results for au:Babu_N in:cs
Apr 21 2017 cs.CV
The ability to automatically learn task specific feature representations has led to a huge success of deep learning methods. When large training data is scarce, such as in medical imaging problems, transfer learning has been very effective. In this paper, we systematically investigate the process of transferring a Convolutional Neural Network, trained on ImageNet images to perform image classification, to kidney detection problem in ultrasound images. We study how the detection performance depends on the extent of transfer. We show that a transferred and tuned CNN can outperform a state-of-the-art feature engineered pipeline and a hybridization of these two techniques achieves 20\% higher performance. We also investigate how the evolution of intermediate response images from our network. Finally, we compare these responses to state-of-the-art image processing filters in order to gain greater insight into how transfer learning is able to effectively manage widely varying imaging regimes.
In this work, we consider a two-level hierarchical MIMO antenna array system, where each antenna of the upper level is made up of a subarray on the lower one. The concept of spatial multiplexing is applied twice in this situation: Firstly, the spatial multiplexing of a Line-of-Sight (LoS) MIMO system is exploited. It is based on appropriate (sub-)array distances and achieves multiplexing gain due to phase differences among the signals at the receive (sub-)arrays. Secondly, one or more additional reflected paths of different angles (separated from the LoS path by different spatial beams at the subarrays) are used to exploit spatial multiplexing between paths. By exploiting the above two multiplexing kinds simultaneously, a high dimensional system with maximum spatial multiplexing is proposed by jointly using 'phase differences' within paths and 'angular differences' between paths. The system includes an advanced hybrid beamforming architecture with large subarray separation, which could occur in millimeter wave backhaul scenarios. The possible gains of the system w.r.t. a pure LOS MIMO system are illustrated by evaluating the capacities with total transmit power constraints.