Lightweight Neural Network Implementation and its application

Prof. Kim started researches in the field of neural network implementation from his sabbatical year. For the lightweight neural networks, complex data format and operations are converted into simple data format and operations with negligible performance degradation. 

He has interests in applying the approximate computing to the neural network implementation. Firstly, he developed several iterative structures based on logarithmic multiplications. Secondly, he is studying the application of approximate computing in the inference system. At this time, it is thought that there are many things to be researched in the training of neural networks with approximate computing and stochastic processing. At this time, the application of approximate multiplier and AdderNet in neural networks have been intensively studied.  

Besides, we have researched lightweight neural network models. Firstly, binarized neural networks are one of our research topic. Secondly, lightweight models for semantic segmentation have been developed.