Linqing YU

Education

• 

  Technische Universität Darmstadt

  -

  Master of Science - MS Information and Communication Engineering

  Master Thesis: "Informative SWA-Gaussian Priors for Transfer Learning".• Implemented Stochastic Weight Averaging Gaussian (SWAG) in PyTorch on Python.• Implemented evaluation metrics such as AUSE and ECE to evaluate the predictive epistemic uncertainty which is inherited in the model parameters and compared the results with the MC Dropout approach.• Proposed Transfer Learning algorithms using SWAG posteriors as informative priors to reduce the training time and training dataset as… Mehr anzeigen Master Thesis: "Informative SWA-Gaussian Priors for Transfer Learning".• Implemented Stochastic Weight Averaging Gaussian (SWAG) in PyTorch on Python.• Implemented evaluation metrics such as AUSE and ECE to evaluate the predictive epistemic uncertainty which is inherited in the model parameters and compared the results with the MC Dropout approach.• Proposed Transfer Learning algorithms using SWAG posteriors as informative priors to reduce the training time and training dataset as well as to provide better prediction accuracy. Weniger anzeigen

• 

  Dalian University of Technology

  -

  Bachelor of Engineering - BE Electronic Information Engineering (English Intensive)

  Bachelor Thesis: "Vision-based Robust Autonomous Landing Algorithm for UAV".• Gathered dataset from real UAV-filmed video sequences to build a benchmarkfor UAV landing sign tracking.• Extracted "SURF” feature which is a fast and scale-invariant feature inorder to detect interest points in the landing sign template and the testingimages. Extracted "Color names" feature which transfers the 3-dimensionalRGB pixels to a feature with 11 commonly used colors in English.•… Mehr anzeigen Bachelor Thesis: "Vision-based Robust Autonomous Landing Algorithm for UAV".• Gathered dataset from real UAV-filmed video sequences to build a benchmarkfor UAV landing sign tracking.• Extracted "SURF” feature which is a fast and scale-invariant feature inorder to detect interest points in the landing sign template and the testingimages. Extracted "Color names" feature which transfers the 3-dimensionalRGB pixels to a feature with 11 commonly used colors in English.• Mapped the detected interest points in the images with the extracted feature.• Used DSST tracker to track the landing sign from the real world UAV filming sequences. Weniger anzeigen