Marija Kukic

Articles in conference proceedings [5]

2024

Kukic M., Bierlaire M. (2024) Hybrid Simulator for Projecting Synthetic Households in Unforeseen Events. Proceedings of the 24th Swiss Transport Research Conference (STRC), 15-17 May, Ascona, Switzerland


@InProceedings{KukicBierlaire_STRC2024,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Hybrid Simulator for Projecting Synthetic Households in

Unforeseen Events},
  booktime = {Proceedings of the 24th Swiss Transport Research Conference (STRC), 15-17 May},
  year = {2024},
  address = {Ascona, Switzerland}}}

2023

Kukic M., Benchelabi S., Bierlaire M. (2023) Hybrid Simulator for Capturing Dynamics of Synthetic Population. Proceedings of the 26th IEEE International Conference on Intelligent Transportation Systems (ITSC), 24 September -28 September, Bilbao, Bizkaia, Spain


@InProceedings{KukicB_ITSC2023,
  author = {Marija Kukic and Salim Benchelabi and Michel Bierlaire},
  title = {Hybrid Simulator for Capturing Dynamics of

Synthetic Population},
  booktime = {Proceedings of the 26th IEEE International Conference on Intelligent Transportation Systems (ITSC), 24 September -28 September},
  year = {2023},
  address = {Bilbao, Bizkaia, Spain}}}

Kukic M., Bierlaire M. (2023) Hybrid Simulator for Capturing Dynamics of Synthetic Population. Proceedings of the 23rd Swiss Transport Research Conference (STRC), 10-12 May, Ascona, Switzerland


@InProceedings{KukicBierlaire_STRC2023,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Hybrid Simulator for Capturing Dynamics of

Synthetic Population},
  booktime = {Proceedings of the 23rd Swiss Transport Research Conference (STRC), 10-12 May},
  year = {2023},
  address = {Ascona, Switzerland}}}

2022

Kukic M., Bierlaire M. (2022) One-step simulator for synthetic household generation. Proceedings of the 22nd Swiss Transport Research Conference (STRC), 18-20 May, Ascona, Switzerland


@InProceedings{KukicBierlaire_STRC2022,
  author = {Marija Kukic and Michel Bierlaire},
  title = {One-step simulator for synthetic household generation},
  booktime = {Proceedings of the 22nd Swiss Transport Research Conference (STRC), 18-20 May},
  year = {2022},
  address = {Ascona, Switzerland}}}

2021

Kukic M., Bierlaire M. (2021) Population synthesis at the level of households. Proceedings of the 21st Swiss Transport Research Conference (STRC), 12-14 September, Ascona, Switzerland


@InProceedings{KuBi_STRC2021,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Population synthesis at the level of households},
  booktime = {Proceedings of the 21st Swiss Transport Research Conference (STRC), 12-14 September},
  year = {2021},
  address = {Ascona, Switzerland}}}

Talks [10]

2024

Kukic M., Bierlaire M. Hybrid Simulator for Projecting Synthetic Households in Unforeseen Events September 04, 2024. Conference in Emerging Technologies in Transportation Systems (TRC-30), , Heraklion, Crete.


@Misc{2024Kukic_TRC30,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Hybrid Simulator for Projecting Synthetic Households in Unforeseen Events},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/2024Kukic_TRC30.pdf}},
  year = {2024},
  note = {Presented at Conference in Emerging Technologies in Transportation Systems (TRC-30), , Heraklion, Crete}}}

Kukic M., Bierlaire M. Hybrid Simulator for Projecting Synthetic Households in Unforeseen Events June 07, 2024. 16th workshop on Discrete Choice Models, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland.


@Misc{2024Kukic_DCM,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Hybrid Simulator for Projecting Synthetic Households in Unforeseen Events},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/2024Kukic_DCM.pdf}},
  year = {2024},
  note = {Presented at 16th workshop on Discrete Choice Models, Ecole Polytechnique F\éd\érale de Lausanne, Lausanne, Switzerland}}}

Kukic M., Bierlaire M. Hybrid Simulator for Projecting Synthetic Households in Unforeseen Events May 15, 2024. STRC 2024, Monte Veritá, Ascona, Switzerland.


@Misc{Kukic_STRC2024,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Hybrid Simulator for Projecting Synthetic Households in Unforeseen Events},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/Kukic_STRC2024.pdf}},
  year = {2024},
  note = {Presented at STRC 2024, Monte Verit\á, Ascona, Switzerland}}}

Kukic M., Bierlaire M. Synthetic Population Projections and Unforeseen Events February 06, 2024. 2nd EPFL Symposium on Transportation Research, EPFL, Barcelona, Spain.


@Misc{2024KukicBarcelona,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Synthetic Population Projections and Unforeseen Events},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/2024KukicBarcelona.pdf}},
  year = {2024},
  note = {Presented at 2nd EPFL Symposium on Transportation Research, EPFL, Barcelona, Spain}}}

2023

Kukic M., Bierlaire M. Hybrid Simulator for Capturing Dynamics of Synthetic Populations September 23, 2023. IEEE Intelligent Transportation Systems Conference (ITSC), , Bilbao, Spain.


@Misc{2023Kukic_ITSC,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Hybrid Simulator for Capturing Dynamics of Synthetic Populations},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/2023Kukic_ITSC.pdf}},
  year = {2023},
  note = {Presented at IEEE Intelligent Transportation Systems Conference (ITSC), , Bilbao, Spain}}}

Kukic M., Bierlaire M. Hybrid Simulator for Capturing Dynamics of Synthetic Population May 10, 2023. STRC 2023, Monte Veritá, Ascona, Switzerland.


@Misc{Kukic_STRC2023,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Hybrid Simulator for Capturing Dynamics of Synthetic Population},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/Kukic_STRC2023.pdf}},
  year = {2023},
  note = {Presented at STRC 2023, Monte Verit\á, Ascona, Switzerland}}}

Kukic M., Bierlaire M. Synthetic Population Generation: Hierarchies and Activities February 17, 2023. 1st EPFL Symposium on Transportation Research, EPFL, Annecy, France.


@Misc{KukicInternal,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Synthetic Population Generation: Hierarchies and Activities},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/KukicInternal.pdf}},
  year = {2023},
  note = {Presented at 1st EPFL Symposium on Transportation Research, EPFL, Annecy, France}}}

2022

Kukic M., Bierlaire M. One-step simulator for synthetic household generation May 19, 2022. STRC 2022, Monte Veritá, Ascona, Switzerland.


@Misc{Kukic_STRC2022,
  author = {Marija Kukic and Michel Bierlaire},
  title = {One-step simulator for synthetic household generation},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/Kukic_STRC2022.pdf}},
  year = {2022},
  note = {Presented at STRC 2022, Monte Verit\á, Ascona, Switzerland}}}

Kukic M., Bierlaire M. One-step simulator for the generation of synthetic households April 27, 2022. ETH workshop on agent-based models , Institute for Transport Planning and Systems (ETHZ-IVT), Zurich, Switzerland.


@Misc{2022Kukic_ETHZ,
  author = {Marija Kukic and Michel Bierlaire},
  title = {One-step simulator for the generation of synthetic households},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/2022Kukic_ETHZ.pdf}},
  year = {2022},
  note = {Presented at ETH workshop on agent-based models , Institute for Transport Planning and Systems (ETHZ-IVT), Zurich, Switzerland}}}

2021

Kukic M., Bierlaire M. Population synthesis at the level of households September 14, 2021. STRC 2021, Monte Veritá, Ascona, Switzerland.


@Misc{Kukic_STRC2021,
  author = {Marija Kukic and Michel Bierlaire},
  title = {Population synthesis at the level of households},
  howpublished = {\url{https://transp-or.epfl.ch/documents/talks/Kukic_STRC2021.pdf}},
  year = {2021},
  note = {Presented at STRC 2021, Monte Verit\á, Ascona, Switzerland}}}

Reviewing [1]

Transportation Research Part C: Emerging Technologies (1)

Research projects [1]

2020

Activity scheduling an rhythmic style: multi-day modeling of mobility habits

A growing body of research shows that the traditional approaches employed to plan and forecast travel behaviors are not equipped to deal with the heterogeneity of behaviors over time, space, and social spheres. Therefore, travel policies struggle to reconcile social inclusivity, sustainability and network efficiency. Two ENAC laboratories propose to join forces, in mathematical modeling and quantitative sociology, to develop a novel multi-day activity-scheduling framework to forecast travel demand. Integrating day-to-day correlations in travel behaviors will lead to a better understanding of the motives behind travel decisions, and will unveil more facets of individual decision making for better predictions of their daily mobility choices. This research uses the MOBIS dataset, a 8-week / 3700-respondent travel survey conducted in Switzerland in 2019. The forecasting model will be applied to the Swiss synthetic population and several scenarios will be considered.
- Oct 01, 2020 - Oct 01, 2021
- Sponsor: School of Architecture, Civil and Environmental Engineering (ENAC), École polytechnique fédérale de Lausanne
- Principal investigator: Janody Pougala
- Project manager: Janody Pougala
- Collaborator: Marija Kukic

EPFL courses [7]

Spring 2024

Introduction to Transportation Systems.
- Level: Bachelor
- Section: Civil Engineering
- Lecturer: Michel Bierlaire
- Teaching assistants: Marija Kukic, Negar Rezvany

Fall 2023

Introduction to optimization and operations research .
- Level: Bachelor
- Sections: Mechanical Engineering, Civil Engineering, Environmental Sciences and Engineering, Electrical and Electronics Engineering
- Lecturer: Michel Bierlaire
- Teaching assistants: Fabian Torres, L�a Ricard, Marija Kukic, Tom Haering, Negar Rezvany

Spring 2023

Introduction to Transportation Systems.
- Level: Bachelor
- Section: Civil Engineering
- Lecturer: Michel Bierlaire
- Teaching assistants: Marija Kukic, Janody Pougala, Mengyi Wang

Fall 2022

Introduction to optimization and operations research .
- Level: Bachelor
- Sections: Mechanical Engineering, Civil Engineering, Environmental Sciences and Engineering, Electrical and Electronics Engineering
- Lecturer: Michel Bierlaire
- Teaching assistants: Fabian Torres, Selin Atac, Marija Kukic, Negar Rezvany

Spring 2022

Decision-aid methodologies in transportation.
- Level: Master
- Section: Civil Engineering
- Lecturers: Tim Hillel, Nour Dougui
- Teaching assistants: Selin Atac, Janody Pougala, Marija Kukic, Negar Rezvany, Tom Haering

Spring 2021

Decision-aid methodologies in transportation.
- Level: Master
- Section: Civil Engineering
- Lecturers: Tim Hillel, Nour Dougui
- Teaching assistants: Selin Atac, Janody Pougala, Marija Kukic, Negar Rezvany, Nicola Ortelli, Stefano Bortolomiol
- Webpage: https://transp-or.epfl.ch/courses/decisionAid2020/index.php

Fall 2020

Introduction to optimization and operations research .
- Level: Bachelor
- Sections: Mechanical Engineering, Civil Engineering, Environmental Sciences and Engineering, Electrical and Electronics Engineering
- Lecturer: Michel Bierlaire
- Teaching assistants: Nour Dougui, Selin Atac, Stefano Bortolomiol, Marija Kukic, None None, Gael Lederrey, Cloe Cortes Balcells, Negar Rezvany

Lectures [3]

2024

Bayesian Estimation; Individual Prediction.
- Discrete Choice Analysis: Bayesian estimation and Individual prediction, EPFL, Lausanne, Switzerland
- Feb 15, 2024
- Lecturer: Pavel Ilinov
- Assistants: Marija Kukic, Evangelos Paschalidis

2023

Hierarchical Bayesian Estimation; Individual Prediction.
- Discrete Choice Analysis: Predicting Individual Behavior and Market Demand, EPFL, Lausanne, Switzerland
- Feb 02, 2023
- Lecturer: Janody Pougala
- Assistants: Marija Kukic, Nicola Ortelli
Logit Estimation and Specification Testing.
- Discrete Choice Analysis: Predicting Individual Behavior and Market Demand, EPFL, Lausanne, Switzerland
- Jan 29, 2023
- Lecturer: Tom Haering
- Assistants: Cloe Cortes Balcells, Marija Kukic, Nicola Ortelli

MSc theses [4]

2023

Capturing dynamics in synthetic population, Salim Benchelabi (Mathematics)
- Supervision: Marija Kukic, Janody Pougala, Michel Bierlaire
- External expert: Antonin Danalet
Synthetic generation of activity motifs, Quentin Bochud (Civil Engineering)
- Supervision: Marija Kukic, Janody Pougala, Michel Bierlaire
- External expert: Gael Lederrey

2021-2022

Modélisation du réseau de transports publics genevois à la suite d'un événement réseau, Nicolas Richter (Civil Engineering)
- Supervision: Marija Kukic, Janody Pougala, Michel Bierlaire
- External expert: Antoine Stroh

2021

The railway timetable rescheduling problem with capacity constraints, Benoit Pahud (Civil Engineering)
- Supervision: Nour Dougui, Marija Kukic, Stefano Bortolomiol
- External expert: Hellwig Matthias

Semester projects [9]

2024-2025

Optimization of railway maintenance planning, Younis Bouzar

Supervision: Léa Ricard, Marija Kukic, Michel Bierlaire
This project aims to develop a model for optimizing railway maintenance planning. The objective is to implement an efficient algorithm that can be applied in real-world networks. Cracks that develop on the rail surface, such as 'head checks', can lead to catastrophic failures if not promptly addressed. To mitigate this risk, preventive grinding is applied. In this work, the objective is to optimize the frequency of preventive grinding interventions for each geometric element of the railway, ensuring both rail safety and operational efficiency. The optimization model utilizes head check forecasts and existing interval scheduling data as inputs.

2024

Comparison of different methodologies for interpolating traffic count data, Mya Jamal Lahjouji (Civil Engineering)

Supervision: Marija Kukic, Nicholas Molyneaux, Michel Bierlaire
In this project, our objective is to evaluate various methodologies for interpolating gaps in traffic count data, assessing their effectiveness in relation to computational and data requirements. We focus on the case study of Switzerland using the data provided by Canton Vaud. This data has been collected from different locations over several years using different data collection procedures which resulted in having unbalanced data with different missing information and inconsistencies. In order to construct a realistic network, we want to investigate the different methodologies to fill these missing information and compare it with the existing solution provided by Canton.

2023-2024

Generating a synthetic population with health and COVID-19 information, Aline Janvier

Supervision: Cloe Cortes Balcells, Marija Kukic, Michel Bierlaire

2022-2023

Comparison of machine learning techniques for synthetic activity generation, Xinling Li (Civil Engineering)

Supervision: Marija Kukic, Nicola Ortelli, Michel Bierlaire
This project aims to investigate the good and the bad sides of state-of-the-art Machine Learning techniques applied in the context of synthetic population generation, focusing on the generation of attributes that de- scribe synthetic activities. Recently, there has been an increase in research focus on the application of deep generative learning algorithms for generating synthetic populations and travel behaviour modelling, including Variational Auto-Encoders (VAEs) and Generative Adversarial Networks (GANs) (Borysov et al., 2019, Wong and Farooq, 2020). Whilst initial experiments show promising results, there are still significant drawbacks with using generative modelling on population synthesis, which includes vanishing gradients, training instability or mode collapse, nor it allows for exact evaluation of the probability density of new points. Based on the literature review, we try to propose possible improvements for existing methodologies by identifying if there is a way to capture better the correlations between variables. One of the techniques that could be improved is Generative Adversarial Networks (GANs). For instance, DATGAN (Lederrey et al., 2022) is a GAN architecture that tries to give more control to the modellers when generating synthetic tabular data. It uses a Directed Acyclic Graph (DAG) to represent the correlation between variables in the dataset. This DAG is then used to define the structure of the Generator. Each variable is represented by an LSTM cell. However, due to the linear nature of LSTM cells, it is required to have a DAG instead of an acyclic graph. This directionality can be an issue when designing the DAG since it represents causality between variables. Thus, we could improve the DATGAN by removing the directionality and only specifying correlations between variables. Graph Neural Networks (GNNs) are a specific type of NN designed to mimic any type of graph, and they can be used to design the architecture for the Generator. This change would give more freedom to the user and make the method more flexible since DATGAN would not require any specific type of graph

Synthetic generation of activity motifs, Bochud Quentin Philippe (Civil Engineering)

Supervision: Marija Kukic, Janody Pougala, Michel Bierlaire
This project aims to work on the extension of the existing simulation framework (Kukic et al., 2021). In the existing methodology, we developed a new approach so-called Divide and Conquer Gibbs Sampler, for the generation of synthetic households. The current version of the framework can generate socio-demographic attributes of individuals (such as gender and age) grouped into households (described by their own set of features) in a realistic and representative way. However, in ABMs, there is a need to generate activities, not only socio-demographic characteristics. Some of the features that describe activities are type, location, start time, etc. However, as shown in the past, the simulation methods struggle to deliver accurate results in a reasonable time while dealing with high-dimensional datasets. According to that, instead of gener- ating disaggregated data on individuals� activities, it would be better to generate aggregated data while preserving the same amount of information. To do so, we will generate synthetic activity motifs. The daily activity motifs are representative graphs where the nodes describe the activity type and location, and the links indicate the sequential order between activities. The advantage of generating activity motifs instead of disaggregated activities is that we can reveal daily mobility patterns while saving computational time and preserving privacy. The first step is to retrieve and replicate the real activity motifs at the level of individu- als. The additional step would be to investigate how to generate the activity motifs of households depending on activities that individuals performed, subject to household constraints (e.g. sharing of resources). Note that the work of pre-study should be focused on the generation of activities that describe the movements of individuals. The investigation of activity motifs at the level of households can be part of the master thesis.

2022

Divide and conquer one-step simulator for synthetic household generation, Xinling Li (Civil Engineering)[Internship]

Supervision: Marija Kukic, Michel Bierlaire
Transportation science today is tasked with predicting the complex mobility needs of individuals, which necessitates the use of advanced mobility and travel demand models. However, the quality of the model outputs depends on the data quality. In transport, data privacy and data availability are two limitations. Therefore, transportation scientists rely increasingly on the usage of synthetic populations. Typically, a synthetic population is generated either on the level of individuals or on the level of households using simulation or machine learning approaches. This project aims to solve the problems of existing simulation techniques for the generation of synthetic households, addressing several literature gaps. In existing methodologies, the generation of individuals and their matching into households is done separately, through two sequential processes. Although the marginal distributions of key generated attributes might show a perfect fit, the �two-step� household generator produces unrealistic households. The generation of illogical observations is caused by neglecting the dependencies between individuals while grouping them into households. In order to create realistic households, this paper suggests a �single-step� household simulator where relationships between individuals are considered simultaneously within the generation process by imposing various statistical constraints. However, as shown in the past, the simulation methods struggle to deliver accurate results in a reasonable time while dealing with high-dimensional datasets. This project changes the traditional Gibbs Sampler into so-called �Divide and Conquer Gibbs Sampler� that solves this problem by decomposing and parallelizing the generation process based on the level of correlation. The expectation of this approach is that it will increase accuracy and efficiency, as highly correlated areas are isolated, enabling a better representation of less probable values. This project aims to work on the extension of the aforementioned existing framework. The extension will contain several steps where we will have to add more attributes that will be generated, which includes modeling of all necessary relationships. The addition of more variables will cause the curse of dimensionality phenomena. To overcome this issue, we will work on improving the divisive Gibbs Sampler in the context of synthetic household generation. We will compare it with the traditional approach by investigating the cost of adding more attributes regarding computational efficiency. Moreover, we will focus on proving the advantages of a one-step simulator compared to a two-step simulator in terms of the realism of generated observations. The methodology will be tested using 2015 Swiss census data.

Road sections clustering, Salim Benchelabi (Mathematics)

Supervision: Nour Dougui, Marija Kukic, Michel Bierlaire
Pavement maintenance is one of the major issues of public agencies. Inefficient maintenance strategies lead to high economic expenses in the long term. Public agencies rely on pavement management system (PMS) to recommend pavement rehabilitation and maintenance. The first step of a PMS is to create homogeneous worksites of reasonable sizes (considered practical and economically relevant). This is performed by grouping road sections based on conditions data and degradation laws over time for the network. In practice, the grouping or clustering phase is done by a basic algorithm whose results then need to be manually checked and repaired. This makes the task tedious and time consuming. The goal of this project is in a first instance to implement a deterministic clustering algorithm that takes as input road sections information (position, type of surface, type of traffic, age...) together with a deterministic degradation law and regroups the sections in clusters using Machine Learning algorithms such as k-means. In a second instance, a probabilistic degradation law will be considered and a probabilistic clustering algorithm such as Expectation�Maximization will be implemented. Both algorithms will be tested on real world case study and the result will be compared to the results of an existing PMS clustering methodology.

2021

Modeling Geneva's public transport network following disruption events, Nicolas Richter (Civil Engineering)

Supervision: Marija Kukic, Janody Pougala, Michel Bierlaire
The project is conducted in collaboration with the Transports Publics Genevois (TPG). The principal objective is to simulate the impact of disruptions (e.g. accidents) on the Geneva road network, and to optimize its resilience by proposing targeted and ecient intervention strategies. A particular emphasis will be put on the public transport network.

The railway timetable rescheduling problem, Benoit Pahud (Civil Engineering)

Supervision: Nour Dougui, Marija Kukic, Stefano Bortolomiol
In railway networks, unexpected disruptions may occur for different reasons and cause delays, service denial, and, consequently, passenger inconvenience. This pre-study project will look at remediation strategies such as canceling, delaying or rerouting trains in case of unexpected disruptions. This pre-study project follows two main lines of research. First, the student will familiarize with state-of-the-art methodologies and algorithms to generate railway timetables in a disrupted network. Second, the student will learn how these methodologies can be integrated in the commercial software Viriato, produced by the company SMA und Partner AG.