Application of Ensemble learning for computing semantic textual similarity

Semantic Textual Similarity (STS) evaluation
assesses the degree to which two parts
of texts are similar, based on their semantic
evaluation. In this paper, we describe three
models submitted to STS SemEval 2017.
Given two English parts of a text, each of
proposed methods outputs the assessment
of their semantic similarity.
We propose an approach for computing
monolingual semantic textual similarity
based on an ensemble of three distinct
methods. Our model consists of recursive
neural network (RNN) text auto-encoders
ensemble with supervised a model of vectorized
sentences using reduced part of
speech (PoS) weighted word embeddings
as well as unsupervised a method based
on word coverage (TakeLab). Additionally,
we enrich our model with additional
features that allow disambiguation of ensemble
methods based on their efficiency.
We have used Multi-Layer Perceptron as an
ensemble classifier basing on estimations
of trained Gradient Boosting Regressors.
Results of our research proves that using
such ensemble leads to a higher accuracy
due to a fact that each memberalgorithm
tends to specialize in particular
type of sentences. Simple model based
on PoS weighted Word2Vec word embeddings
seem to improve performance of
more complex RNN based auto-encoders in
the ensemble. In the monolingual EnglishEnglish
STS subtask our Ensemble based
model achieved mean Pearson correlation
of .785 compared with human annotators.


Evaluation of interaction dynamics of concurrent processes : Journal of Electrical Engineering

The purpose of this paper is to present the wavelet tools that enable the detection of temporal interactions of concurrent processes. In particular, the determination of interaction coherence of time-varying signals is achieved using a complex continuous wavelet transform. This paper has used electrocardiogram (ECG) and seismocardiogram (SCG) data set to show multiple continuous wavelet analysis techniques based on Morlet wavelet transform. MATLAB Graphical User Interface (GUI), developed in the reported research to assist in quick and simple data analysis, is presented. These software tools can discover the interaction dynamics of time-varying signals, hence they can reveal their correlation in phase and amplitude, as well as their non-linear interconnections. The user-friendly MATLAB GUI enables effective use of the developed software what enables to load two processes under investigation, make choice of the required processing parameters, and then perform the analysis. The software developed is a useful tool for researchers who have a need for investigation of interaction dynamics of concurrent processes.

MRI imaging texture features in prostate lesions classification | SpringerLink

(PCa) is the most common diagnosed cancer and cause of cancer-related death among men. This paper describes novel, deep learning based PCa CAD system that uses statistical central moments and Haralick features extracted from MR images, integrated with anamnestic data. Developed system has been trained on the dataset consisting of 330 lesions and evaluated on the challenge dataset using area under curve (AUC) related to estimated receiver operating characteristic (ROC). Two configurations of our method, based on statistical and Haralick features, scored 0.63 and 0.73 of AUC values. We draw conclusions from the challenge participation and discussed further improvements that could be made to the model to improve prostate classification.

Cross-correlation of bio-signals using continuous wavelet transform and genetic algorithm. – PubMed – NCBI

We have used genetic algorithm with a fitness function based on signals convolution to find time delay between investigated signals. Two methods of cross-correlation are proposed: one that finds single delay for analyzed signals, and one returns a vector of delay values for each of wavelet transform sub-band center frequencies. Algorithms were implemented using MATLAB.

I’ve changed my surname from Sukiennik to Sobecki in September, 2015.