Dynamical Sampling with Langevin Normalization Flows

Gu, M. H.; Sun, Shiliang; Liu, Yan

doi:10.3390/e21111096

Cited by 3 publications

(9 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our setup however, the exact sampling is guaranteed by the M-H process, thus the KL divergence loss is no longer applicable. Like in variational inference, the normalizing flow Langevin MC (NFLMC) [11] also used a KL divergence loss. Strictly speaking, this model is a normalizing flow but not an MCMC method.…”

Section: Related Work: Other Samplers Inspired By Hmcmentioning

confidence: 99%

“…Recently, approaches have been proposed that inherit the exact sampling property from the MCMC method, while potentially mitigating the described issues of unfavorable geometry. One approach is MCMC samplers augmented with neural networks [9][10][11]; the other approach is neural transport MCMC techniques [12,13]. A disadvantage of these recent techniques is that their objectives optimize the quality of proposed samples, but do not explicitly encourage the exploration speed of the sampler.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Neural Network MCMC Sampler That Maximizes Proposal Entropy

Chen

Sommer

2021

Entropy

View full text Add to dashboard Cite

Markov Chain Monte Carlo (MCMC) methods sample from unnormalized probability distributions and offer guarantees of exact sampling. However, in the continuous case, unfavorable geometry of the target distribution can greatly limit the efficiency of MCMC methods. Augmenting samplers with neural networks can potentially improve their efficiency. Previous neural network-based samplers were trained with objectives that either did not explicitly encourage exploration, or contained a term that encouraged exploration but only for well structured distributions. Here we propose to maximize proposal entropy for adapting the proposal to distributions of any shape. To optimize proposal entropy directly, we devised a neural network MCMC sampler that has a flexible and tractable proposal distribution. Specifically, our network architecture utilizes the gradient of the target distribution for generating proposals. Our model achieved significantly higher efficiency than previous neural network MCMC techniques in a variety of sampling tasks, sometimes by more than an order magnitude. Further, the sampler was demonstrated through the training of a convergent energy-based model of natural images. The adaptive sampler achieved unbiased sampling with significantly higher proposal entropy than a Langevin dynamics sample. The trained sampler also achieved better sample quality.

show abstract

Section: Related Work: Other Samplers Inspired By Hmcmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A Neural Network MCMC Sampler That Maximizes Proposal Entropy

Chen

Sommer

2021

Entropy

View full text Add to dashboard Cite

show abstract

“…Some methods only use permissions to classify the Android app [ 11 , 12 , 21 ], while other methods combine other features (such as API, CFG, etc.) to classify [ 22 , 23 , 24 ]. Wang et al [ 12 ] analyzed the risks of individual permissions and collaborative permissions.…”

Section: Introductionmentioning

confidence: 99%

“…In machine learning, complex probabilistic models usually need to calculate complex high-dimensional integrals [ 24 ]. For example, for a classification task, we need to predict the class of the instance.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

MADFU: An Improved Malicious Application Detection Method Based on Features Uncertainty

Yuan

Tang

2020

Entropy

View full text Add to dashboard Cite

Millions of Android applications (apps) are widely used today. Meanwhile, the number of malicious apps has increased exponentially. Currently, there are many security detection technologies for Android apps, such as static detection and dynamic detection. However, the uncertainty of the features in detection is not considered sufficiently in these technologies. Permissions play an important role in the security detection of Android apps. In this paper, a malicious application detection model based on features uncertainty (MADFU) is proposed. MADFU uses logistic regression function to describe the input (permissions) and output (labels) relationship. Moreover, it uses the Markov chain Monte Carlo (MCMC) algorithm to solve features’ uncertainty. After experimenting with 2037 samples, for malware detection, MADFU achieves an accuracy of up to 95.5%, and the false positive rate (FPR) is 1.2%. MADFU's Android app detection accuracy is higher than the accuracy of directly using 24 dangerous permission. The results also indicate that the method for an unknown/new sample’s detection accuracy is 92.7%. Compared to other state-of-the-art approaches, the proposed method is more effective and efficient, by detecting malware.

show abstract

A Neural Network MCMC sampler that maximizes Proposal Entropy

Chen

Sommer

2020

Preprint

View full text Add to dashboard Cite

Markov Chain Monte Carlo (MCMC) methods sample from unnormalized probability distributions and offer guarantees of exact sampling. However, in the continuous case, unfavorable geometry of the target distribution can greatly limit the efficiency of MCMC methods. Augmenting samplers with neural networks can potentially improve their efficiency. Previous neural network based samplers were trained with objectives that either did not explicitly encourage exploration, or used a L2 jump objective which could only be applied to well structured distributions. Thus it seems promising to instead maximize the proposal entropy for adapting the proposal to distributions of any shape. To allow direct optimization of the proposal entropy, we propose a neural network MCMC sampler that has a flexible and tractable proposal distribution. Specifically, our network architecture utilizes the gradient of the target distribution for generating proposals. Our model achieves significantly higher efficiency than previous neural network MCMC techniques in a variety of sampling tasks. Further, the sampler is applied on training of a convergent energy-based model of natural images. The adaptive sampler achieves unbiased sampling with significantly higher proposal entropy than Langevin dynamics sampler.

show abstract

Dynamical Sampling with Langevin Normalization Flows

Cited by 3 publications

References 32 publications

A Neural Network MCMC Sampler That Maximizes Proposal Entropy

A Neural Network MCMC Sampler That Maximizes Proposal Entropy

MADFU: An Improved Malicious Application Detection Method Based on Features Uncertainty

A Neural Network MCMC sampler that maximizes Proposal Entropy

Contact Info

Product

Resources

About