[postdoc position] Postdoc position in machine learning and computer vision at Henan University

[Fadi Dornaika]

Dear Colleagues,

Prof. Fadi Dornaika is an adjunct professor at Henan University, China.  He is interested in recruiting postdoctoral researchers at Henan University.

Interested candidates can contact Prof. Fadi Dornaika at    fdor...@gmail.com

Salary: 200,000 RMB in total per year, before tax, tax rate: about 10%.

Research funding: 100,000 RMB, once for all.

Period: 2 years.

Age: must be <= 35

The cost of living in Kaifeng is quite low. For example, the cost of housing (one bedroom, one living room) in Kaifeng is about 150 EUR per month; the cost of food in Kaifeng is about 100 EUR per month.

 Subject / Area:

Machine Learning, Pattern Recognition, Computer Vision

Possible research line:  Medical Image Analysis

Quantitative analysis of medical images plays an important role in the analysis of anatomical and pathological structures, as well as in the diagnosis of cancer and various neurological diseases. Such analysis can help clinicians make an accurate diagnosis, and this depends on accurate segmentation of the structures of interest. It is necessary to segment some important objects in medical images and extract features from the segmented areas. As 2D, 3D, and 4D imaging continues to advance and image data becomes more extensive and complex, manual processing of such data becomes increasingly difficult, tedious, and impractical. Therefore, the use of machine learning and deep neural networks to analyze such data is required. This will be the main goal of this work.

Deep learning is often divided into three main categories: supervised learning, weakly supervised learning (semi-supervised), and unsupervised learning. The advantage of supervised learning is that we can train models based on carefully labeled data. In contrast, unsupervised learning does not require labeled data, but the difficulty of learning is higher. Weakly supervised learning is in between supervised and unsupervised learning because only a small portion of the data needs to be labeled, while the majority of the data is unlabeled.

Despite of the fact that CNN achieves excellent performance, it is unable to learn global and long-range semantic information interactions due to the limitations of convolutional operations. Most segmentation results seriously depend on high-quality labels. In fact, it is challenging to build datasets with high-quality labels, especially in the medical field, since data acquisition and labeling have high costs. Thus, studies with weakly supervised learning might also be beneficial and promising. Recently proposed Transformer-based architectures that leverage the self-attention mechanism encode long-range dependencies and learn representations that are highly expressive. The self-attention mechanism is an improvement of the attention mechanism that reduces the dependence on external information to effectively capture the internal correlation of features.

Transformers have shown impressive performance on various computer vision tasks. In 2021, Alexey Dosovitskiy's Vision Transformer (ViT) was introduced for image classification. Transformers have attracted increasing interest in the context of semantic image segmentation.

The goal of this is work is to develop a new segmentation approach for medical images that can meet the existing challenges in various segmentation tasks. The plan is to develop a new CNN transformer architecture to combine their strengths and achieve stable and high performance in various segmentation tasks. In addition, the proposal will use of transfer learning and semi-supervised learning. In this thesis, the following tracks will be addressed.

First objective:  Consideration of existing methods

Second objective: Use of Dedicated Transformers-based models

Third objective: Application of weakly supervised methods

About Prof.  Fadi Dornaika

Homepage:  https://www.ikerbasque.net/es/fadi-dornaika

Fadi Dornaika is an Ikerbasque research Professor. He got his PhD degree from INRIA, France in 1995. His PhD thesis work was about geometric modeling for the integration vision/robotics. Before joining IKERBASQUE, he worked at nine international research institutes:  INRIA (France), GMD, (Germany), The Chinese University of Hong Kong (China), Linkoping University (Sweden), University of Alberta (Canada), York University (Canada), Heudiasyc laboratory (CNRS France),  Computer Vision Center (Spain), and French Geographical Institute (France). He has a rich research experience in computer vision and pattern recognition. His research interests cover a broad spectrum in computer vision, image processing, pattern recognition, and machine learning. His current research interests include Manifold Learning, Supervised Learning, Multiview Clustering, Scalable Semi-Supervised Learning, Structured Semi-Supervised Learning, Deep Learning with applications to facial age estimation, facial beauty prediction, emotion recognition, fatigue detection, and kinship verification, and Deep learning for medical image analysis.

His h-index is 36 (Google scholar). According to Stanford University's current ranking, he is in the top 2% of scholars (DOI: 10.17632/btchxktzyw.4) based on his citations on career-long data updated to end-of-2021  and single year (2021) impact. He has published more than 350 papers in the field of computer vision, pattern recognition and machine learning, including 145 JCR indexed journal articles in (IEEE Trans. Robotics & Automation, IEEE Trans. Cybernetics, IEEE Trans. Neural Networks and Learning Systems, IEEE Trans. CSVT, IEEE Trans. SMC, Information Fusion, Information Sciences, Neural Networks, Pattern Recognition, Artificial Intelligence Review,  Knowledge-Based Systems, Int. Journal of Computer Vision, Int. Journal of Robotics Research, etc.).

-----------------------------------------------------------------------------------------