TY - JOUR AU - Yoo, Sunghwan AU - Haider, Masoom A. AU - Khalvati, Farzad PY - 2017/10/15 Y2 - 2024/03/28 TI - Estimating Optimal Depth of VGG Net with Tree-Structured Parzen Estimators JF - Journal of Computational Vision and Imaging Systems JA - J. Comp. Vis. Imag. Sys. VL - 3 IS - 1 SE - Articles DO - 10.15353/vsnl.v3i1.175 UR - https://openjournals.uwaterloo.ca/index.php/vsl/article/view/175 SP - AB - <p>Deep convolutional neural networks (CNNs) have shown astonishing<br />performances in variety of fields. However, different architectures<br />of the networks are required for different datasets, and finding<br />right architecture for given data has been a topic of great interest in<br />computer vision communities. One of the most important factors of<br />the CNNs architecture is the depth of the networks, which plays a<br />significant role in avoiding over-fitting. Grid Search is widely used<br />for estimating the depth, but it requires huge computation time. Motivated<br />by this, a method for finding an optimal architecture depth is<br />introduced, which is based on a hyper-parameter optimizer called<br />Tree-Structured Parzen Estimators (TPE). In this work, we show<br />that the TPE is capable of estimating the CNNs architecture depth<br />with an accuracy of 83.33% with CIFAR-10 dataset and 60.00%<br />with CIFAR-100 dataset while it reduces the computation time by<br />more 70% compared to the Grid Search.</p> ER -