In this work, we report a new concept of adaptive "ensemble aptamers" (ENSaptamers) that exploits the collective recognition abilities small set rationally designed, nonspecific DNA sequences to identify molecular or cellular targets discriminatively. contrast in vitro-selected aptamers, which possess specific "lock-and-key" recognition, ENSaptamers rely on pattern mimics natural olfactory gustatory systems. Nanographene oxide was employed provide low-background and highly reproducible...

We demonstrate 3D differential phase-contrast (DPC) microscopy, based on computational illumination with a programmable LED array. By capturing intensity images various angles generated by sequentially patterning an array source, we digitally refocus through depths via light field processing. The differences from taken at complementary are then used to generate DPC images, which related the gradient of phase. proposed method achieves simple, inexpensive optics and no moving parts....

Bag-of-features based approaches have become prominent for image retrieval and classification tasks in the past decade. Such methods represent an as a collection of local features, such patches key points with scale invariant feature transform (SIFT) descriptors. To improve bag-of-features methods, we first model assignments descriptors contribution functions, then propose novel multiple assignment strategy. Assuming features can be reconstructed by their neighboring visual words vocabulary,...

This paper presents a closed queuing network model to address bike queues in bike-sharing systems with finite docks. The tackles issues of spillover and user attrition due fully occupied docks shortages at stations. objective is determine throughput rates other performance metrics for these systems. To overcome computational challenges, we propose an approximation algorithm based on the developed model. Our analysis reveals intrinsic properties docks: (i) effective system rate increases...

In the machine learning problems, performance measure is used to evaluate models. Recently, number positive data points ranked at top positions (Pos@Top) has been a popular in community. this paper, we propose learn convolutional neural network (CNN) model maximize Pos@Top measure. The CNN represent multi-instance point, and classifier function predict label from its representation. We minimize loss of over training set filters parameter. parameter vector solved by Lagrange multiplier...

Automatically classifying the tissues types of region interest (ROI) in medical imaging has been a important application computer-aided diagnosis, such as classification breast parenchymal tissue mammogram. Recently, bag-of-features method show its power this field, treating each image set local features. In paper, we investigate using strategy to classify applications. Two issues are considered here: visual vocabulary learning and weighting. Although there already plenty algorithms deal...

In this paper, we investigate the bag-of-feature based medical image retrieval methods, which represent an as a bag of local features, such patchs. To describe feature bags effectively, most methods learn visual vocabulary containing number words via clustering. Different have different importance in image. The word weighting assign appropriate weights to improve performance retrieval. method, propose effective approach paper. We first analysis each word's discriminating power by modeling...

Biometrics is an effective technology for personnel identity authentication (PIA), but unimodal biometric systems which use a single trait authentication, will suffer from problems like noisy sensor data, nonuniversality, lack of distinctiveness the trait, unacceptable error rates, and spoof attacks. These can be tackled by using multi-biometrics in system. This paper investigates fusion palmprint iris features. A new scheme at score level that combines Gaussian mixture model (GMM)...

Nonnegative Matrix Factorization (NMF) has been continuously evolving in several areas like pattern recognition and information retrieval methods. It factorizes a matrix into product of 2 low-rank non-negative matrices that will define parts-based, linear representation nonnegative data. Recently, Graph regularized NMF (GrNMF) is proposed to find compact representation,which uncovers the hidden semantics simultaneously respects intrinsic geometric structure. In GNMF, an affinity graph...

In this paper, we propose a novel multitask learning method based on the deep convolutional network. The proposed network has four layers, three max-pooling and two parallel fully connected layers. To adjust to problem, learn low-rank so that relation among different tasks can be explored. We minimize number of independent parameter rows one layer explore relations tasks, which is measured by nuclear norm layer, seek matrix. Meanwhile, also regularize another sparsity penalty useful features...

Transformers are widely used in Natural Language Processing (NLP) and computer vision; the Bidirectional Encoder Representations from (BERT) is one of most popular pre-trained models for NLP applications. This paper considers dependable operations transformers using BERT studies impact soft errors; as a case study, single-precision floating point weights considered emotion classification text. Simulation by error injection conducted to assess errors on different parts model well bits...

Biometric solution for embedded device gained significant attention in the commercial and research sectors over recent years. Combining multiple biometrics may enhance performance of personal verification system accuracy reliability. This paper presents a new multi-biometric aimed at implementing on an within wide range applications. The combines voiceprint fingerprint makes decision score level. Fusion strategy is based normalization support vector machine (SVM) classifier. We tested...

Deep learning models, such as deep convolutional neural network and long-short term memory model, have achieved great successes in many pattern classification applications over shadow machine models with hand-crafted features. The main reason is the ability of to automatically extract hierarchical features from massive data by multiple layers neurons. However, other situations, existing still cannot gain satisfying results due limitation inputs models. only take instances an input point but...

Considering that proteins with similar 3D structures have functions or biological actions, classification of based on structure can lead biologists to the investigation new protein's structural, evolutionary, and functional relatedness. However, protein remains a hard task, because different classes may discriminant features unbalanced data distribution. This means standard pairwise distance computation like Euclidean does not always equal strength in feature space. Most existing methods...