Lab's Research: Deep Adversarial Transition Learning using Cross-Grafted Generative Stacks

 Abstract: As a common approach of deep domain adaptation in computer vision, current works have mainly focused on learning domain-invariant features from different domains, achieving limited success in transfer learning. In this paper, we present a novel ``deep adversarial transition learning'' (DATL) framework that bridges the domain gap by generating some intermediate, transitional spaces between the source and target domains through the employment of adjustable, cross-grafted generative network stacks and effective adversarial learning between transitions. Specifically, variational auto-encoders (VAEs) are constructed for the domains, and bidirectional transitions are formed by cross-grafting the VAEs' decoder stacks. Generative adversarial networks are then employed to map the target domain data to the label space of the source domain, which is achieved by aligning the transitions initiated by different domains. This results in a new, effective learning paradigm, where training and testing are carried out in the associated transitional spaces instead of the original domains. Experimental results demonstrate that our method outperforms the state-of-the-art on a number of unsupervised domain adaptation benchmarks.

This is an extended version of task-cv paper.

Reference:

Hou, J.; Ding, X.; Deng, J. D. & Cranefield, S. Deep Adversarial Transition Learning using Cross-Grafted Generative Stacks Neural Networks, 2022, 149, 172-183

Comments

Post a Comment

Popular posts from this blog

Lab's weekly topic - week 50 2023 - Detailed explanation for encoder of U-Net

Image
In U-Net the encoder block comprises two components: 2D convolution and max-pooling operations. The 2D convolution is a technique in image processing that generates a feature map. While max-pooling can reduce the size/shape of the feature map. The 2D convolution equation is as follows: \begin{equation} \mathbf{F}(m, n) = \sum_j\sum_k \mathbf{K}(i, j)\mathbf{I}(m-i)(n-j) \end{equation} Where 𝑲 is the kernel or filter, pass it over the image 𝑰. The indexes of rows and columns of the results feature map are marked with m and n respectively. In Figure 1, we show what are the operations in the encoder block, including 2D convolution, max pooling and related technologies.     Figure 1.  Operations in the encoder block In Figure 1, we add a layer with the number “0” called padding, which is used to maintain the size of the original input. Then the filter K shaped as a 3 by 3 matrix performs an element-wise multiplication with the input image 𝑰. The values are summed to generate the featur
Read more

Lab's weekly topic - week 49, 2023 - What's limitations and possible solutions for a new AI tools to detect blood poisoning?

Image
  A New, Transparent AI Tool May Help Detect Blood Poisoning It is a promising application using AI for medical services. It is a hot topic using AI to help people’s health issues. This novel algorithm can scan electric records and may reduce sepsis death, but the wide application still needs more time. Summary A solution based on machine learning was developed to help physicians detect sepsis earlier, which could save lives. Johns Hopkins Hospital developed the solution and has been deployed at three different hospitals for a try. Electronic records are used directly for an automation alert.  Limitations User side: Physicians could not have a high willingness to try a new tool they are not familiar with. Solution side: The solution depends on the electronic records. It has missing fields and values in the records, which could block the solution. Model side: It is better that we can explain the results predicted by the model detecting sepsis. Deep learning usually has a better perform
Read more

Lab's recommendation: Awesome books to learn machine learning and AI (continue updating)

Image
Machine Learning in Action (MLiA) This is an old book, which was published ten years ago, in 2012. The author is Peter Harrington. At that time, there is not so popular ML tools like scikit learn or the DL framework like TF, and Pytorch. This book describes the ML concept and the necessary formula to understand the learning algorithms. Most importantly, the author gives the implementation bit by bit (not just call API) for the algorithms. These can get the reader a deep understanding of how the models/algorithms work and are built. It's super helpful! This book is good for beginner or intermediate readers. But please forget that the top 10 models in ML as the book says, actually there is no best one but just a suitable one.  Also, there is a  GitHub repo  for the code (though I think could dirt your hands following the book and get the best learning).  The code is in python 2, a popular version a decade ago (currently has stopped the support to python 2). I do not test the code in
Read more