Accelerating Continuous Normalizing Flow with Trajectory Polynomial Regularization

Authors: Han-Hsien Huang, Mi-Yen Yeh7832-7839

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that our proposed method can result in 42.3% to 71.3% reduction of NFE on the task of density estimation, and 19.3% to 32.1% reduction of NFE on variational auto-encoder, while the testing losses are not affected.
Researcher Affiliation Academia Han-Hsien Huang,1,2 Mi-Yen Yeh 1 1 Institute of Information Science, Academia Sinica 2 Department of Computer Science and Engineering, Texas A&M University hanhsien.huang@tamu.edu, miyen@iis.sinica.edu.tw
Pseudocode No The paper describes the approach mathematically and textually but does not include any explicit pseudocode blocks or algorithms.
Open Source Code Yes Our code is published in our Github repository1. 1https://github.com/hanhsienhuang/CNF-TPR
Open Datasets Yes We experiment density estimation on three 2D toy data and real data sets, which are five tabular data sets from (Papamakarios, Pavlakou, and Murray 2017) and two image data sets, MNIST and CIFAR10.
Dataset Splits Yes The learning rate is set to be 5 × 10−4 and is divided by 10 when the validation error does not decrease for 35 epochs.
Hardware Specification No The paper mentions "due to our hardware limitation" and subsequent adjustments like smaller batch sizes, but it does not specify any particular hardware components such as CPU or GPU models.
Software Dependencies No The paper mentions using "the package of ODE solvers Torchdiffeq2 (Chen et al. 2018)" but does not specify a version number for this package or any other software dependencies.
Experiment Setup Yes We choose the number of sampled point as n = 4, with τ0 = t0 and τn−1 = t1 as the start and end of time steps. The other τs are randomly sampled from a uniform distribution. The degree of the polynomial regression is set to be d = 1. ... The coefficient of the polynomial regularization is α = 5. The tolerance coefficient of Dopri5 is set to be atol = rtol = 10−4 in training, and atol = rtol = 10−5 in testing to ensure precision of solving ODEs.