On the evaluation of unsupervised outlier detection: measures,datasets, and an empirical study

The evaluation of unsupervised outlier detection algorithms is a constant challenge in data mining research. Little is known regarding the strengths and weaknesses of different standard outlier detection models, and the impact of parameter choices for these algorithms. The scarcity of appropriate benchmark datasets with ground truth annotation is a significant impediment to the evaluation of outlier methods. Even when labeled datasets are available, their suitability for the outlier detection task is typically unknown. Furthermore, the biases of commonly-used evaluation measures are not fully understood. It is thus difficult to ascertain the extent to which newly-proposed outlier detection methods improve over established methods. In this paper, we perform an extensive experimental study on the performance of a representative set of standard k nearest neighborhood-based methods for unsupervised outlier detection, across a wide variety of datasets prepared for this purpose. Based on the overall performance of the outlier detection methods, we provide a characterization of the datasets themselves, and discuss their suitability as outlier detection benchmark sets. We also examine the most commonly-used measures for comparing the performance of different methods, and suggest adaptations that are more suitable for the evaluation of outlier detection results.     

Optimizing one-shot learning with binary synapses

A network of excitatory synapses trained with a conservative version of Hebbian learning is used as a model for recognizing the familiarity of thousands of once-seen stimuli from those never seen before. Such networks were initially proposed for modeling memory retrieval (selective delay activity). We show that the same framework allows the incorporation of both familiarity recognition and memory retrieval, and estimate the network's capacity. In the case of binary neurons, we extend the analysis of Amit and Fusi (1994) to obtain capacity limits based on computations of signal-to-noise ratio of the field difference between selective and non-selective neurons of learned signals. We show that with fast learning (potentiation probability approximately 1), the most recently learned patterns can be retrieved in working memory (selective delay activity). A much higher number of once-seen learned patterns elicit a realistic familiarity signal in the presence of an external field. With potentiation probability much less than 1 (slow learning), memory retrieval disappears, whereas familiarity recognition capacity is maintained at a similarly high level. This analysis is corroborated in simulations. For analog neurons, where such analysis is more difficult, we simplify the capacity analysis by studying the excess number of potentiated synapses above the steady-state distribution. In this framework, we derive the optimal constraint between potentiation and depression probabilities that maximizes the capacity.     

Control of the polychromatic response of an optical system through the use of annular color filters

The use of annular color filters as a tool to modify the polychromatic response of an optical system is investigated. The introduction of filters with transmission that depends on the wavelength produces a significant modification of the chromaticity response. In contrast, the position in the pupil of the annuli in which the color filters are placed modifies mainly the illuminance response. The influence of different types of annular color filter on the transverse and axial responses of the aberration-free system is studied.     

Utilizing Iron's Attractive Chemical and Magnetic Properties in Microrocket Design,Extended Motion,and Unique Performance

All‐in‐one material for microrocket propulsion featuring acid‐based bubble generation and magnetic guidance is presented. Electrochemically deposited iron serves as both a propellant, toward highly efficient self‐propulsion in acidic environments, and as a magnetic component enabling complete motion control. The new microrockets display longer lifetime and higher propulsion efficiency compared to previously reported active metal zinc‐based microrockets due to the chemical properties of iron and the unique structure of the microrockets. These iron‐based microrockets also demonstrate unique and attractive cargo towing and autonomous release capabilities. The latter is realized upon loss of the magnetic properties due to acid‐driven iron dissolution. More interestingly, these bubble‐propelled microrockets assemble via magnetic interactions into a variety of complex configurations and train structures, which enrich the behavior of micromachines. Modeling of the magnetic forces during the microrocket assembly and cargo capture confirms these unique experimentally observed assembly and cargo‐towing behaviors. These findings provide a new concept of blending propellant and magnetic components into one, toward simplifying the design and fabrication of artificial micro/nanomachines, realizing new functions and capabilities for a variety of future applications.     

Application of the Galerkin and Least-Squares Finite Element Methods in the solution of 3D Poisson and Helmholtz equations

This paper presents the numerical solution, by the Galerkin and Least Squares Finite Element Methods, of the three-dimensional Poisson and Helmholtz equations, representing heat diffusion in solids. For the two applications proposed, the analytical solutions found in the literature review were used to compare with the numerical solutions. The analysis of results was made from the L₂ norm (average error throughout the domain) and L_∞ norm (maximum error in the entire domain). The results of the two applications (Poisson and Helmholtz equations) are presented and discussed for testing of the efficiency of the methods.     

Comparing sensitivities of differently oriented multi-walled carbon nanotubes integrated on silicon wafer for electrochemical biosensors

In this study, we report on multi-walled carbon nanotubes fabricated on silicon substrate with four different orientations via chemical vapor deposition. It is well-known that chemical treatments improve the nanotube electrochemical reactivity by creating edge-like defects on their exposed sidewalls. Before use, we performed an acid treatment on carbon nanotubes. To prove the effect of the treatment on these nanostructured electrodes, contact angles were measured. Then, sensitivities and detection limits were evaluated performing cyclic voltammetry. Two target molecules were used: potassium ferricyanide, an inorganic electroactive molecule, and hydrogen peroxide that is a product of reactions catalyzed by many enzymes, such as oxidases and peroxidases. Carbon nanotubes with tilted tips become hydrophilic after the treatment showing a contact angle of 22° ± 2°. This kind of electrode has shown also the best electrochemical performance. Sensitivity and detection limit values are 110.0 ± 0.5 μA/(mM cm²) and 8 μM for potassium ferricyanide solutions and 16.4 ± 0.1 μA/(mM cm²) and 24 μM using hydrogen peroxide as target compound. Considering the results of wettability and voltammetric measurements, nanotubes with tilted tips-based electrodes are found to be the most promising for future biosensing applications.     

Discrete Time Non-Homogeneous Semi-Markov Reliability Transition Credit Risk Models and the Default Distribution Functions

This paper shows how to apply discrete time non-homogeneous semi-Markov processes (DTNHSMP) with an age index to credit risk. The idea is to consider the credit risk problem as a reliability model indexed by the age and in this way, many semi-Markov results could be adapted to describe credit risk problem. The default state is seen as a “non working state”. As the semi-Markov process is a generalization of Markov process, the presented model can be seen as a more general migration model. In fact, in semi-Markov environment the distribution function (d.f.) of the waiting time before a transition can be of any type and without the strong constraints of the Markov model. Furthermore, some results on the asymptotic behavior of the presented model are given. This permits the construction of the d.f. of the default random variable for each “working state”. An example, constructed manipulating some Standard & Poor’s (S&P) data, is presented.     

Convertibility of Function Points into COSMIC Function Points: A study using Piecewise Linear Regression

Background

COSMIC Function Points and traditional Function Points (i.e., IFPUG Function Points and more recent variation of Function Points, such as NESMA and FISMA) are probably the best known and most widely used Functional Size Measurement methods. The relationship between the two kinds of Function Points still needs to be investigated. If traditional Function Points could be accurately converted into COSMIC Function Points and vice versa, then, by measuring one kind of Function Points, one would be able to obtain the other kind of Function Points, and one might measure one or the other kind interchangeably. Several studies have been performed to evaluate whether a correlation or a conversion function between the two measures exists. Specifically, it has been suggested that the relationship between traditional Function Points and COSMIC Function Points may not be linear, i.e., the value of COSMIC Function Points seems to increase more than proportionally to an increase of traditional Function Points.

Objective

This paper aims at verifying this hypothesis using available datasets that collect both FP and CFP size measures.

Method

Rigorous statistical analysis techniques are used, specifically Piecewise Linear Regression, whose applicability conditions are systematically checked. The Piecewise Linear Regression curve is a series of interconnected segments. In this paper, we focused on Piecewise Linear Regression curves composed of two segments. We also used Linear and Parabolic Regressions, to check if and to what extent Piecewise Linear Regression may provide an advantage over other regression techniques. We used two categories of regression techniques: Ordinary Least Squares regression is based on the usual minimization of the sum of squares of the residuals, or, equivalently, on the minimization of the average squared residual; Least Median of Squares regression is a robust regression technique that is based on the minimization of the median squared residual. Using a robust regression technique helps filter out the excessive influence of outliers.

Results

It appears that the analysis of the relationship between traditional Function Points and COSMIC Function Points based on the aforementioned data analysis techniques yields valid significant models. However, different results for the various available datasets are achieved. In practice, we obtained statistically valid linear, piecewise linear, and non-linear conversion formulas for several datasets. In general, none of these is better than the others in a statistically significant manner.

Conclusions

Practitioners interested in the conversion of FP measures into CFP (or vice versa) cannot just pick a conversion model and be sure that it will yield the best results. All the regression models we tested provide good results with some datasets. In practice, all the models described in the paper - in particular, both linear and non-linear ones - should be evaluated in order to identify the ones that are best suited for the specific dataset at hand.