Ljupco Todorovski

Jožef Stefan Institute, Ljubljana, Ljubljana, Slovenia

Are you Ljupco Todorovski?

Claim your profile

Publications (47)19.73 Total impact

  • [show abstract] [hide abstract]
    ABSTRACT: Modeling dynamical systems involves two subtasks: structure identification and parameter estimation. ProBMoT is a tool for automated modeling of dynamical systems that addresses both tasks simultaneously. It takes into account domain knowledge formalized as templates for components of the process-based models: entities and processes. Taking a conceptual model of the system, the library of domain knowledge, and measurements of a particular dynamical system, it identifies both the structure and numerical parameters of the appropriate process-based model. ProBMoT has two main components corresponding to the two subtasks of modeling. The first component is concerned with generating candidate model structures that adhere to the conceptual model specified as input. The second subsystem uses the measured data to find suitable values for the constant parameters of a given model by using parameter estimation methods. ProBMoT uses model error to rank model structures and select the one that fits measured data best. In this paper, we investigate the influence of the selection of the parameter estimation methods on the structure identification. We consider one local (derivative-based) and one global (meta-heuristic) parameter estimation method. As opposed to other comparative studies of parameter estimation methods that focus on identifying parameters of a single model structure, we compare the parameter estimation methods in the context of repetitive parameter estimation for a number of candidate model structures. The results confirm the superiority of the global optimization methods over the local ones in the context of structure identification.
    Ecological Modelling 02/2013; 245:136-165. · 2.07 Impact Factor
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: The aim of this paper is to discover a model equation for predicting the concentration of the algal species Peridinium gatunense (Dinoflagellate) in Lake Kinneret. This is a rather difficult task, due to the sudden ecosystem changes that occurred in the mid-1990s. Namely, the stable ecosystem (with regular Peridinium blooms until 1993) underwent changes and has transformed into an unstable system, with cyanobacterial blooms now occurring regularly. This shift in the algal succession is expected to influence attempts to model the lake ecosystem. Namely, the model structure before and after the change is likely to be different. Our modelling experiments were directed to discover a single model equation that can simulate dinoflagellate dynamics in both periods. We apply an automated modelling tool (Lagramge), which integrates the knowledge- and the data-driven modelling approach. In addition we include an expert visual estimation of the models discovered by Lagramge to assist in the selection of the optimal model. The dataset used included time-series measurements of typical data from the periods 1988 to 1992 and 1997 to 1999. Using the data and expert knowledge coded in a modelling knowledge library, Lagramge successfully discovered several suitable mathematical models for Peridinium. After the expert’s visual estimation and validation of the models, we propose one optimal model capable of long-term predictions.
    Environmental Modelling and Software 01/2011; 26(5):658-668. · 3.48 Impact Factor
  • Source
    Will Bridewell, Ljupco Todorovski
    Proceedings of the Twenty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2010, Atlanta, Georgia, USA, July 11-15, 2010; 01/2010
  • Saso Dzeroski, Ljupco Todorovski
    [show abstract] [hide abstract]
    ABSTRACT: Reconstructing biological networks, such as metabolic and signaling networks, is at the heart of systems biology. Although many approaches exist for reconstructing network structure, few approaches recover the full dynamic behavior of a network. We survey such approaches that originate from computational scientific discovery, a subfield of machine learning. These take as input measured time course data, as well as existing domain knowledge, such as partial knowledge of the network structure. We demonstrate the use of these approaches on illustrative tasks of finding the complete dynamics of biological networks, which include examples of rediscovering known networks and their dynamics, as well as examples of proposing models for unknown networks.
    Current Opinion in Biotechnology 09/2008; 19(4):360-8. · 7.86 Impact Factor
  • Aleksandar Peckov, Saso Dzeroski, Ljupco Todorovski
    [show abstract] [hide abstract]
    ABSTRACT: The paper addresses the task of polynomial regression, i.e., the task of inducing polynomials from numeric data that can be used to predict the value of a selected numeric variable. As in other learning tasks, we face the problem of finding an optimal trade-off between the complexity of the induced model and its predictive error. One of the approaches to finding this optimal trade-off is the minimal description length (MDL) principle. In this paper, we propose an MDL scheme for polynomial regression, which includes coding schemes for polynomials and the errors they make on data. We empirically compare this principled MDL scheme to an ad-hoc MDL scheme and show that it performs better. The improvements in performance are such that the polynomial regression approach we propose is now comparable in performance to other commonly used methods for regression, such as model trees.
    Advances in Knowledge Discovery and Data Mining, 12th Pacific-Asia Conference, PAKDD 2008, Osaka, Japan, May 20-23, 2008 Proceedings; 01/2008
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: In this paper, we apply automated modelling method Lagramge to the task of modelling phytoplankton dynamics in Lake Glumsø, Denmark. The approach is based on integrating expert knowledge in the process of automated model induction from measured data. It supports modelling of ecosystem dynamics with ordinary differential equations by following the mass conservation law. The data set used in this paper comprises 2 years daily measurements of data needed for phytoplankton modelling in lake. In order to have sufficient data set for training and testing the models, the entire data set was divided in two parts, each containing 1 year of daily measurements. The expert knowledge supplied to Lagramge consists of elementary models of the basic ecological processes related to the food web dynamics and rules for combining elementary into complex models of the whole system. By applying Lagramge on Lake Glumsø we discovered a set of phytoplankton models that showed good fit on the training data set. The models were evaluated by simulating them on testing data set, which revealed good performance of the models.
    Ecological Modelling 01/2008; 212(1-2):92-98. · 2.07 Impact Factor
  • Saso Dzeroski, Pat Langley, Ljupco Todorovski
    Computational Discovery of Scientific Knowledge, Introduction, Techniques, and Applications in Environmental and Life Sciences; 01/2007
  • Valentin Gjorgjioski, Ljupco Todorovski, Saso Dzeroski
    [show abstract] [hide abstract]
    ABSTRACT: Ordinary dierential equations are one of the most widely accepted formalisms for model- ing dynamic systems. The space of possible model structures can be specified by taking into account dierent types of domain knowl- edge. These include existing models, par- tially specified models, and building blocks from which models are composed. The do- main knowledge can be expressed in a form of context-free grammars or as generic pro- cesses. In the LAGRAMGE approach to process-based modeling, the domain knowl- edge in a form of generic processes is trans- formed into a context-free grammar that con- strains the space of dierential equations ex- plored by the system. The computational complexity of the task addressed by LA- GRAMGE is very high. Fitting the constant parameters in a single equation structure is computationally demanding, as it involves it- erative methods for nonlinear optimization. In this work, we propose to speed-up the pro- cess of ODE discovery by LAGRAMGE by parallelizing the LAGRAMGE algorithm.
    01/2007;
  • Saso Dzeroski, Ljupco Todorovski
    01/2007;
  • Conference Proceeding: Learning Declarative Bias.
    Will Bridewell, Ljupco Todorovski
    Inductive Logic Programming, 17th International Conference, ILP 2007, Corvallis, OR, USA, June 19-21, 2007, Revised Selected Papers; 01/2007
  • Ljupco Todorovski, Saso Dzeroski
    Computational Discovery of Scientific Knowledge, Introduction, Techniques, and Applications in Environmental and Life Sciences; 01/2007
  • Source
    Will Bridewell, Stuart R. Borrett, Ljupco Todorovski
    [show abstract] [hide abstract]
    ABSTRACT: In this paper, we introduce an approach for extracting constraints on process model construction. We begin by clarifying the type of knowledge produced by our method and how one may apply it. Next, we review the task of inductive process modeling, which provides the required data. We then introduce a logical formal- ism and a computational method for acquiring scientific knowledge from candidate process models. Results sug- gest that the learned constraints make sense ecologically and may provide insight into the nature of the modeled domain. We conclude the paper by discussing related and future work.
    Proceedings of the 4th International Conference on Knowledge Capture (K-CAP 2007), October 28-31, 2007, Whistler, BC, Canada; 01/2007
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: We address the task of inducing explanatory models from observations and knowledge about candidate biological processes, using the illustrative problem of modeling photosynthesis regulation. We cast both models and background knowledge in terms of processes that interact to account for behavior. We also describe IPM, an algorithm for inducing quantitative process models from such input. We demonstrate IPM's use both on photosynthesis and on a second domain, biochemical kinetics, reporting the models induced and their fit to observations. We consider the generality of our approach, discuss related research on biological modeling, and suggest directions for future work.
    Artificial Intelligence in Medicine 08/2006; 37(3):191-201. · 1.36 Impact Factor
  • Ljupco Todorovski, Nada Lavrac, Klaus P. Jantke
    01/2006;
  • Source
    Ljupco Todorovski, Will Bridewell, Oren Shiran, Pat Langley
    [show abstract] [hide abstract]
    ABSTRACT: Research on inductive process modeling combines back- ground knowledge with time-series data to construct explana- tory models, but previous work has placed few constraints on search through the model space. We present an extended formalism that organizes process knowledge in a hierarchi- cal manner, and we describe HIPM, a system that carries out constrained search for hierarchical process models. We report experiments that suggest this approach produces more accu- rate and plausible models with less effort. We conclude by discussing related research and directions for future work.
    Proceedings, The Twentieth National Conference on Artificial Intelligence and the Seventeenth Innovative Applications of Artificial Intelligence Conference, July 9-13, 2005, Pittsburgh, Pennsylvania, USA; 01/2005
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: In this paper, we review the paradigm of in- ductive process modeling, which uses back- ground knowledge about possible component processes to construct quantitative models of dynamical systems. We note that previ- ous methods for this task tend to overt the training data, which suggests ensemble learn- ing as a likely response. However, such tech- niques combine models in ways that reduce comprehensibility, making their output much less accessible to domain scientists. As an al- ternative, we introduce a new approach that induces a set of process models from dieren t samples of the training data and uses them to guide a nal search through the space of model structures. Experiments with syn- thetic and natural data suggest this method reduces error and decreases the chance of in- cluding unnecessary processes in the model. We conclude by discussing related work and suggesting directions for additional research.
    Machine Learning, Proceedings of the Twenty-Second International Conference (ICML 2005), Bonn, Germany, August 7-11, 2005; 01/2005
  • [show abstract] [hide abstract]
    ABSTRACT: In this report, the prediction performances of two rule evaluation measures, accuracy and weighted relative accuracy, are compared.
    02/2004;
  • Saso Dzeroski, Ljupco Todorovski, Peter Ljubic
    Constraint-Based Mining and Inductive Databases, European Workshop on Inductive Databases and Constraint Based Mining, Hinterzarten, Germany, March 11-13, 2004, Revised Selected Papers; 01/2004
  • Source
    Journal of Machine Learning Research. 01/2004; 5:153-188.
  • Source
    Branko Kavsek, Nada Lavrac, Ljupco Todorovski
    ROC Analysis in Artificial Intelligence, 1st International Workshop, ROCAI-2004, Valencia, Spain, August 22, 2004; 01/2004