Papers by Daniel VOLOVICI
Qualitative and Quantitative Methods in Libraries, Mar 5, 2019
Most of the literatures that deal with library security take into account the security of collect... more Most of the literatures that deal with library security take into account the security of collections, buildings, or equipment. The security of the libraries is an important concern for library staff. There are security systems for collections, but in the context of terrorism, we are interested that people who have access inside the library to be in good faith. Checking the library card is the key to enter the library. It is nontransferable. We present a complementary system of the entrance permit through the facial recognition of users, developed and tested in a public library as collaboration between two Romanian universities:
International Conference on Artificial Intelligence, Feb 21, 2009
... Authors: Maria Rodica Volovici, University Library, Lucian Blaga University of Sibiu, România... more ... Authors: Maria Rodica Volovici, University Library, Lucian Blaga University of Sibiu, România.Daniel Volovici, University Library, Lucian Blaga University of Sibiu, România. Published in: ... Downloads (12 Months), 0. View colleagues of Maria Rodica Volovici. Daniel Volovici ...
Studies in Informatics and Control - ICI Bucharest, 2012
IFAC Proceedings Volumes, Oct 1, 1995
TIle most popular methods for modifying feed forward and recurrent neural networks' weights, acco... more TIle most popular methods for modifying feed forward and recurrent neural networks' weights, accordingly the backpropagation methods, actually are gradient methods. Due to the fact that the gradient is a local measure, the step towards the gradient's direction, which step is given by the learning rate p, must be infinitesimal, which implies the choosing of a very small p. But this leads to a very slow convergence. Hence, still a large p is chosen. On the other hand, a too large p leads to strong oscillations of the aim function. Moreover, the values of p are relative to the problem to be solved.
Studies in Informatics and Control, Jun 15, 2012
The impact of the social networks nowadays is impossible to ignore. Sales companies turn towards ... more The impact of the social networks nowadays is impossible to ignore. Sales companies turn towards them as they acknowledge the opportunities provided by the socio-emotional value of the information that spreads quickly within the communities of these networks. Facebook, at more than 600 million users, is the largest of these, to date. We made a study on this very network, in order to observe the way an application can spread among its users for a week, without using any paid publicity. The use of Facebook's framework for the promotion of digital libraries can constitute an opportunity and a challenge at the same time. In this paper, we propose an architecture to enable the promotion of digital libraries within the social network environment. The results prove that such an approach can have a significant impact in the online community.
Studies in Informatics and Control, Jun 15, 2010
The principal aim of this paper is to make a review of main statistical methods for classifying d... more The principal aim of this paper is to make a review of main statistical methods for classifying documents that could be easily adapted in the context of Web document retrieval. After presenting the most popular methods of classification we will also define the most accurate indicators for assessment of classifiers performance. Thus we will refer to the recall, precision, fscore, sensitivity and specificity. We will also describe how these indicators can be calculated in the context of Web documents.
International Journal of Advanced Statistics and IT&C for Economics and Life Sciences, Jun 1, 2019
Document clustering is a problem of automatically grouping similar document into categories based... more Document clustering is a problem of automatically grouping similar document into categories based on some similarity metrics. Almost all available data, usually on the web, are unclassified so we need powerful clustering algorithms that work with these types of data. All common search engines return a list of pages relevant to the user query. This list needs to be generated fast and as correct as possible. For this type of problems, because the web pages are unclassified, we need powerful clustering algorithms. In this paper we present a clustering algorithm called DBSCAN-Density-Based Spatial Clustering of Applications with Noise-and its limitations on documents (or web pages) clustering. Documents are represented using the "bag-of-words" representation (word occurrence frequency). For this type o representation usually a lot of algorithms fail. In this paper we use Information Gain as feature selection method and evaluate the DBSCAN algorithm by its capacity to integrate in the clusters all the samples from the dataset.
International Journal of Advanced Statistics and IT&C for Economics and Life Sciences, Dec 1, 2021
This paper presents a design method and tool developed to support the skill forming activities in... more This paper presents a design method and tool developed to support the skill forming activities in the DigiFoF network (https://www.digifof.eu/). The focus is on training of manufacturing system design skills both as HEI education and vocational training, but preliminary design of new manufacturing systems is also supported (e.g in the development of small business process scenarios). We proposed a model-based methodology for solving of the manufacturing system design problems The methodology and the supporting tool are centred around a less abstract Domain-Specific Modelling Language (DSML). The language is easy to learn due to its few components. A modelling and simulation environment named Digital Production Planner Tool (DPPT) was generated from the metamodel of the DSML. The degree of abstraction used by this tool corresponds well to the intended use in training and preliminary design. Our method incorporates by design the possibility to impose constraints at the modelling language level to limit the modelling space to feasible/possible solutions. The resulting tool enforces these constraints in the use and supports the development of feasible designs even by inexperienced designers. The access to the conceptual model allows the translation of the model to other modelling language like Petri net. This extends the support for the design methodology. The whitepaper presents a use case for the developed method and tool: the design of a chocolate manufacturing line.
ABSTRACT The main weakness of fractal image compression is its long encoding time needed to searc... more ABSTRACT The main weakness of fractal image compression is its long encoding time needed to search the entire domain pool to find the best domain-range mapping. To solve the problem, some solutions were proposed but most of them do not employ neural networks (only the use of Kohonen SOM for clustering was reported). The paper proposes a new method based on Karhunen-Loeve transform (PCA networks), which attempts to use neural networks' well-known adaptability in order to find a good feature vector for a block. Performance regarding network generality, quantization of the transform coefficients, comparison with DCT and kd-tree search, was explored. Results prove that the proposed method slightly outperforms state-of-the-art methods
Studies in Informatics and Control - ICI Bucharest, 2012
Domain-Specific Conceptual Modeling, 2022
In terms of class organization, an important factor for creating the conditions for the teaching ... more In terms of class organization, an important factor for creating the conditions for the teaching process is the venue at which the teaching takes place. School field trips represent a special type of teaching, which involves visits to museums, libraries and other cultural heritage institutions. The aim of this research, conducted in a Croatian elementary school, is to determine how well the pupils remember their visits to these institutions and how motivated they are for such trips. Taking into consideration that the research involved 41 participants who visited a total of 31 museums, libraries and other institutions during the course of 8 years, the total number of possible answers in the survey was 1271 and the participants only provided 239 answers, which means that they remembered only 18.8 percent of institutions visited. The survey also probed the students to see how they prepared for these visits during their school excursions and field trips. Out of 41 participants, 16 reported that their homeroom or class teachers had acquainted them with the institution they would visit; only 4 students obtained the information on their own from online or some alternative sources; and 21 students, more than a half, did not try to obtain any information about the institution they would visit. A way to change this lies in appropriate intellectual and emotional motivation of students, and one of the possible concrete measures is the project "A Backpack Full of Culture", conducted by the Ministry of Culture of the Republic of Croatia.
Abstract. We use simulated soccer to study multi-agent learning. Each team member tries to learn ... more Abstract. We use simulated soccer to study multi-agent learning. Each team member tries to learn from the corresponding human player in a real game. Following a unified approach, strategic and tactical behavior is learned synergistically by training a feed-forward neural network (ANN) with a modified back-propagation algorithm. It aims at decreasing the learning time and avoiding the local maximums. We tried to minimize the computation effort, as required in classic back-propagation (BKP) methods. 1
Machine Learning is the most important part of Articial Intelligence in the same sense as we cann... more Machine Learning is the most important part of Articial Intelligence in the same sense as we cannot speak about intelligence without the capacity of learning. One of the basics type of learning is to learn to classify objects or putting labels on objects. If you are able to recognize that an object have the attributes of a class C or not (meaning that it is part of class non C), than you will be able to classify in more than one classes: with the strategy one-vs-all or with the strategy one-vs-one. Classication as a learning task imply training with examples of objects a priori labeled with the class which they belong. But if in data we do not have denitions of classes, splitting data into groups has the name of clustering. The idea behind clustering is that probably the data are produced by different processes or that they belong naturally to diferent groups. So, the best way to evaluate the quality of the clustering is to try to cluster data generated to be part of dierent classes...
Identifying Parts of Speech (PoS) represents the process by which grammar tags containing their c... more Identifying Parts of Speech (PoS) represents the process by which grammar tags containing their corresponding PoS are attached automatic to every word within a sentence. Since no word acts as just one single PoS—their syntactic value depending on the context they are used in—identifying parts of speech is not a trivial matter. In this paper we have taken into account two tagging methods, based on Naive Bayes’ classifier probabilities and the occurring context of the word for which the PoS must be identified. We have called these methods Backward Naive Bayes and Forward Naive Bayes. For Romanian language, we have taken into account seven different PoS as: noun, verb, adjective, adverb, article, preposition plus the „and others” category. From conducted experiments, we have observed that identifying the PoS for a word based on the PoS for the previous word produces better results in all respects. We have studied each PoS separately and have concluded that there also are more easily id...
Machine Learning is the most important part of Artificial Intelligence in the same sense as we ca... more Machine Learning is the most important part of Artificial Intelligence in the same sense as we cannot speak about intelligence without the capacity of learning. One of the basics type of learning is to learn to classify objects or putting labels on objects. If you are able to recognize that an object have the attributes of a class C or not (meaning that it is part of class non C), than you will be able to classify in more than one classes: with the strategy one-vs-all or with the strategy one-vs-one. Classification as a learning task imply training with examples of objects a priori labeled with the class which they belong. But if in data we do not have definitions of classes, splitting data into groups has the name of clustering. The idea behind clustering is that probably the data are produced by different processes or that they belong naturally to different groups. So, the best way to evaluate the quality of the clustering is to try to cluster data generated to be part of differen...
Abstract: In many medical applications, computer vision is required to step behind the image prep... more Abstract: In many medical applications, computer vision is required to step behind the image preprocessing level and perform recognition tasks without human interaction. We focused our work on blood tests for leucocyte detection and counting. This test is generally performed manually and required for infection diagnosis. We developed a method for cells classification and counting using images from a laboratory microscope. This is a step forward for the automation of laboratory tests, producing good results without supervision and interaction. A camera calibration is required for the precise determination of the leucocyte formula.
Digital documents as the real ones have to be classified and indexed in a library for proper futu... more Digital documents as the real ones have to be classified and indexed in a library for proper future exploitation. Classification and indexation process is a hard one for librarians all over the world. A software system can ease their work and make the process more accurate. We present in our paper methods for classifying and indexing publications, suitable for such a system and analyze different storage and index database management systems capabilities in order to use them as support for classification, indexation and retrieval processes in an integrated software system for libraries. Furthermore, the problem of storing and retrieval of full content of a publication is taken into consideration.
... Authors: Maria Rodica Volovici, University Library, Lucian Blaga University of Sibiu, România... more ... Authors: Maria Rodica Volovici, University Library, Lucian Blaga University of Sibiu, România.Daniel Volovici, University Library, Lucian Blaga University of Sibiu, România. Published in: ... Downloads (12 Months), 0. View colleagues of Maria Rodica Volovici. Daniel Volovici ...
Classical image compression methods are based on measuring the error only at entire image level. ... more Classical image compression methods are based on measuring the error only at entire image level. In some areas there is an obvious need for getting an upper bound for the error at the pixel level. In the paper we propose such a near-lossess method based on LZW dictionary algorithm. The modifications needed to adapt LZW to become a near-lossless method are presented. As far as we know this is the first attempt to use LZW as a near-lossless method. Experimental results done and presented in the paper prove that the method gives better that the one based on quadtree partitioning so the proposed method is promising.
Uploads
Papers by Daniel VOLOVICI