need not tune u and γ in Underdamped Langevin Dynamics based methods. For example, in SG-UL-MCMC/SRVR-HMC/HSG-MCMC, u is usually fixed to 1, and γ is chosen to make e γη = 0.9. Thus, we do not include them as hyperparameters. We will discuss this more in the appendix. |