2020/04/14 by Ashish Rana, Rana, Ashish, Taranveer Singh +7
Biochemistry, Genetics and Molecular Biology · Computer Science · #Cell Image Analysis Techniques #Digital Imaging for Blood Diseases #FOS: Computer and information sciences #Image and Object Detection Techniques #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Neural Networks and Applications #Neural and Evolutionary Computing (cs.NE)
paper · pdf · doi:10.48550/arxiv.2004.06674
openalex publication_date 2020/04/14 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
The big problem for neural network models which are trained to count\ninstances is that whenever test range goes high training range generalization\nerror increases i.e. they are not good generalizers outside training range.\nConsider the case of automating cell counting process where more dense images\nwith higher cell counts are commonly encountered as compared to images used in\ntraining data. By making better predictions for higher ranges of cell count we\nare aiming to create better generalization systems for cell counting. With\narchitecture proposal of neural arithmetic logic units (NALU) for arithmetic\noperations, task of counting has become feasible for higher numeric ranges\nwhich were not included in training data with better accuracy. As a part of our\nstudy we used these units and different other activation functions for learning\ncell counting task with two different architectures namely Fully Convolutional\nRegression Network and U-Net. These numerically biased units are added in the\nform of residual concatenated layers to original architectures and a\ncomparative experimental study is done with these newly proposed changes. This\ncomparative study is described in terms of optimizing regression loss problem\nfrom these models trained with extensive data augmentation techniques. We were\nable to achieve better results in our experiments of cell counting tasks with\nintroduction of these numerically biased units to already existing\narchitectures in the form of residual layer concatenation connections. Our\nresults confirm that above stated numerically biased units does help models to\nlearn numeric quantities for better generalization results.\n