Active hashing and its application to image and text retrieval

In recent years, hashing-based methods for large-scale similarity search have sparked considerable research interests in the data mining and machine learning communities. While unsupervised hashing-based methods have achieved promising successes for metric similarity, they cannot handle semantic similarity which is usually given in the form of labeled point pairs. To overcome this limitation, some attempts have recently been made on semi-supervised hashing which aims at learning hash functions from both metric and semantic similarity simultaneously. Existing semi-supervised hashing methods can be regarded as passive hashing since they assume that the labeled pairs are provided in advance. In this paper, we propose a novel framework, called active hashing, which can actively select the most informative labeled pairs for hash function learning. Specifically, it identifies the most informative points to label and constructs labeled pairs accordingly. Under this framework, we use data uncertainty as a measure of informativeness and develop a batch mode algorithm to speed up active selection. We empirically compare our method with a state-of-the-art passive hashing method on two benchmark data sets, showing that the proposed method can reduce labeling cost as well as overcome the limitations of passive hashing.     

Ketamine effects on the urogenital system--changes in the urinary bladder and sperm motility

Different doses of ketamine (10 mg/kg, 20 mg/kg, 30 mg/kg, 40 mg/kg, 50 mg/kg, and 60 mg/kg) were injected i.p. (I.P.), respectively, to male ICR mice to determine the optimal dosage for chronic administration. At and above 40 mg/kg I.P. injection, mice had almost no hindlimb movement during swimming test. Subsequently, 30 mg/kg was used as the dose for the study in the toxicity of long-term ketamine administration on urinary bladder and sperm motility. The treatment group were subdivided into two (n = 10 each group); one received daily ketamine treatment i.p. for 3 months and another group for 6 months. Corresponding number of mice in control groups (n = 5 each group) received saline injection instead of ketamine. Terminal dUTP nick and labeling (TUNEL) study and Sirius red staining were carried out on the sectioned slides of the urinary bladders to study the degree of apoptosis in both epithelium and muscular layers of the urinary bladder and the relative thickness of the muscular layers in this organ was also computed. Apoptosis in the bladder epithelium was observed initially in the 3-month ketamine treated mice and the number of apoptotic cells was significantly different (P < 0.05) between the 3-month and 6-month ketamine treated mice and the control. The relative thickness of muscular layers in the bladder wall also decreased significantly (P < 0.05) when the 6-month treated mice and the control were compared. Sirius red staining revealed increase of collagen in the urinary bladder of the treated mice, most evidently 6 months after ketamine treatment. In addition, the sperm motility was studied and there was a statistically significant difference between the control and ketamine treated groups in the percentages of sperms which were motile (P < 0.05). This suggested that the chronic administration of ketamine affected the genital system as well.     

Robust path-based spectral clustering

Spectral clustering and path-based clustering are two recently developed clustering approaches that have delivered impressive results in a number of challenging clustering tasks. However, they are not robust enough against noise and outliers in the data. In this paper, based on M-estimation from robust statistics, we develop a robust path-based spectral clustering method by defining a robust path-based similarity measure for spectral clustering under both unsupervised and semi-supervised settings. Our proposed method is significantly more robust than spectral clustering and path-based clustering. We have performed experiments based on both synthetic and real-world data, comparing our method with some other methods. In particular, color images from the Berkeley segmentation data set and benchmark are used in the image segmentation experiments. Experimental results show that our method consistently outperforms other methods due to its higher robustness.     

A scalable kernel-based semisupervised metric learning algorithm with out-of-sample generalization ability

In recent years, metric learning in the semisupervised setting has aroused a lot of research interest. One type of semisupervised metric learning utilizes supervisory information in the form of pairwise similarity or dissimilarity constraints. However, most methods proposed so far are either limited to linear metric learning or unable to scale well with the data set size. In this letter, we propose a nonlinear metric learning method based on the kernel approach. By applying low-rank approximation to the kernel matrix, our method can handle significantly larger data sets. Moreover, our low-rank approximation scheme can naturally lead to out-of-sample generalization. Experiments performed on both artificial and real-world data show very promising results.     

Approximations and reducts with covering generalized rough sets

The covering generalized rough sets are an improvement of traditional rough set model to deal with more complex practical problems which the traditional one cannot handle. It is well known that any generalization of traditional rough set theory should first have practical applied background and two important theoretical issues must be addressed. The first one is to present reasonable definitions of set approximations, and the second one is to develop reasonable algorithms for attributes reduct. The existing covering generalized rough sets, however, mainly pay attention to constructing approximation operators. The ideas of constructing lower approximations are similar but the ideas of constructing upper approximations are different and they all seem to be unreasonable. Furthermore, less effort has been put on the discussion of the applied background and the attributes reduct of covering generalized rough sets. In this paper we concentrate our discussion on the above two issues. We first discuss the applied background of covering generalized rough sets by proposing three kinds of datasets which the traditional rough sets cannot handle and improve the definition of upper approximation for covering generalized rough sets to make it more reasonable than the existing ones. Then we study the attributes reduct with covering generalized rough sets and present an algorithm by using discernibility matrix to compute all the attributes reducts with covering generalized rough sets. With these discussions we can set up a basic foundation of the covering generalized rough set theory and broaden its applications.     

Mosaic organization of DNA nucleotides

Long-range power-law correlations have been reported recently for DNA sequences containing noncoding regions. We address the question of whether such correlations may be a trivial consequence of the known mosaic structure ("patchiness") of DNA. We analyze two classes of controls consisting of patchy nucleotide sequences generated by different algorithms--one without and one with long-range power-law correlations. Although both types of sequences are highly heterogenous, they are quantitatively distinguishable by an alternative fluctuation analysis method that differentiates local patchiness from long-range correlations. Application of this analysis to selected DNA sequences demonstrates that patchiness is not sufficient to account for long-range correlation properties.     

Long-range correlation properties of coding and noncoding DNA sequences: GenBank analysis

An open question in computational molecular biology is whether long-range correlations are present in both coding and noncoding DNA or only in the latter. To answer this question, we consider all 33301 coding and all 29453 noncoding eukaryotic sequences--each of length larger than 512 base pairs (bp)--in the present release of the GenBank to dtermine whether there is any statistically significant distinction in their long-range correlation properties. Standard fast Fourier transform (FFT) analysis indicates that coding sequences have practically no correlations in the range from 10 bp to 100 bp (spectral exponent beta=0.00 +/- 0.04, where the uncertainty is two standard deviations). In contrast, for noncoding sequences, the average value of the spectral exponent beta is positive (0.16 +/- 0.05) which unambiguously shows the presence of long-range correlations. We also separately analyze the 874 coding and the 1157 noncoding sequences that have more than 4096 bp and find a larger region of power-law behavior. We calculate the probability that these two data sets (coding and noncoding) were drawn from the same distribution and we find that it is less than 10(-10). We obtain independent confirmation of these findings using the method of detrended fluctuation analysis (DFA), which is designed to treat sequences with statistical heterogeneity, such as DNA's known mosaic structure ("patchiness") arising from the nonstationarity of nucleotide concentration. The near-perfect agreement between the two independent analysis methods, FFT and DFA, increases the confidence in the reliability of our conclusion.     

Characterization of Nonbonded Poly(ethylene oxide) Coating for Capillary Electrophoresis via Continuous Monitoring of Electroosmotic Flow

We examined changes in a poly(ethylene oxide) (PEO) coating by continuously monitoring the electroosmotic flow (EOF) in a fused-silica capillary during electrophoresis. An imaging CCD camera was used to follow the motion of a fluorescent neutral marker zone along the length of the capillary. The PEO coating was shown to reduce the velocity of EOF by more than 1 order of magnitude compared to a bare capillary at pH 7.0. However, it did not reduce EOF efficiently at pH 8.2. The coating protocol was important, especially at an intermediate pH of 7.7. Capillary reconditioning with an acidified solution of PEO was necessary in order to create a stable and efficient coating. In all cases we observed a gradual increase of EOF during extended runs, suggesting that the coating is slowly being degraded. The increase of pH in the cathodic (detection-end) buffer reservoir beyond pH ～8.0, e.g., as a result of electrolysis, had a large impact on the stability of the coating. This phenomenon may be used for the efficient and fast regeneration of the column surface and provides a simpler and more reliable alternative to pressure flushing of the capillary.     

An economical piezoelectric phase modulator for fiber optic sensors

A homemade piezoelectric phase modulator for interfero-metric fiber optic sensors was fabricated using piezoelectric buzzers as strain elements. Six piezoelectric elements were embedded between the two halves of a bakelite cylinder split along its axis and secured tightly together again to form a cylinder. Single-mode optical fiber was then wound around the cylinder to complete the unit. Up to a frequency of 500 Hz, the phase shift produced by the modulator is linearly proportional to the amplitude of the applied voltage. The sensitivity of the phase modulator is about 3.6 rad/V and has a dynamic range of 1,000 rad, which is sufficient for most phase modulation purposes.