Reducing bit-widths of activations and weights of deep networks makes it efficient to compute and store them in memory, which is crucial in their deployments to resourcelimited devices, such as mobile phones. However, decreasing bit-widths with quantization generally yields drastically degraded accuracy. To tackle this problem, we propose to learn to quantize activations and weights via a trainable quantizer that transforms and discretizes them. Specifically, we parameterize the quantization intervals and obtain their optimal values by directly minimizing the task loss of the network. This quantization-interval-learning (QIL) allows the quantized networks to maintain the accuracy of the fullprecision (32-bit) networks with bit-width as low as 4-bit and minimize the accuracy degeneration with further bitwidth reduction (i.e., 3 and 2-bit). Moreover, our quantizer can be trained on a heterogeneous dataset, and thus can be used to quantize pretrained networks without access to their training data. We demonstrate the effectiveness of our trainable quantizer on ImageNet dataset with various network architectures such as ResNet-18, -34 and AlexNet, on which it outperforms existing methods to achieve the stateof-the-art accuracy.
A sensor network operating in open environments requires a network-wide group key for confidentiality of exchanged messages between sensor nodes. When a node behaves abnormally due to its malfunction or a compromise attack by adversaries, the central sink node should update the group key of other nodes. The major concern of this group key update procedure will be the multi-hop communication overheads of the rekeying messages due to the energy constraints of sensor nodes. Many researchers have tried to reduce the number of rekeying messages by using the logical key tree. In this paper, we propose an energy-efficient group key management scheme called Topological Key Hierarchy (TKH). TKH generates a key tree by using the underlying sensor network topology with consideration of subtreebased key tree separation and wireless multicast advantage. Based on our detailed analysis and simulation study, we compare the total rekeying costs of our scheme with the previous logical key tree schemes and demonstrate its energy efficiency.
We aimed to predict molecular subtypes of breast cancer using radiomics signatures extracted from synthetic mammography reconstructed from digital breast tomosynthesis (DBT). A total of 365 patients with invasive breast cancer with three different molecular subtypes (luminal A + B, luminal; HER2-positive, HER2; triple-negative, TN) were assigned to the training set and temporally independent validation cohort. A total of 129 radiomics features were extracted from synthetic mammograms. The radiomics signature was built using the elastic-net approach. Clinical features included patient age, lesion size and image features assessed by radiologists. In the validation cohort, the radiomics signature yielded an AUC of 0.838, 0.556, and 0.645 for the TN, HER2 and luminal subtypes, respectively. In a multivariate analysis, the radiomics signature was the only independent predictor of the molecular subtype. The combination of the radiomics signature and clinical features showed significantly higher AUC values than clinical features only for distinguishing the TN subtype. In conclusion, the radiomics signature showed high performance for distinguishing TN breast cancer. Radiomics signatures may serve as biomarkers for TN breast cancer and may help to determine the direction of treatment for these patients.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.