Convolutional Neural Networks Optimization
Currently the Artificial Intelligence group is investigating Convolutional Neural Networks (CNN) optimization using binarization and Semantic Pruning. The purpose of these approaches and method is to:
-
Reduce the size of CNNs and determine the minimal complexity of CNN parameters in order to make them current hardware friendly.
-
Understand the learning and the process of storing learned patterns in the featre space
-
Provide a theory for understanding the learning and inference processes (Explainable AI)
Publications
- Abdiyeva, K. and Lukac, M. and Ahuja, N., Remove to Improve?, In: Del Bimbo A. et al. (eds) Pattern Recognition. ICPR International Workshops and Challenges. ICPR 2021. Lecture Notes in Computer Science, vol 12663. Springer, Cham. https://doi.org/10.1007/978-3-030-68796-0_11
- Abdiyeva, K. and Tlebeyev, T. and Lukac, M. Capacity Limits of Fully Binary CNN, ISMVL 2020, Accepted
- Abdiyeva K., Yap K.H., Wang G., Ahuja N., Lukac M., QL-Net: Quantized-by-LookUp CNN, ICARCV 2018, pp. 413-418
- Lukac, M., Pruned Single Class Deep Convolutional Detectors, The 6th International Symposium on Brainware LSI, Research Institute of Electrical Communication, Tohoku University, Sendai, Japan