Loan default prediction has been playing a key role in credit risk management throughout the years. Existing solutions usually involve classical machine learning classifiers, e.g. logistics and SVM, but most of them need extensive feature engineering such as feature cross which requires plenty of hand-crafted feature design. In this paper, we propose a novel method to implement feature cross based on the convolutional neural network. This method is designed to extract automatically important cross features and generate cross-feature embedding from structured data which reduces the need to generate hand-crafted cross features. The experimental results show that our method can improve the performance of predicting loan default probability compared with the methods based only on classical machine learning algorithms that are widely used in loan default prediction.
Prediction Intervals (PIs) provide a method to quantify the uncertainty of deep neural networks' point forecasts. High-quality PIs should be as narrow as possible while covering a designated percentage of data points, which is an axiomatic theory. Lower Upper Bound Estimation (LUBE) method is the first to incorporate this axiom into the neural networks loss function, however it pays much attention to the interval width, and thus the coverage probability does not achieve the desired result. Consequently, the PIs are unreliable and their practical application risks are elevated. In this paper, prioritizing coverage probability, we propose a coverage-driven approach that is generalized to any neural networks model, combining bootstrap method with improved LUBE method. We show that PIs constructed by our method are more reliable, and model uncertainty is quantified using bootstrap. Moreover, compared with two novel PI methods, benchmark experiments show our method is able to reduce the mean PI width by more than 7.5% while obtaining the better results in coverage probability.
Malignant melanoma, developing from melanocytes, is a kind of high metastatic tumor. Circulating melanoma cells, as a marker for metastasis development, are found in blood or lymphatic system at the early stage. Thus, quantitative detection of circulating melanoma cells has great significance to diagnose carcinoma and monitor tumor metastasis. In contrast to in vitro detection methods and in vivo fluorescence-based flow cytometry (IVFC), the in vivo photoacoustic flow cytometry (PAFC) utilizes melanoma cells’ predominant optical absorption in the near-infrared range over other absorbers to receive the photoacoustic (PA) signals without fluorescent dye labeling in a non-invasive way. The sensitivity of the PAFC system was verified by in vitro and in vivo experiments. Besides, we solves the technical problem that blood vessels cannot be positioned in the process of in vivo detection by designing the laser signal positioning navigation system using frequency doubling technique. PAFC provides a new tool for in vivo, label-free, and noninvasive detection of circulating tumor cells (CTCs) and has strong practicality and favorable clinical prospects.
Melanoma, developing from melanocytes, is the most serious type of skin cancer. Circulating melanoma cells, the prognosis marker for metastasis, are present in the circulation at the early stage. Thus, quantitative detection of rare circulating melanoma cells is essential for monitoring tumor metastasis and prognosis evaluation. Compared with in vitro assays, in vivo flow cytometry is able to identify circulating tumor cells without drawing blood. Here, we built in vivo photoacoustic flow cytometry based on the high absorption coefficient of melanoma cells, which is applied to labelfree counting of circulating melanoma cells in tumor-bearing mice.
Melanoma is a kind of a malignant tumor of melanocytes with the properties of high mortality and high metastasis rate. The circulating melanoma cells with the high content of melanin can be detected by light absorption to diagnose and treat cancer at an early stage. Compared with conventional detection methods such as in vivo flow cytometry (IVFC) based on fluorescence, the in vivo photoacoustic flow cytometry (PAFC) utilizes melanin cells as biomarkers to collect the photoacoustic (PA) signals without toxic fluorescent dyes labeling in a non-invasive way. The information of target tumor cells is helpful for data analysis and cell counting. However, the raw signals in PAFC system contain numerous noises such as environmental noise, device noise and in vivo motion noise. Conventional denoising algorithms such as wavelet denoising (WD) method and means filter (MF) method are based on the local information to extract the data of clinical interest, which remove the subtle feature and leave many noises. To address the above questions, the nonlocal means (NLM) method based on nonlocal data has been proposed to suppress the noise in PA signals. Extensive experiments on in vivo PA signals from the mice with the injection of B16F10 cells in caudal vein have been conducted. All the results indicate that the NLM method has superior noise reduction performance and subtle information reservation.
Melanoma is known as a malignant tumor of melanocytes, which usually appear in the blood circulation at the metastasis stage of cancer. Thus the detection of circulating melanoma cells is useful for early diagnosis and therapy of cancer. Here we have developed an in vivo photoacoustic flow cytometry (PAFC) based on the photoacoustic effect to detect melanoma cells. However, the raw signals we obtain from the target cells contain noises such as environmental sonic noises and electronic noises. Therefore we apply correlation comparison and feature separation methods to the detection and verification of the in vivo signals. Due to similar shape and structure of cells, the photoacoustic signals usually have similar vibration mode. By analyzing the correlations and the signal features in time domain and frequency domain, we are able to provide a method for separating photoacoustic signals generated by target cells from background noises. The method introduced here has proved to optimize the signal acquisition and signal processing, which can improve the detection accuracy in PAFC.
In recent years, the in vivo flow cytometry (IVFC) has been a useful technology in detecting and quantifying the circulating cells dynamically in living animals, especially in the research related to the cell tracking and the cancer metastasis. In practice, however, the unstable background signals caused by the experiment animals’ respiratory movement, limb movement and photo-bleaching of tissues’ auto-fluorescence exist in many IVFC data, which could affect the accuracy of cell counting results in the following post-processing procedure, making the IVFC signals less available. Here we developed a signal processing method that could effectively correct the unstable background signals by using methods combining interpolating, fitting, automatic segmenting and wavelet-based denoising. Compared with the previously used non-correction methods, i.e., the “line-gating” method or the automatic threshold method, this method showed a higher accuracy and efficiency in counting cell numbers of IVFC signals, as well as demonstrating a better statistic results in the Pearson’s correlation coefficient R2 and the mean-squared error (MSE).
Melanoma is a malignant tumor of melanocytes. Melanoma cells have high light absorption due to melanin highly contained in melanoma cells. This property is employed for the detection of circulating melanoma cell by in vivo photoacoustic flow cytometry (PAFC), which is based on photoacoustic effect. Compared to in vivo flow cytometry based on fluorescence, PAFC can employ high melanin content of melanoma cells as endogenous biomarkers to detect circulating melanoma cells in vivo. We have developed in vitro experiments to prove the ability of PAFC system of detecting photoacoustic signals from melanoma cells. For in vivo experiments, we have constructed a model of melanoma tumor bearing mice by inoculating highly metastatic murine melanoma cancer cells, B16F10 with subcutaneous injection. PA signals are detected in the blood vessels of mouse ears in vivo. The raw signal detected from target cells often contains some noise caused by electronic devices, such as background noise and thermal noise. We choose the Wavelet denoising method to effectively distinguish the target signal from background noise. Processing in time domain and frequency domain would be combined to analyze the signal after denoising. This algorithm contains time domain filter and frequency transformation. The frequency spectrum image of the signal contains distinctive features that can be used to analyze the property of target cells or particles. The processing methods have a great potential for analyzing signals accurately and rapidly. By counting circulating melanoma cells termly, we obtain the number variation of circulating melanoma cells as melanoma metastasized. Those results show that PAFC is a noninvasive and label-free method to detect melanoma metastases in blood or lymph circulation.
Melanoma is a malignant tumor of melanocytes. Circulating melanoma cell has high light absorption due to melanin highly contained in melanoma cells. This property is employed for the detection of circulating melanoma cell by in vivo photoacoustic flow cytometry (PAFC). PAFC is based on photoacoustic effect. Compared to in vivo flow cytometry based on fluorescence, PAFC can employ high melanin content of melanoma cells as endogenous biomarkers to detect circulating melanoma cells in vivo. In our research, we developed in vitro experiments to prove the ability of PAFC system of detecting PA signals from melanoma cells. For in vivo experiments, we constructed a model of melanoma tumor bearing mice by inoculating highly metastatic murine melanoma cancer cells B16F10 with subcutaneous injection. PA signals were detected in the blood vessels of mouse ears in vivo. By counting circulating melanoma cells termly, we obtained the number variation of circulating melanoma cells as melanoma metastasized. Those results show that PAFC is a noninvasive and label-free method to detect melanoma metastases in blood or lymph circulation. Our PAFC system is an efficient tool to monitor melanoma metastases, cancer recurrence and therapeutic efficacy.
We propose a physical-layer energy-efficient receiving method based on selective sampling in an orthogonal frequency division multiplexing access passive optical network (OFDMA-PON). By using the special designed frame head, the receiver within an optical network unit (ONU) can identify the destination of the incoming frame. The receiver only samples at the time when the destination is in agreement with the ONU, while it stays in standby during the rest of the time. We clarify its feasibility through an experiment and analyze the downstream traffic delay by simulation. The results indicate that under limited delay conditions, ∼60% energy can be saved compared with the traditional receiving method in the OFDMA-PON system with 512 ONUs.
Since the underlying demand for network capacity can be satisfied only by extremely increasing transmission bit rate,
processing speed, and switching capacity, it definitely will lead to increased power consumption of network nodes.
Energy crisis is global crisis nowadays and it has brought up many problems. Power consumption is becoming a crucial
issue in designing high-performance network devices. This paper studies the power consumption model of a kind of
Optical Packet Switching (OPS) node based on Semiconductor Optical Amplifiers (SOA). Our results show that by
dynamically adjusting the bias current of SOAs, the power consumption on an OPS node can be reduced considerably,
especially when the traffic distribution across multiple ports is not balanced and inter-nodal distances vary. The total
electrical power consumption can be cut off significantly. We also show the power consumption can be further reduced
by jointly optimizing the thermoelectric cooler current.
KEYWORDS: Orthogonal frequency division multiplexing, Modulation, Signal to noise ratio, Monte Carlo methods, Receivers, Telecommunications, Optical communications, Passive optical networks, MATLAB, Computer programming
A new symbol timing synchronization scheme based on a tailored training sequence is proposed and implemented in
direct modulation and direct detection optical OFDM PON (DM-DD OFDM PON) systems. This tailored training
sequence is added in front of OFDM data symbols. By using the received signal correlation operation, receiver calculates
the peak value of metrics to achieve accurate symbol timing synchronization. The performance of this new timing
synchronization scheme is presented by simulation and experiment. The results show that the proposed scheme has high
synchronization accuracy and is suit for DM-DD OFDM PON system.
KEYWORDS: Orthogonal frequency division multiplexing, Telecommunications, Digital signal processing, Error analysis, Optical communications, Data transmission, Signal generators, Receivers, Channel projecting optics, Fiber optic communications
Pilot pattern and pilot density are often used for channel estimation in optical/electronic orthogonal frequency division
multiplexing (OFDM) systems. To enhance the performance of the channel estimation of OFDM systems, the
transmission of the OFDM signals is experimentally investigated via different pilot tone arrangements for effective
channel estimation. Three types of pilot patterns (comb-type, block-type and incline-type) are used to evaluate the output
OFDM signal in terms of error vector magnitude (EVM), based on the most commonly adopted low complexity Least
Square (LS) channel estimation algorithm. It has been shown that, the pilot insertion will definitely improve the quality
of the transmitted OFDM signals up to 6 dB in EVM. The best channel estimation is obtained by the block-type pilot
pattern, which gives nearly 1dB gain in EVM compared with the other two patterns, implying the measured channel of
the electronic cable is relatively quasi-static.
As a special case of continuous phase frequency-shift-keying (FSK), minimum-shift keying (MSK) exhibits some
different properties compared with the traditional optical phase modulation formats, such as return-to-zero (RZ)
differential-phase-shift-keying (DPSK) and differential-quadrature-phase-shift-keying (DQPSK). In this paper, we
investigated the receiver performance degradation caused by laser frequency offset between laser frequency and delay
interferometer (DI) phase, which is found to be the most critical impairment for the receiver performance in a practical
optical phase modulated system. Results show that MSK system is about double times and six times more robust to
frequency offsets than RZ-DPSK and RZ-DQPSK systems operating at the same 10Gb/s bit rate, respectively.
This paper demonstrates bandwidth requirement for optical DPQSK transmitter using one Dual-drive Mach Zehnder
Modulator (DDMZM) for 100Gb/s physical transmission. The result shows that at receiver bandwidth of 40GHz, NRZDQPSK
signal requires 60GHz modulation bandwidth at least, while RZ-DQPSK scheme demands less bandwidth,
about 40GHz for 50% duty cycle RZ-DQPSK signal.
In this article optical minimum-shift keying (MSK) modulation format in 9×40 Gbit/s wavelength-division multiplexing
(WDM) spacing of 100 GHz transmission systems is investigated and compared with return-to-zero differential
phase-shift keying (RZ-DPSK) and return-to-zero (RZ). The performance of optical MSK in presence of amplification
noise, chromatic dispersion and nonlinear effects is analytically determined. Operations over a wide range of parameters,
such as span input power, precompensation value, and chromatic dispersion parameter are performed; experimental
results show that optical MSK demonstrates special performance in transmission distance, nonlinear effects, and
dispersion tolerance.
As bandwidth-intensive and time-sensitive streams applications such as high-definition television (HDTV) get popular, there rises a demand of supporting multicast communication directly at optical layer on next-generation optical networks. The multicast-capable optical cross-connect (MC-OXCs) is the necessary device to implement multicasting at optical layer. The functional building blocks for MC-OXC are light splitters (space splitters) or multi-wavelength converters (frequency splitters). The space splitter has no wavelength conversion capability, whereas the frequency splitter has limited fanout, decreased optical signal-to-noise ratio and also is expensive. We therefore proposed a tradeoff architecture called joint multicasting capable optical cross-connect (jMC-OXC) integrating both space splitters and multi-wavelength converters. In this paper, we investigate the network performance and the physical transmission performance on the jMC-OXC architecture by dynamic simulation and experimental demonstration. Three simulation schemes are designed and the results show that the network performance of the jMC-OXCs with limited multi-wavelength converters can obtain a close performance to that with full multi-wavelength converters. A prototype of jMC-OXC is examined and its bit error rate (BER) performance is tested after passing the multi-wavelength converters.
In this paper, a novel in-band optical spectra and filter shape monitoring technique is experimentally demonstrated. Based on swept coherent detection, the proposed technique simultaneously measures the signal and ASE spectra by adjusting the polarization states of the signal and local oscillator. In our experiment, a high resolution of 0.002 nm is achieved.
IPTV-based broadband services such as interactive multimedia and video conferencing are considered as promising revenue-adding services, and multicast is proven to be a good supplier to support these applications for its reduced consumption of network bandwidth. Generally there are two approaches to implement optical layer multicast. One is space-domain multicast using space-splitter which is low cost but has wavelength continuity constraint, the other is
frequency-domain multicast using wavelength converter which resolves the wavelength continuity but with high costs. A new multicasting node which adopts both space-domain multicast and frequency-domain multicast is recently discussed. In this paper we present an experimental demonstration of the new multicasting node architecture based on space splitters and wavelength converters, measurements to characterize such a node are provided.
1+1 protection is usually used to implement fast failure recovery and provide high reliability to data connections. A typical scheme of 1+1 protection employs a primary working lightpath and a dedicated diversely-routed backup ligthpath to carry the same data, in which the same wavelengths are always used for the two lightpath, thus putting edge-disjoint restrictions on the wavelength routing. Because current standard requires <50-ms protection-switching time, a mechanical switches with several millisecond switching time can be used before the receiver for channel switching. However, with the transmission speed of single wavelength channel increasing to tens of Gb/s, a faster switching time is highly desired to reduce the amount of data loss. In this paper, we propose an out-of-band 1+1 protection scheme, which employs two different wavelengths for primary working lightpath and protection backup path, respectively. At the destination node, a fast tunable heterodyne-detection receiver is used to select the desired wavelength lightpath. The proposed fast tunable receiver significantly reduces the protection switching time between the two paths. To investigate the performance of the proposed protection scheme, a fast channel selection experiment is performed. Two optical wavelength channels were modulated simultaneously by a
Mach-Zehnder (MZ) external modulator with data rate 1.25Gb/s. The tunable heterodyne receiver switches between the working lightpath and protection lightpath to demonstrate the capability of 1+1 protection. A <0.8-ns switching time is obtained which demonstrates the potential of hitless protection switching, if the switching occurs at the transition of two neighboring '1's.
From element automatic control view, we propose a configuration algorithm for three-level cross-connects in data plane to handle with bypass, grooming and local add/drop traffic of fiber-level, band-level and wavelength-level by a abstract bipartite graph of MG-OXC and bandwidth utilization spectra graph. The configuration algorithm is evaluated by computer simulation as well as validated by experiment on our flexible Multi-functional Optical Switching Testbed (MOST).
To solve the scalability and flexibility issue in current optical network testbeds, a Multi-functional Optical Switching Testbed (MOST) has been built. Benefiting from design of modular hardware and layered software, network experiments of scalability and complicated node architecture can be made based on it. The hardware and software structure of the MOST system is explicitly analyzed in the article and demonstrations of a 12-node ASON and shared wavelength conversion ASON on MOST are also reported.
The increasing switching capacity brings the optical node with considerable complexity. Due to the limitation in cost and technology, an optical node is often designed with partial switching capability and partial resource sharing. It means that the node is of blocking to some extent, for example multi-granularity switching node, which in fact is a structure using pass wavelength to reduce the dimension of OXC, and partial sharing wavelength converter (WC) OXC. It is conceivable that these blocking nodes will have great effects on the problem of routing and wavelength assignment. Some previous works studied the blocking case, partial WC OXC, using complicated wavelength assignment algorithm. But the complexities of these schemes decide them to be not in practice in real networks. In this paper, we propose a new scheme based on the node blocking state advertisement to reduce the retry or rerouting probability and improve the efficiency of routing in the networks with blocking nodes. In the scheme, node blocking state are advertised to the other nodes in networks, which will be used for subsequent route calculation to find a path with lowest blocking probability. The performance of the scheme is evaluated using discrete event model in 14-node NSFNET, all the nodes of which employ a kind of partial sharing WC OXC structure. In the simulation, a simple First-Fit wavelength assignment algorithm is used. The simulation results demonstrate that the new scheme considerably reduces the retry or rerouting probability in routing process.
This article presents an FPGA-based Dynamic Multicasting emulator for Wide-Area Transport Networks. We setup overlay prototype network environment based on the emulator and compare the performance of pure IP multicasting and Layer 2 multicasting.
This paper provides an overview of prototyping research activities of automatically switched optical networks and transport networks (ASONs/ASTNs) in China. In recent years, China has recognized the importance and benefits of the emerging ASON/ASTN techniques. During the period of 2001 and 2002, the national 863 Program of China started the preliminary ASON research projects with the main objectives to build preliminary ASON testbeds, develop control plane protocols and test their performance in the testbeds. During the period of 2003 and 2004, the 863 program started ASTN prototyping equipment projects for more practical applications. Totally 12 ASTN equipments are being developed by three groups led by Chinese venders: ZTE with Beijing University of Posts and Telecommunications (BUPT), Wuhan Research Institute of Posts and Telecommunication (WRI) with Shanghai Jiao Tong University (SJTU), and Huawei Inc. Meanwhile, as the ASTN is maturing, some of the China’s carries are participating in the OIF’s World Interoperability Demonstration, carrying out ASTN test, or deploying ASTN backbone networks. Finally, several ASTN backbone networks being tested or deployed now will be operated by the carries in 2005. The 863 Program will carry out an ASTN field trail in Yangtse River Delta, and finally deploy the 3TNET. 3TNET stands for Tbps transmission, Tbps switching, and Tbps routing, as well as a network integrating the above techniques. A task force under the “863” program is responsible for ASTN equipment specifications and interoperation agreements, technical coordination among all the participants, schedule of the whole project during the project undergoing, and organization of internetworking of all the equipments in the laboratories and field trials.
This paper presents a multicasting extension to OIF User Network Interface (UNI) standard, which allows dynamically setup/teardown/graft/prune a transport network multicasting tree. Transport network multicasting tree means a point-to-multipoint connection that originates from an ingress Transport Network Element (TNE) port and end at multiple egress TNE ports across the transport network for transporting user services. A global multicasting tree ID is defined to identify the transport network multicasting tree. Without addition of any new abstract messages, we just add some new attributes to existing UNI abstract messages to implement the signaling actions needed to support optical multicasting. Experimental results show the performance of our proposal under different strategy.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.