Evolving Gradient Boost: A Pruning Scheme Based on Loss Improvement Ratio for Learning Under Concept Drift.

In nonstationary environments, data distributions can change over time. This phenomenon is known as concept drift, and the related models need to adapt if they are to remain accurate. With gradient boosting (GB) ensemble models, selecting which weak learners to keep/prune to maintain model accuracy under concept drift is nontrivial research. Unlike existing models such as AdaBoost, which can directly compare weak learners' performance by their accuracy (a metric between [0, 1]), in GB, weak learners' performance is measured with different scales. To address the performance measurement scaling issue, we propose a novel criterion to evaluate weak learners in GB models, called the loss improvement ratio (LIR). Based on LIR, we develop two pruning strategies: 1) naive pruning (NP), which simply deletes all learners with increasing loss and 2) statistical pruning (SP), which removes learners if their loss increase meets a significance threshold. We also devise a scheme to dynamically switch between NP and SP to achieve the best performance. We implement the scheme as a concept drift learning algorithm, called evolving gradient boost (LIR-eGB). On average, LIR-eGB delivered the best performance against state-of-the-art methods on both stationary and nonstationary data.

Dynamics of high-contrast grating vertical-cavity surface-emitting laser with lateral optical feedback by a heterostructure interface.

We investigate the dynamics of high-contrast grating vertical-cavity surface-emitting laser (HCG-VCSEL) with a lateral optical feedback cavity. The lateral optical feedback is realized by the reflection at the heterostructure interface between two different HCGs. The lateral optical feedback cavity possesses slow light which can be tuned by changing HCG parameters, and can control the dynamics of the HCG-VCSEL. The optical feedback can enhance the -3-dB bandwidth and enlarge the eye openings of diagrams of the HCG-VCSEL, and can also reduce the frequency chirp. The HCG-VCSEL with a lateral optical feedback cavity can achieve a -3-dB bandwidth of 37.7 GHz at 12 mA and eye diagrams at 60 Gbps (non-return to zero format) and 50 GBaud (4-level pulse amplitude modulation format) with sufficient openings.

A Segment-Based Drift Adaptation Method for Data Streams.

In concept drift adaptation, we aim to design a blind or an informed strategy to update our best predictor for future data at each time point. However, existing informed drift adaptation methods need to wait for an entire batch of data to detect drift and then update the predictor (if drift is detected), which causes adaptation delay. To overcome the adaptation delay, we propose a sequentially updated statistic, called drift-gradient to quantify the increase of distributional discrepancy when every new instance arrives. Based on drift-gradient, a segment-based drift adaptation (SEGA) method is developed to online update our best predictor. Drift-gradient is defined on a segment in the training set. It can precisely quantify the increase of distributional discrepancy between the old segment and the newest segment when only one new instance is available at each time point. A lower value of drift-gradient on the old segment represents that the distribution of the new instance is closer to the distribution of the old segment. Based on the drift-gradient, SEGA retrains our best predictors with the segments that have the minimum drift-gradient when every new instance arrives. SEGA has been validated by extensive experiments on both synthetic and real-world, classification and regression data streams. The experimental results show that SEGA outperforms competitive blind and informed drift adaptation methods.

Concept Drift Detection via Equal Intensity k-Means Space Partitioning.

The data stream poses additional challenges to statistical classification tasks because distributions of the training and target samples may differ as time passes. Such a distribution change in streaming data is called concept drift. Numerous histogram-based distribution change detection methods have been proposed to detect drift. Most histograms are developed on the grid-based or tree-based space partitioning algorithms which makes the space partitions arbitrary, unexplainable, and may cause drift blind spots. There is a need to improve the drift detection accuracy for the histogram-based methods with the unsupervised setting. To address this problem, we propose a cluster-based histogram, called equal intensity k -means space partitioning (EI-kMeans). In addition, a heuristic method to improve the sensitivity of drift detection is introduced. The fundamental idea of improving the sensitivity is to minimize the risk of creating partitions in distribution offset regions. Pearson's chi-square test is used as the statistical hypothesis test so that the test statistics remain independent of the sample distribution. The number of bins and their shapes, which strongly influence the ability to detect drift, are determined dynamically from the sample based on an asymptotic constraint in the chi-square test. Accordingly, three algorithms are developed to implement concept drift detection, including a greedy centroids initialization algorithm, a cluster amplify-shrink algorithm, and a drift detection algorithm. For drift adaptation, we recommend retraining the learner if a drift is detected. The results of experiments on the synthetic and real-world datasets demonstrate the advantages of EI-kMeans and show its efficacy in detecting concept drift.

Diverse Instance-Weighting Ensemble Based on Region Drift Disagreement for Concept Drift Adaptation.

Concept drift refers to changes in the distribution of underlying data and is an inherent property of evolving data streams. Ensemble learning, with dynamic classifiers, has proved to be an efficient method of handling concept drift. However, the best way to create and maintain ensemble diversity with evolving streams is still a challenging problem. In contrast to estimating diversity via inputs, outputs, or classifier parameters, we propose a diversity measurement based on whether the ensemble members agree on the probability of a regional distribution change. In our method, estimations over regional distribution changes are used as instance weights. Constructing different region sets through different schemes will lead to different drift estimation results, thereby creating diversity. The classifiers that disagree the most are selected to maximize diversity. Accordingly, an instance-based ensemble learning algorithm, called the diverse instance-weighting ensemble (DiwE), is developed to address concept drift for data stream classification problems. Evaluations of various synthetic and real-world data stream benchmarks show the effectiveness and advantages of the proposed algorithm.

Single-mode semiconductor lasers fabricated by standard photolithography for direct modulation.

The output characteristics, modal properties, far-field profiles, and dynamic modulation responses of semiconductor lasers with surface higher-order gratings fabricated by the standard photolithography are presented. Single-mode semiconductor lasers with 20th- and 37th-order gratings for the 1.55 µm wavelength range are realized. The single-mode semiconductor lasers with 20th-order gratings have lower threshold currents and higher slope efficiencies than those with 37th-order gratings. The surface higher-order grating placed closed to the output facet can deteriorate the vertical far-field profile of the semiconductor laser. However, the properties of the semiconductor laser's single-mode operation are not affected by the surface higher-order grating's position in the ridge waveguide. The -3 dB bandwidth of these single-mode semiconductor lasers can achieve 9 GHz at 100 mA, which is the highest, to the best of our knowledge, for such a kind of single-mode semiconductor laser with a surface higher-order grating.

Genome-Wide Association Analysis Dissects the Genetic Basis of the Grain Carbon and Nitrogen Contents in Milled Rice.

BACKGROUND: Carbon (C) and nitrogen (N) are two fundamental components of starch and protein, which are important determinants of grain yield and quality. The food preferences of consumers and the expected end-use of grains in different rice-growing regions require diverse varieties that differ in terms of the grain N content (GNC) and grain C content (GCC) of milled rice. Thus, it is important that quantitative trait loci (QTLs)/genes with large effects on the variation of GNC and GCC are identified in breeding programs. RESULTS: To dissect the genetic basis of the variation of GNC and GCC in rice, the Dumas combustion method was used to analyze 751 diverse accessions regarding the GNC, GCC, and C/N ratio of the milled grains. The GCC and GNC differed significantly among the rice subgroups, especially between Xian/Indica (XI) and Geng/Japonica (GJ). Interestingly, in the GJ subgroup, the GNC was significantly lower in modern varieties (MV) than in landraces (LAN). In the XI subgroup, the GCC was significantly higher in MV than in LAN. One, six, and nine QTLs, with 55 suggestively associated single nucleotide polymorphisms, were detected for the GNC, GCC, and C/N ratio in three panels during a single-locus genome-wide association study (GWAS). Three of these QTLs were also identified in a multi-locus GWAS. We screened 113 candidate genes in the 16 QTLs in gene-based haplotype analyses. Among these candidate genes, LOC_Os01g06240 at qNC-1.1, LOC_Os05g33300 at qCC-5.1, LOC_Os01g04360 at qCN-1.1, and LOC_Os05g43880 at qCN-5.2 may partially explain the significant differences between the LAN and MV. These candidate genes should be cloned and may be useful for molecular breeding to rapidly improve the GNC, GCC, and C/N ratio of rice. CONCLUSIONS: Our findings represent valuable information regarding the genetic basis of the GNC and GCC and may be relevant for enhancing the application of favorable haplotypes of candidate genes for the molecular breeding of new rice varieties with specific grain N and C contents.

Eight-channel laser array with 100 GHz channel spacing based on surface-slotted structures fabricated by standard lithography.

An eight-channel laser array with 100 GHz channel spacing based on surface-slotted structures is demonstrated. Wavelength selection is realized by using a group of micrometer-order slots fabricated by standard lithography technology. An output power of over 14 mW and a side-mode suppression ratio of better than 35 dB for each laser within the array are achieved. A laser array without tuning for 100 GHz channel spacing by standard lithography was realized for the first time, to the best of our knowledge. The laser array can be used to enhance the capacity of communication links and provides a promising light source for photonic-integrated circuits.

Two dimensional analysis of finite size high-contrast gratings for applications in VCSELs.

2-dimensional simulations of high-contrast gratings (HCGs) of finite size are carried out, targeting at their applications in vertical-cavity surface-emitting lasers (VCSELs). Finite HCGs show a very different behavior from infinite grating ones. The reflectivity of a finite HCG strongly depends on the HCG size and the source size. Our simulation results predict finite reflectivity and transmission values, well consistent with reported experimental results. The band of high reflectivity (>99.5%) of finite HCGs is less broad as compared to the infinite case. Losses into a guided mode excited in the HCG plane are identified as being at the root. This guided mode is excited due to the nonzero angular components in the finite source size, and greatly enhances the transmission and the light leakage from the slab. In addition, the simulation results show that the details of the finite HCG can shape the output beam, whilst a Gaussian-like reflected wave is typically achieved. Our simulations can explain the current discrepancies between numerical predictions of reflectivities approaching 100% and working HCG-VCSELs showing finite reflectivities and nearly Gaussian-like output. Consequently, our analysis of finite HCGs is indispensable for HCG-VCSEL design.

Polarization-insensitive subwavelength grating reflector based on a semiconductor-insulator-metal structure.

We present a polarization-insensitive subwavelength grating reflector based on a semiconductor-insulator-metal structure. The polarization-insensitive characteristic originates from the combined effect of the TM-polarized high-reflectivity high-index-contrast subwavelength grating and the TE-polarized metallic (Au) subwavelength grating with the addition of the insulator layer. The overlapped high reflectivity (>99.5%) bandwidth between the transverse electric polarization and the transverse magnetic polarization is 89 nm. This polarization-insensitive subwavelength grating reflector can be used in the applications without a preferred polarization.

Ultracompact 1×4 TM-polarized beam splitter based on photonic crystal surface mode.

We provide an improved surface-mode photonic crystal (PhC) T-junction waveguide, combine it with an improved PhC bandgap T-junction waveguide, and then provide an ultracompact 1×4 TM-polarized beam splitter. The energy is split equally into the four output waveguides. The maximal transmission ratio of each output waveguide branch equals 24.7%, and the corresponding total transmission ratio of the ultracompact 1×4 beam splitter equals 98.8%. The normalized frequency of maximal transmission ratio is 0.397(2πc/a), and the bandwidth of the ultracompact 1×4 TM-polarized beam splitter is 0.0106(2πc/a). To the best of our knowledge, this is the first time such a high-efficiency 1×4 beam splitter exploiting the nonradiative surface mode as a guided mode has been proposed. Although we only employed a 1×4 beam splitter, our design can easily be extended to other 1×n beam splitters.

Lateral cavity photonic crystal surface-emitting laser with ultralow threshold.

A lateral cavity photonic crystal (PhC) surface-emitting laser is realized on a commercial epitaxial waveguide wafer without a distributed Bragg reflector first. Energy is coupled from the lateral resonance to surface-emitting light through the Γ band edge of the PhC with a square lattice. Electrically driven 1553.8 nm surface-emitting lasing action is observed at room temperature. The threshold current density of 667 A/cm2 is ultralow for the surface-emitting laser.

Improved fake mode free plane wave expansion method.

We analyze the origin of the fake modes introduced by the plane wave expansion method with three-dimension (3D) supercell approximation. Through the detailed analysis of the energy distribution of fake modes and real modes, we propose the plane wave expansion-three planar-slab waveguides method to remove the fake modes and obtain the fake mode free band structure of a two-dimensional air hole photonic crystal slab. To the best of our knowledge, this is the first time that such a fake mode free photonic crystal band structure is presented. Our method is also definitely useful in designing other 3D devices.

Optical properties of the crescent and coherent applications.

By out-of-particle surface plasmon (SP) excitation in the near infrared range, the influences of key parameters on the basic optical properties of the Au crescent are qualitatively studied from the mode dispersion. Based on the coherent control of SP wave, a crescent pair sensor with the intensified extracted signal and the controllability of sensing is proposed. In addition, the crescent half replaced by Ag functioning as a position detector is also proposed. The particular phase of the detecting light as a detection parameter is used to improve the accuracy of the position detection.