Search
mgr inż. Karol Draszawka
Publications
Filters
total: 18
Catalog Publications
Year 2023
-
From Scores to Predictions in Multi-Label Classification: Neural Thresholding Strategies
PublicationIn this paper, we propose a novel approach for obtaining predictions from per-class scores to improve the accuracy of multi-label classification systems. In a multi-label classification task, the expected output is a set of predicted labels per each testing sample. Typically, these predictions are calculated by implicit or explicit thresholding of per-class real-valued scores: classes with scores exceeding a given threshold value...
Year 2021
-
Jak wykraść złoto smokowi? - uczenie ze wzmocnieniem w świecie Wumpusa
PublicationNiniejszy rozdział zawiera łagodne wprowadzenie do problematyki uczenia ze wzmocnieniem, w którym podstawy teoretyczne wyjaśniane są na przykładzie przewodnim, jakim jest zagadnienie nauczenia agenta poruszania się w świecie potwora o imieniu Wumpus (ang. Wumpus world), klasycznym środowisku do testowania logicznego rozumowania agentów (problem nietrywialny dla algorytmów uczenia ze wzmocnieniem). Przedstawiona jest główna idea...
-
Multi-Aspect Quality Assessment Of Mobile Image Classifiers For Companion Applications In The Publishing Sector
PublicationThe paper presents the problem of quality assessment of image classifiers used in mobile phones for complimentary companion applications. The advantages of using this kind of applications have been described and a Narrator on Demand (NoD) functionality has been described as one of the examples, where the application plays an audio file related to a book page that is physically in front of the phone's camera. For such a NoD application,...
Year 2020
-
Automated Classifier Development Process for Recognizing Book Pages from Video Frames
PublicationOne of the latest developments made by publishing companies is introducing mixed and augmented reality to their printed media (e.g. to produce augmented books). An important computer vision problem that they are facing is classification of book pages from video frames. The problem is non-trivial, especially considering that typical training data is limited to only one digital original per book page, while the trained classifier...
Year 2018
-
Evaluating Asymmetric N-Grams as Spell-Checking Mechanism
PublicationTypical approaches to string comparing marks two strings as either different or equal without taking into account any similarity measures. Being able to judge similarity is however required for spelling error corrections, as we want to find the best match for a given word. In this paper we present a bi2quadro-grams method for spelling errors correction. The method proposed uses different n-grams dimension for the source (checked)...
Year 2017
-
Analysis of Denoising Autoencoder Properties Through Misspelling Correction Task
PublicationThe paper analyzes some properties of denoising autoencoders using the problem of misspellings correction as an exemplary task. We evaluate the capacity of the network in its classical feed-forward form. We also propose a modification to the output layer of the net, which we called multi-softmax. Experiments show that the model trained with this output layer outperforms traditional network both in learning time and accuracy. We...
Year 2016
-
Towards increasing F-measure of approximate string matching in O(1) complexity
PublicationThe paper analyzes existing approaches for approximate string matching based on linear search with Levenshtein distance, AllScan and CPMerge algorithms using cosine, Jaccard and Dice distance measures. The methods are presented and compared to our approach that improves indexing time using Locally Sensitive Hashing. Advantages and drawbacks of the methods are identified based on theoretical considerations as well as empirical evaluations...
Year 2015
-
Improving css-KNN Classification Performance by Shifts in Training Data
PublicationThis paper presents a new approach to improve the performance of a css-k-NN classifier for categorization of text documents. The css-k-NN classifier (i.e., a threshold-based variation of a standard k-NN classifier we proposed in [1]) is a lazy-learning instance-based classifier. It does not have parameters associated with features and/or classes of objects, that would be optimized during off-line learning. In this paper we propose...
-
Improving Effectiveness of SVM Classifier for Large Scale Data
PublicationThe paper presents our approach to SVM implementation in parallel environment. We describe how classification learning and prediction phases were pararellised. We also propose a method for limiting the number of necessary computations during classifier construction. Our method, named one-vs-near, is an extension of typical one-vs-all approach that is used for binary classifiers to work with multiclass problems. We perform experiments...
Year 2014
-
Emotion Recognition Based on Facial Expressions of Gamers
PublicationThis article presents an approach to emotion recognition based on facial expressions of gamers. With application of certain methods crucial features of an analyzed face like eyebrows' shape, eyes and mouth width, height were extracted. Afterwards a group of artificial intelligence methods was applied to classify a given feature set as one of the following emotions: happiness, sadness, anger and fear. The approach presented in this...
-
How Specific Can We Be with k-NN Classifier?
PublicationThis paper discusses the possibility of designing a two stage classifier for large-scale hierarchical and multilabel text classification task, that will be a compromise between two common approaches to this task. First of it is called big-bang, where there is only one classifier that aims to do all the job at once. Top-down approach is the second popular option, in which at each node of categories’ hierarchy, there is a flat classifier...
-
Towards Increasing Density of Relations in Category Graphs
PublicationIn the chapter we propose methods for identifying new associations between Wikipedia categories. The first method is based on Bag-of-Words (BOW) representation of Wikipedia articles. Using similarity of the articles belonging to different categories allows to calculate the information about categories similarity. The second method is based on average scores given to categories while categorizing documents by our dedicated score-based...
Year 2013
-
Thresholding Strategies for Large Scale Multi-Label Text Classifier
PublicationThis article presents an overview of thresholding methods for labeling objects given a list of candidate classes’ scores. These methods are essential to multi-label classification tasks, especially when there are a lot of classes which are organized in a hierarchy. Presented techniques are evaluated using the state-of-the-art dedicated classifier on medium scale text corpora extracted from Wikipedia. Obtained results show that the...
Year 2012
-
Emotion Recognition Based on Facial Expressions of Gamers
PublicationThis article presents an approach to emotion recognition based on facial expressions of gamers. With application of certain methods crucial features of an analysed face like eyebrows' shape, eyes and mouth width, height were extracted. Afterwards a group of artificial intelligence methods was applied to classify a given feature set as one of the following emotions: happiness, sadness, anger and fear.The approach presented in this...
-
Wielkoskalowa hierarchiczna klasyfikacja dokumentów tekstowych
PublicationNiniejszy rozdział przedstawia problematykę wielkoskalowej, hie-rarchicznej i wieloetykietowej klasykacji dokumentów tekstowych naprzykładzie problemu automatycznego przyporządkowywania artykułuencyklopedycznego do jednej lub kilku (wieloetykietowość) kategorii,spośród setek tysięcy (wielkoskalowość) kategorii tematycznych Wi-kipedii zorganizowanych hierarchicznie. Praca opisuje różne wariantyrozwiązania zagadnienia, analizując...
Year 2011
-
External Validation Measures for Nested Clustering of Text Documents
PublicationAbstract. This article handles the problem of validating the results of nested (as opposed to "flat") clusterings. It shows that standard external validation indices used for partitioning clustering validation, like Rand statistics, Hubert Γ statistic or F-measure are not applicable in nested clustering cases. Additionally to the work, where F-measure was adopted to hierarchical classification as hF-measure, here some methods to...
Year 2010
-
Scenariusze hierarchicznej klasteryzacji wykonywane w środowisku BeesyCluster
PublicationPrzedstawiono szczególnego rodzaju porządkowanie zbioru danychz użyciem tzw. hierarchicznej klasteryzacji. Metoda ta, przy użyciu określonej miary podobieństwa, łączy podobne do siebie dane w grupy tworząc tzw. klastry, które wcześniej nie były dane explicite. Opisano klasteryzację typu skupiającego, typu dzielącego oraz zaproponowano typ mieszany. Przedstawiono koncepcje realizacji algorytmów klasteryzacji poprzez scenariusze...
Year 2009
-
Uniwersalny system RPG do zastosowań w przestrzeniach inteligentnych
PublicationArtykuł dotyczy systemów rozpoznawania poleceń głosowych(RPG). Przedstawiono dwa podstawowe rodzaje systemów RPG i przeprowadzono dyskusję nad wyborem architektury odpowiedniej do zastosowań w przestrzeniach inteligentnych (PI). Zaprezentowano algorytm czasowego dopasowania sygnałów (ang. Dinamic Time Warping - DTW) oraz budowę elementu decyzyjnego zaimplementowanego systemu. Przedstawiono wyniki oceny tego systemu.
seen 1572 times