2019 spring
-
25 Apr
Bogdan Zavalnij (Rényi Intézet): A gráfokkal való modellezés erősségéről
A matematika egyik (tudománytörténeti) vívmánya, hogy a világ bonyolult dolgait tudja a saját eszközeivel hatékonyan modellezni. Előadásunk egy ilyen modellt próbál bemutatni, nevezetesen a gráfokkal való modellezést. Bemutatunk és részletezünk néhány problémát, amely hatékonyan megfogalmazható mint gráfprobléma, és ahol a megoldást a gráf egy maximális vagy k-klikkje szolgáltatja. A bevezető példák után bemutatunk több átfogalmazást, ahol a gráfok színezése vezethető vissza klikk keresésre. Ezen példákhoz hasonlóan bemutatjuk, ahogy a hipergráfok esetén is alkalmazhatóak ezek a technikák, és összevetjük a lehetséges megoldásokat. Bemutatjuk, hogy az elemzett technikák segítségével hogy lehet megoldani bonyolult, nyitott kérdéseket. Annak céljából, hogy bemutassuk a gráf-átfogalmazások valóban sokrétű erősségét meg egy speciális esetet mutatunk be. A gráfszínezések LP átfogalmazásánál ismert módszer van arra, hogy bizonyos szimmetriatöréseket is meg lehessen az LP egyenlőtlenségekben fogalmazni. Megmutatjuk, hogy ezen szimmetriatörések a gráfátfogalmazásokban is lehetségesek. Az így kapott gráfok meglepő tulajdonsággal bírnak: bár nagyobb gráfokról van szó, mégis könnyebb a bennük a k-klikk feladat megoldása.
11 Apr
Kincses Tamás és Veréb Dániel (SZTE ÁOK, Neurológiai Klinika): Funkcionális MRI-vizsgálatok neurológiai kórképekben
21 Marc
István Miklós (Rényi Institute): Computational Complexity of Counting and Sampling
It is a natural task to count mathematical objects with a given property. For example, we might want to count the k-regular graphs of n vertices, the increasing sub-permutations of a given permutation, the sequences of a given length from alphabet {a,b} that do not contain the substring aba, etc. Modern statistics requires the random generations of mathematical objects. For example, random Latin squares are used in experimental designs. Random objects are also frequently generated to estimate the background distribution for hypothesis testing. It is a well-known phenomenon that counting and sampling of mathematical objects usually have the same computational complexity, that is, the same amount of running time of a computer program performing these tasks. The aim of the lecture is to give an overview of the computational complexity of counting and sampling. The lecture will start with the very basics like the Fibonacci series, then will gradually move to advanced topics like perfect matchings in planar graphs or division-free algorithms to compute the determinant to eventually arrive to cutting-edge problems like holographic reductions or #BIS-complete counting problems. The lecture is based on the recently published book with the same title
2018 fall
-
28 Nov
Juhász Sándor (MTA ANET Lab): Brokering the core with periphery - collaboration networks and individual success in the Hungarian film industry
In collaboration-based creative industries, such as film production, connections to central creators can facilitate individual success; while brokering loosely connected communities provide additional advantages. However, there is still limited evidence on how brokerage in core/periphery networks influence nodal outcomes. In this paper, we argue that links to peripheral creators provide additional and complementary benefits for core members, such as access to novel creative ideas or free capacities. Consequently, those creators who broker the core with the periphery of the network enjoy both type of benefits and thus are more likely to achieve success. To verify the argument, we construct a dynamic network of movie creators based on a unique dataset of Hungarian feature films for the 1990-2009 period. We provide evidence that being in the core and brokering the network at the same time induce individual success together. Then, we propose a new way to capture brokers’ role in core/periphery networks and find that those creators who bridge the core with the periphery have an increased likelihood of award winning.
-
26 Sept
Jamolbek Mattiev (University of Primorska): Using constrained exhaustive search vs. greedy heuristic search for classification rule learning
Existing classification rule learning algorithms use mainly greedy heuristic search to find regularities (e.g., a decision tree or a set of rules) in data for classification. In recent years, extensive research was done in the machine learning community on learning rules using exhaustive search – association rule mining. The objective there is to find all rules in data that satisfy the user-specified minimum support and minimum confidence constraints. Although the whole set of rules may not be used directly for accurate classification, effective and efficient classifiers have been built using these, so called, classification association rules. In this seminar the author will present the comparison between a “classical” classification rule learning algorithm that uses greedy heuristic search to produce the final classifier (a set of decision rules) with a class association rule learner that uses constrained exhaustive search to find classification rules on a “real life” dataset. Pro's and con's of each approach will be discussed.
2018 spring
-
20 June
Pablo San Segundo (Universidad Politecnica de Madrid (UPM), Madrid, Spain): Upper bounds for the maximum clique problem
The Maximum Clique Problem is a fundamental NP-hard problem in graph theory which finds numerous real-life applications. In the last decade a number of new upper bounds and implementation techniques have come to light, which have improved the performance of prior exact solvers by orders of magnitude. The relevance of these improvements is shifting the attention of researchers to solve other fundamental combinatorial problems and applications by reducing them to a maximum clique problem. In this talk, some of the cutting edge improvements concerning exact maximum clique search will be discussed. Special attention will be devoted to the relation of these techniques with other combinatorial problems. At the end of the talk we will discuss how different combinatorial problems can combine to solve a constraint-satisfaction problem
-
24 May
Gábor Szederkényi (Pázmány University): Optimization-based computation of biochemical reaction network structures
Kinetic systems form a general class of nonlinear dynamical models that is suitable to describe complex nonlinear phenomena not only in (bio)chemical processes but in other fields as well, where the state variables are constrained to be nonnegative [1]. A dynamical model given in the form of ODEs is kinetic if a weighted directed graph (i.e., a chemical reaction network) can be assigned to it, which realizes the dynamics. Necessary and sufficient conditions for the kinetic property of a polynomial ODE were given in [2] and it is known that the reaction graph corresponding to a given dynamics is non-unique. However, in chemical reaction network theory, strong results have been developed on the relations between network structure and the dynamical properties (stability, existence/uniqueness of equilibria, boundedness of solutions etc.) of the system. Therefore, it is of interest to study whether there exist such network structures/parametrizations for a set of kinetic ODEs that guarantee a desired property. In this lecture, a summary of computational methods mostly based on linear programming will be presented that can be used to study the existence and compute reaction graphs with preferred properties such as minimal/zero deficiency, complex/detailed balance or weak reversibility [3-6]
[1] P Érdi, J Tóth. Mathematical models of chemical reactions: Theory and applications of deterministic and stochastic models. Manchester University Press, Princeton Univ. Press, 1989 [2] Hárs, V.; Tóth, J.: On the inverse problem of reaction kinetics, In: Colloquia Mathematica Societatis János Bolyai, (Szeged, Hungary, 1979) Qualitative Theory of Differential Equations (M. Farkas ed.), 30 North-Holland - János Bolyai Mathematical Society, Budapest, 1981, pp. 363-379. [3] Szederkényi G, Hangos K M. Finding complex balanced and detailed balanced realizations of chemical reaction networks. JOURNAL OF MATHEMATICAL CHEMISTRY 49: pp. 1163-1179. (2011) [4] Johnston M D , Siegel D, Szederkényi G. Computing weakly reversible linearly conjugate chemical reaction networks with minimal deficiency. MATHEMATICAL BIOSCIENCES 241:(1) pp. 88-98. (2013) [5] Lipták G, Szederkényi G, Hangos KM. Computing zero deficiency realizations of kinetic systems. SYSTEMS & CONTROL LETTERS 81: pp. 24-30. (2015) [6] Ács B, Szederkényi G, Tuza Zs, Tuza A A. Computing all possible graph structures describing linearly conjugate realizations of kinetic systems. COMPUTER PHYSICS COMMUNICATIONS 204: pp. 11-20. (2016)
-
10 May
Gábor Danner (University of Szeged): Robust Decentralized Mean Estimation with Limited Communication
Mean estimation, also known as average consensus, is an important computational primitive in decentralized systems. When the average of large vectors has to be computed, like in distributed data mining applications, reducing the communication cost becomes a key design goal. One way of reducing communication cost is to add dynamic stateful encoders and decoders to traditional mean estimation protocols. In this approach, each element of a vector message is encoded in a few bits (often only one bit) and decoded by the recipient node. However, due to this encoding and decoding mechanism, these protocols are much more sensitive to benign failure such as message drop and message delay. Properties such as mass conservation are harder to guarantee. Accordingly, known approaches are formulated under strong assumptions such as reliable communication, atomic non-overlapping transactions or even full synchrony. In this work, we propose a communication efficient algorithm that supports known codecs even if transactions overlap and the nodes are not synchronized. The algorithm is based on push-pull averaging, with novel features to support fault tolerance and compression. As an independent contribution, we also propose a novel codec, called the pivot codec. We demonstrate experimentally that our algorithm improves the performance of existing codecs and the novel pivot codec dominates the competing codecs in the scenarios we studied.
-
26 Apr
Branko Kavsek (University of Primorska; Faculty of Mathematics,Natural Sciences and Information Technologies; Jožef Stefan Institute; Artificial Intelligence Laboratory): Using Words from Daily News Headlines to Predict the Movement of Stock Market Indices
Stock market analysis is one of the biggest areas of interest for text mining. Many researchers proposed different approaches that use text information for predicting the movement of stock market indices. Many of these approaches focus either on maximising the predictive accuracy of the model or on devising alternative methods for model evaluation. On the seminar, we will describe a more descriptive approach focusing on the models themselves, trying to identify the individual words in the text that most affect the movement of stock market indices. Data from two sources will be used: the daily data for the Dow Jones Industrial Average index (“open” and “close” values for each trading day) and the headlines of the most voted 25 news on the Reddit WorldNews Channel for the previous “trading days” (data has been gathered for the past 8 years). By applying machine learning algorithms on these data and analysing individual words we will show that certain words have a "positive" effect on stock indices while other words have a clearly "negative" effect.
-
12 Apr
Claudius Gräbner (Johannes Kepler University, Linz, Austria): Structural spillovers and the emergence of institutions: Agent-based modelling as a means to bridge network science and the humanities
Built upon the suggestion of Bednar & Page (2007), game theory is used as a means for studying culture and institutions. Here, economic agents are modelled as Moore machines that need to play different games sequentially. Thus, there is an incentive to economise cognitive effort and to develop strategies that work not only in one, but in several games. This is consistent with numerous empirical studies in decision sciences indicating that human beings develop behavioural receipts and heuristics which they apply in numerous contexts. Based on the basic idea of modelling culture and institutions using multiple games, the related literature is extended into a network dimension: while there is some work on how agents use behavioural receipts in one context as a vantage point for their behaviour in other contexts (behavioural spillovers), the impact of social ties formed in one context to the behaviour in another context (structural spillovers) is less understood. Here an agent-based model that addresses this challenge is introduced: how the social ties developed in one interaction context can affect the performance of players in another interaction context. The impact of various fixed interaction structures on the performance of the agents in various games is tested, just as the effect of dynamic networks where agents choose their interaction partners endogenously. Such models can help to introduce results from network sciences to social sciences, and serve as a means to facilitate interdisciplinary research. On the other hand, there are also potential pitfalls, such as the heavy reliance on numerical solution techniques and the enormous flexibility in specifying such models.
-
22 Marc
Kovács Gyula (Fókusz Takarékszövetkezet, CEO Andego Tanácsadó Kft): 1+1=2?
Az adatbányászati elemzések, és ezen belül a prediktív modellek, igen széles körben terjedtek el az üzleti szektorban. A vállalati adatvagyon gazdálkodás mára valóság, egyre több cég hoz meg ún. adatvezérelt döntéseket. Az utóbbi években a klasszikus adatbányászati elemzések mellett egyre több új elemzési metódus és technológia jelent meg az üzleti világban, így pl. a szövegbányászat, hanganyagok feldolgozása és elemzése, illetve maga a hálózat elemzés. Ez utóbbit az tette lehetővé, hogy egyre több olyan adatbázis épült fel, melyek valójában kapcsolati hálókat írnak le. Az előadás a következő témákat dolgozza fel: 1.) Milyen hálózatok építhetők fel az üzleti életben? 2.) Milyen üzleti problémák megoldásában nyújthatnak segítséget a hálózatelemzések? 3.) 1+1 = 2? A hálózati elemzés miért ad más eredményt, mint a hagyományos adatbányászat? Melyiket alkalmazzuk? Egyáltalán kell választanunk? 4.) Egy esettanulmány, ahol kiderül, hogy 1 + 1 > 2
-
8 Marc
Dombi József (Professzor emeritus, SZTE): Approximation, neural networks, clustering, continuous valued logic
Among the tools of artificial intelligence we find mostly heuristic procedures. Their goodness is difficult to control and a global optimum is not guaranteed. The connection of different algorithms is not solved either. A continuous logic provides a common ground for different algorithms. With the help of it we can also give semantic meaning to the solution; clustering, neural network, approximation, etc. can be handled as well. During the lecture I would like to draw attention to some possible solutions.
2017 fall
-
14 Dec
Lengyel Balázs (MTA KRKT, Agglomeráció és Társadalmi Kapcsolathálózatok Lendület kutatócsoport vezető): Spatial diffusion and churn over the life-cycle of innovation: the role of social networks
Innovative ideas, products or services spread on social networks that, in the digital age, are maintained via telecommunication tools such as emails or social media. One of the last standing puzzles in social contagion is the role of physical space and it is not fully understood how products disappear from the map at the end of their life-cycle. In this paper, we utilize a unique dataset compiled from a Hungarian on-line social network (OSN) to uncover novel features in the spatial adoption and churn of digital technologies. The studied OSN was established in early 2000s and failed in international competition a decade later. A Bass Diffusion Model describes the process how the product gets adopted in the overall population. However, it does not cope with the prediction of spatial diffusion. The novel ingredients missing from the model are: the assortativity of adoption time, urban scaling of adoption over the product life-cycle and a distance decay function of diffusion probability. We find that early adopter towns also churn early; while individuals tend to follow the churn of nearby friends and are less influenced by the churn of distant contacts.
-
16 Nov
Gyimesi Péter (SZTE)
A szoftverhibák számos okból kifolyólag elkerülhetetlenek egy rendszer fejlesztése során: szűk határidők, pontatlan specifikáció, programozói figyelmetlenség, stb. Ezeknek a hibáknak a megtalálása és kijavítása erőforrás igényes feladat. Számos kutatás foglalkozik a szoftverhibák felderítésével és különböző megközelítéseket alkalmaznak. Egy azonban közös bennük: a módszereket valahogyan tesztelni, azok eredményességét mérni kell. Ebből a célból közzétettek nyilvános hiba-adatbázisokat, melyek benchmark-ként szolgálnak az ilyen jellegű kutatások számára. Az előadás során az ilyen hiba-adatbázisok előállítására mutatunk egy hatékony módszert, mely egy gráf adatbázist (Neo4j) használ.
-
9 Nov
Ivan Luković and Vladimir Ivančević (University of Novi Sad)
Ivan Luković: Formal Education in Data Science – A Perspective of Serbia and Faculty of Technical Sciences
In the last years, Data Science becomes an emerging education and research discipline all over the world. Software industry shows an increasing and even quite intensive interest for academic education in this area. The similar trend has been noticed in Serbia, particularly in Belgrade and Novi Sad. In this talk, we discuss main motivating factors for creating a new study program in data science at Faculty of Technical Sciences of University of Novi Sad. Also, we present a short survey of software industry needs for data science related experts, and discuss how we structured the new study program and addressed the main issues that come from more than evident industry requirements. The program was accredited in year 2015, both at the level of bachelor and master level studies, and this school year is its first execution, from which we expect the new experiences.
Vladimir Ivančević: An Overview of Selected Research Studies in Data Analytics and Data Science at the Faculty of Technical Sciences in Novi Sad, Serbia
Over the past several years, the Faculty of Technical Sciences in Novi Sad, Serbia, has been a home to an increasing number of research studies concerned with extraction of potentially valuable information from diverse data sets. Many of these research efforts were concentrated at the Chair of Applied Computer Science during a period in which data science started emerging as one of the most popular and promising new fields associated with computer science and informatics. The two most notable areas covered by these studies are education and medicine, both of which play an important role in the contemporary society and generate ample quantities of data. The education-focused studies dealt with a wide set of topics: a) exploring patterns in spatial distribution of students in a classroom, b) analysing student grades with respect to different factors, c) constructing programming tests automatically, and d) identifying position of engineering and technology education within the system of higher education in Serbia. The medicine-related studies centered upon epidemiology-oriented topics: a) creating a software system to support business intelligence in epidemiology and b) analysing collected data about early childhood caries to identify risk factors and create predictive models. Studies from both areas have provided some interesting findings and also managed to spur ideas for new research directions in theory and practice of data analytics and data science.
-
26 Oct
Sándor Szabó (University of Pécs): Estimating clique number in high performance computing environment
Suppose A is an algorithm to locate an upper estimate for the clique number of a given graph G. Using algorithm A one can construct a new A' algorithm that provides an improved estimate of the size of the maximum cliques in G. The fact that such algorithm A' exists does not come as a surprise to us. The point is that the construction is practical and the new algorithm A' well suited for various high performance computing environments. We carried out a large scale numerical experiment to test our proposal.
-
28 Sept
Kristóf Kovács (BME): Facility location on networks, with hard to compute objective functions
This talk will be about facility location problems on networks with objective functions that are especially hard to compute. Calculating any point of these functions requires the solution of an NP-hard problem. Due to this property general global optimizer algorithms are inefficient to solve these problems. I will introduce the general Stackelberg problem, where two or more firms compete for demand by locating facilities one after the other. The choice of location for one of the firms influences the choice of the other firms, as both wants to maximize its profit after the facilities are built. Another hard to solve problem I will talk about is the 1-median problem with demand surplus, where one facility has to be located such that only a given percent of the demand has to be covered. Finally I will present the solution and the computational results to a Stackelberg problem, as well as a modified 1-median problem, both of which have hard to compute objective functions.
2017 spring
-
24 May
Nagy Gyula (SZTE): Tudománymetriai és tartalmi elemzések szövegbányászati módszerekkel
Abstract: tudományos élet szereplői számára az olyan nemzetközi hivatkozási adatbázisok használata, mint a Web of Science vagy a Scopus egyre inkább ismertek és megkerülhetetlenek. Mindemellett azonban teljes tudományterületek léteznek hazánkban, amelyek ezen adatbázisokban csak kevéssé reprezentálódnak [Scival, Hungary 2011-2015: Social Sciences 4,4%; Arts and Humanities 2,7%; Other 7,7%]. A jelenség több okra vezethető vissza: egyrészt a "soft science" tudományágak publikálási szokásai jelentősen eltérnek a "hard science" diszciplínák publikálási gyakorlatától (ti. tanulmánykötet, monográfia, stb.), másrészt ezen területek hazai kutatói előszeretettel publikálnak olyan folyóiratokban, amelyeket a nemzetközi hivatkozási adatbázisok nem indexelnek. Azonban részükről ugyanolyan elemi igény mutatkozna saját tudományterületük belső szerkezetének felderítésére. Mindezt a kívánságot a tudományos együttműködés hálózatainak felderítése által és egy-egy folyóirat, vagy akár egy egész diszciplína hivatkozási gráfjának megalkotásával tudjuk teljesíteni. A fenti vizsgálódás elvégzéséhez - a bevett tudománymetriai elemzési módszerek mellett - egyaránt érdemes segítségül hívnunk a szövegbányászat és a hálózattudomány eszközeit. Az előadásban egy pilotprojekt tapasztalatai kerülnek bemutatásra, melynek keretében kísérletet tettünk két magyar nyelvű, társadalomtudományi folyóirat teljes tudománymetriai elemzésére. Kutatásunkban kiemelt figyelmet szenteltünk a társszerzőségi gráfok előállításának, illetve a folyóiratok teljes hivatkozási hálózatának megalkotására, mely a magas elemszám miatt csak a hivatkozások automatikus extrakciója által volt megvalósítható. A tudománymetriai elemzések mellett az egyik tárgyalt folyóirat esetében egy összetett tartalmi elemzésre is kísérletet tettünk a szövegbányászat megoldásain keresztül.
-
27 Apr
István Hegedűs (University of Szeged): Gossip-Based Machine Learning and Matrix Decomposition
Abstract: We are talking about distributed learning, when we wish to build models or compute aggregations on data sets that are stored on a large set of computers. Standard methods for processing distributed data first collect them in data centers and run centralized algorithms. But models can be built on the data sets without centralized algorithms as well. In my presentation, I will talk about the distributed data aggregation and modeling, introduce the basics of machine learning and classification. Finally, I will present a gossip-based solution of the distributed learning and an algorithm that can be used for low-rank matrix decomposition.
-
6 Apr
Péter Erdős (Rényi Institute): Degree Sequences' Realizations
Abstract: This talk can be considered as a prequel to István Miklós' talk on March 23, 2017, but it is completely self contained. It discusses realization and listing problems of different degree sequence type questions. These are fundamental tools for studying synthetic graphs which simulate real life networks.
-
23 Marc
István Miklós (Rényi Institute): Exact sampling of graphs with prescribed degree correlations
Abstract: Many real-world networks exhibit correlations between the node degrees. For instance, in social networks nodes tend to connect to nodes of similar degree. Conversely, in biological and technological networks, high-degree nodes tend to be linked with low-degree nodes. Degree correlations also affect the dynamics of processes supported by a network structure, such as the spread of opinions or epidemics. The proper modelling of these systems, i.e., without uncontrolled biases, requires the sampling of networks with a specified set of constraints. We present a solution to the sampling problem when the constraints imposed are the degree correlations. Furthermore we would like to give a brief survey of our current knowledge of generating such random networks.
-
9 Marc
Federico Musciotto(University of Palermo and CEU): Timely evolution and core of communities of statistically validated projections of bipartite networks
Abstract: Many complex systems are naturally organized in bipartite networks, i.e. networks with two disjoint sets of node, usually characterized by high heterogeneity. Although useful information can be extracted from these networks, their projections are not always stable and robust against errors in the data or other sources of noise. A solution to this issue is based on statistically validated networks, which reduce the original system by considering only the links which are statistically significant. In this way, cores of the real communities of the system are obtained with a high level of precision. Moreover, the methodology of SVN can be extended in order to track the evolution in time of the communities of a system. In this talk I will show the results of these methods on a dataset of stock market Finnish investors, characterized by high heterogeneity with respect to trading activity. This work is the result of a collaboration with Luca Marotta, Jyrki Piilo and Rosario N. Mantegna.
2016 fall
-
20 Dec
András Bóta (The University of New South Wales, Australia): Learning infection processes
Abstract: Infection models can be used to model the spread of disease, information, behaviour and many other things through a network composed of connected nodes. One of the common challenges arising in the application of infection models is the lack of available transmission probabilities. The task of inverse infection is the systematic estimation of these values. Several methods have been proposed recently for solving this task. A common property of these approaches is that they make many specialized assumptions about the problem they are trying to solve. In contrast the method we discuss in this talk gives a general framework for inverse infection tasks and offers much needed flexibility allowing the method to be used with a variety of real-life applications. In the second part of the talk we are going to discuss a specific application of the generalized inverse infection model. Modern transportation infrastructure allows quick and efficient travel between distant parts of the world, but unfortunately also offers many ways to transfer diseases between regions. Recent examples of global outbreaks include the MERS, SARS and swine flu epidemics and most recently the 2015 Zika fever outbreak of Brazil. A critical component of outbreak control is the identification of disease spreading from region to region. We will show how the proposed inverse infection model can be used to 1. accurately model the Zika virus outbreak on the country level, 2. estimate travel risk between regions.
-
17 Nov
Brys Zoltán (LAM, BME-TMIT): Hálózatkutatás a népegészségügy és a gyógyszeripari marketing területén. Esetismertetés és áttekintés. A gyógyszerfelírások, a pattanások és fejfájás valóban "szociálisan fertőznek"?
Kivonat: A gráfelmélet a fertőző betegségek terjedésének modellezésében régóta használt módszer és fontos eredményeket tárt fel (Epstein, Balcan, Lilrejos stb.). A fehérje-fehérje interakciós hálózat elemzése pedig a bioscience szerves részévé vált (genetika, epigenetika, gyógyszerfejlesztés stb.) A képalkotó eljárások és az optogenetika terjedésével a neurológiában is egyre inkább használt a gráfelmélet. A népegészségügy területén nagy vitát váltott ki Christakis és Fowler 2007-es felvetése, amely szerint a noncommunicable diseases egyik jelentős okaként számon tartott elhízás és dohányzás "szociálisan fertőz". A felvetés jelentős kritikát kapott, főként statisztikai (Lyons) és szociálpszichológiai/szociológiai (Buda) területen. Cohen-Cole és mtsai (Yale) szellemes közleményében hasonló módszertannal bizonyította, hogy Christakisék logikájával hálózatos hatások érvényesülnek a pattanások és fejfájások terjedésében is. A hálózatelemzés ígérete, hogy egyfajta "theory of everything", mindent magyarázó "csodamódszer" lesz nem látszik beigazolódni. Sajnos az egyes hálózatelemzéssel dolgozó sztárszerzők a "csodamódszer" birtokában könnyen és észrevétlenül (és biztosan nem tudatosan) negligálták a társadalomtudományokban és a társtudományokban felhalmozott tudásvagyont (pl. szociológia, addiktológia stb.). Az előadó egy valós gyógyszeripari Clinical Opion Leader detektálási eset ismertetése kapcsán próbálja bemutatni, hogy a népegészségügyhöz hasonló folyamat látszik a gyakorlati tudományok területén, különösen a marketing az adatelemzés területére. Az előadó személyes meglátása szerint a hálózatelemzés hype-fázisa lassan lejár és a módszer valódi értéke lassan kibontakozni, ami jelentősen kisebb, mint az ígéret volt, de így is rendkívül jelentős.
-
3 Nov
Ivan Fekete(Semmelweis University and Link-group): Computational prediction personalized of drug combinations
Abstract: Numerous methods and biological networks have been used to model intracellular signal transduction, but so far the clinical applicability of these approaches remained rather limited. Here we describe a novel system (Turbine, http://turbine.hu) for the reliable simulation of intracellular signaling which required, first, to utilize a large, fully dynamically reviewed, manually curated network of major human signaling pathways and their transcriptional regulatory mechanisms, and second, to run ensembles of simulations and extract the resulting attractors - steady states - of the system. The software finds the attractors of the signaling network, and correlates the attractors’ activity patterns with the activity of biological processes, like apoptosis or proliferation. By combining a large set of steady states, we were able to map the cellular attractor landscapes, which varied depending on the presence of different physiological ligands, available membrane receptors, or - most importantly regarding the clinical applicability - mutated proteins. This approach, combined with additional omics data layers (such as cancer genomic and transcriptomic profiles) and artificial intelligence systems made Turbine able to predict potential mono- or combination drug therapies on a personalized basis.
-
20 Oct
Laszlo Kovacs (University of West Hungary, Department of Applied Lingusitics): Linguistic networks
Abstract: Network structures are not new in linguistics: since the 1960’s networks are used to explain linguistic phenomenon. In the first part of the lecture a broad picture of linguistic network research will be given, showing where networked structures in language exist. The main focus of the lecture is on the networked structures of the mind: it will be shown on the example of the Hungarian database "ConnectYourMind" which structures in the mental lexicon ("dictionary of mind") exist. The presented results are joint works done with Andras Bota, Laszlo Hajdu and Miklos Kresz (University of Szeged) and with Peter Pollner and Katalin Orosz (Eotvos University).
-
6 Oct
Gyorgy Turan (University of Szeged and University of Illinois at Chicago): Betweenness centrality
Abstract: The betweenness centrality of a vertex in a network measures the number of shortest paths containing the vertex. The local version considers only shortest paths of bounded length. We review related results, and then discuss the behavior of the local version for trees, including worst-case and scale-free random trees. Joint work with Ben Fish and Rahul Kushwaha.
2016 spring
-
12 May
Nandor Poka, (research assistant (Department of Applied Informatics) and doctoral candidate (Doctoral School of Multidisciplinary Medical Science) University of Szeged): Combinatorial Scientific Computing and (Computational) Biology: When the demand meets the offer
Abstract: With the rise of the BigData and High-Performance Computing era, came the flourishing of many fields of science, including Combinatorial Scientific Computing. Both the business and scientific community produces such vast amounts of data, that new technologies and algorithms are required to analyze them. To efficiently utilize highly parallel computers or clusters, tasks must be decomposed and the data must be partitioned, and these involve graph algorithms themselves. Combinatorial Scientific Computing methods have been used for eg. in the aforementioned load distribution, automatic differentiation, statistical physics, and other enigmatic areas. However more “life-like” applications of these methods were and are put to use in other fields of science like modern biology. In our talk we will present various biological problems (both “old” and more recent) that can be fairly easily represented with mathematical objects such as graphs and matrices. Our main focus of these problems will be the present and future challenges of Next-Generation Sequencing, and how to tackle them using Data Science and Combinatorial Scientific Computing.
-
28 April
Miklos Kresz (Department of Applied Informatics, University of Szeged; joint work with Andras Bota and Andras Pluhar): Dynamic network mining and business intelligence
Abstract: During the last decade social network analysis and mining became a key research area. Apart from the obvious applications in on-line social network services, the field plays central role in many classical business intelligence tasks such as customer attrition, risk analysis and campaign management. In order to capture the characteristics of the above problems, the dynamics of the corresponding network processes and that of the changes in the network structure needs to be studied. In this talk we will consider two relevant problems. Dynamic community detection is an algorithmic tool for the analysis of the lifetime of communities in real graphs. The study of infection processes in networks pose several algorithmic and modelling questions such as maximizing the spread of influence or approximating the real infection values. In addition to review applied models and methods, in the talk real-life applications will be also presented.
-
7 April
Bogdan Zavalnij(University of Pecs): Mathematical modeling of various problems by graphs and solving them with clique search
Abstract: In our talk we would like to demonstrate the modeling powers of graphs. We will demonstrate problems from different fields and show that they can be reformulated as graphs. The solution of these problems will be transformed to usual graph problems as different coloring and clique problems. The problems included in the talk will be from simple games, scheduling, stock exchange, drug design and even graph problems themselves -- including hypergraph coloring problems. We would like to show some useful techniques and also some common methods for such reformulations that could prove useful.
-
24 March
Christian Bongirono (University of Palermo): The dual-projection approach for bipartite networks using statistical link validation
Abstract: Bipartite complex networks are usually analyzed by projecting the two disjoint set of nodes into two networks and then using the standard techniques for them separately. Since complex systems are often very heterogeneous that makes very difficult to distinguish links of the projected network that are just reflecting system’s heterogeneity from links relevant to unveil the properties of the system. To avoid this problem it has been developed a methodology for one-mode projections of bipartite networks using an unsupervised statistical link validation. In order to study the efficiency of the method we investigate the community structure of the projected network using both a simple projection and the statistically validated projection on various synthetic benchmarks and real networks. In all these cases the link validating filtering procedure necessarily increases the precision and suggested to use, even if considering the drawback that it decreases the level of accuracy in certain situations.
-
10 March
Christian Bongirono (University of Palermo): Statistical Regularity in the Air Traffic flow
Abstract: The aircraft trajectories are compounded by a sequence of spatial fixed points (NVP) that typically diverge from the best path route. This infrastructure allows the air traffic controllers to direct the air traffic flow on standard air ways, and focuses their attention to a few numbers of special NVPs where the routes converge. As a drawback the not optimal routes force the air traffic controller to modify, where is possible, the routes to enhance the air traffic flow. The aim of the first part of the talk is to highlight the behaviour of the air traffic controller respect to a network optimization operation named direct by the observation of stylized facts both at the global trajectory level and at the local navigation point level. In the second part of the talk will be discussed how an Agent Based Model could help us in understanding the transition from the current NVP based network to the future new SESAR scenario, where the aircraft will be allowed to follow a free-route path.
2015 fall
-
26 November
Gabor Berend (Institute of Informatics): Learning the transition matrix for weighted PageRank
Abstract: The PageRank algorithm is a widely known and well understood approach to assign importance scores to nodes in networks. This highly applicable approach relies on the assumption that all the connections of a node has equal importance (i.e. inversely proportional to their out-degree). The talk will introduce approaches that aim at determining the strength of the connections between pairs of nodes given that the PageRank scores (or some relative importance scores) of the nodes within the graph is assumed to be known in advance. After presenting these models, both quantitative and qualitative results will be presented on synthetic and real world networks (e.g. citation and collaboration networks, the English and Hungarian Wikipedia and networks generated from language usage).
-
19 November
Gabor London (MTA-SZTE Stereochemistry Research Group): Accomplishments and challenges of networks science in chemistry
Abstract: Approaching chemical problems with the tools of network science has proved very fruitful in recent years. The network approach has provided us with new ways of designing drugs, optimizing multistep chemical syntheses or fighting terrorism. Most of these successes, however, are based on the analyses of available data. One of the big challenges chemists are facing whether it is possible to implement this data-based knowledge for the creation of instructable molecular networks that can be models for early (molecular) evolution or able to perform complex (synthetic) tasks. The talk will discuss both the data-based achievements and the recent attempts towards creating “molecular ecosystems”.
-
5 November
Istvan Kiss (SZTE Knowledge Management Research Center): Communities and central nodes in the network of the mobile inventors in the United States
Abstract: IP-intensive industries accounted for about 33-40% of the gross domestic product in western economies. These intellectual properties are mostly embodied in patents and trademarks. Highest stake of the expenditures in the mentioned sector goes for the wages of white-collar workers what underlines the importance of the intellectual capital and knowledge in the creation and development of IP portfolio. We investigated the flow of knowledge as a crucial resource among organizations by analysis of patent documents from the United States. In our network organizations are the nodes and the mobility of researchers among them are the edges. This graph can be considered as the informal innovation network of US organizations, where firms, universities or governmental institutions competing for knowledge and recombine their innovation capacity through the mobility of inventors.
-
22 October
Tamas Vinko (Institute of Informatics): Network models for BitTorrent communities
Abstract: BItTorrent communities are content sharing systems using peer-to-peer technology. From mathematical point of view these communities can be modeled with graphs of particular structure. For example, there is a straightforward bipartite representation, where we take the users and the files as vertices and the edges represent supply and demand. Using this simple representation one can already consider interesting optimization problems. Moreover, there is a richer graph representation which leads to a flow network. Hence, one can start thinking about applying traditional flow algorithms and their meaning in this particular context.
-
8 October
Andras London (Institute of Informatics): Statistically validated projections of bipartite networks
Abstract: Bipartite networks naturally appear in from social to biological systems, examples include, among many others, the actors--movies network, artists--music network, scientists--research papers cooperation network, network of sexual contacts, diseases--genes network, plants--pollinators mutualistic networks, banks--firms money transfer networks and words co-occurrence networks. Many properties of these networks typically investigated by constructing and analyzing a projected network on one of the two sets of the original network. When one constructs a projected network of nodes only from one set, the original network's heterogeneity (e.g. the heterogeneous degree distribution) makes difficult to determine those links that presence in the projected network cannot be explained by random co-occurrence of their neighbors in the original network. In this talk we present a method based on statistical validation to overcome this problem and show some possible applications on real bipartite systems.