I am an assistant Professor at Léonard de Vinci Pôle Universitaire, Research Center. I have done my Ph.D. in Computer Science at LIPN laboratory, University of Sorbonne Paris Nord. Some of my current research interests are: Machine Learning (Reinforcement Learning, Deep Reinforcement Learning, Markov Decision Processes, Preference Learning), Decision Theory (Qualitative Decision Making, Preference Elicitation), Optimization (Linear Programming, Integer Linear Programming), Machine Learning and Deep learning applications in Industry 4.0, Internet of Things, Natural Language Processing and Location Based Social Networks analysis.
Pegah Alizadeh; Emiliano Traversi; Aomar Osmani
Deterministic policies based on maximum regrets in MDPs with imprecise rewards Article de journal
Dans: Ai Communications, 2021.
@article{alizadeh_1653,
title = {Deterministic policies based on maximum regrets in MDPs with imprecise rewards},
author = {Pegah Alizadeh and Emiliano Traversi and Aomar Osmani},
url = {https://content.iospress.com/articles/ai-communications/aic190632},
year = {2021},
date = {2021-09-01},
journal = {Ai Communications},
abstract = {Markov Decision Process Models (MDPs) are a powerful tool for planning tasks and sequential decision-making issues. In this work we deal with MDPs with imprecise rewards, often used when dealing with situations where the data is uncertain. In this context, we provide algorithms for finding the policy that minimizes the maximum regret. To the best of our knowledge, all the regret-based methods proposed in the literature focus on providing an optimal stochastic policy. We introduce for the first time a method to calculate an optimal deterministic policy using optimization approaches. Deterministic policies are easily interpretable for users because for a given state they provide a unique choice. To better motivate the use of an exact procedure for finding a deterministic policy, we show some (theoretical and experimental) cases where the intuitive idea of using a deterministic policy obtained after ?determinizing? the optimal stochastic policy leads to a policy far from the exact deterministic policy.},
keywords = {},
pubstate = {online},
tppubtype = {article}
}
Pegah Alizadeh; Aomar Osmani; Mohamed Essaid Khanouche; Abdelghani Chibani; Yacine Amirat
Reinforcement Learning for Interactive QoS-Aware Services Composition Article de journal
Dans: Ieee Systems Journal, vol. 15, no. 1, p. 1098-1108, 2020.
@article{alizadeh_1236,
title = {Reinforcement Learning for Interactive QoS-Aware Services Composition},
author = {Pegah Alizadeh and Aomar Osmani and Mohamed Essaid Khanouche and Abdelghani Chibani and Yacine Amirat},
url = {https://ieeexplore.ieee.org/document/9110736},
year = {2020},
date = {2020-06-01},
journal = {Ieee Systems Journal},
volume = {15},
number = {1},
pages = {1098-1108},
abstract = {An important and challenging research problem in web of things is how to select an appropriate composition of concrete services in a dynamic and unpredictable environment. The main goal of this article is to select from all possible compositions the optimal one without knowing a priori the users' quality of service (QoS) preferences. From a theoretical point of view, we give bounds on the problem search space. As the QoS user's preferences are unknown, we propose a vector-valued MDP approach for finding the optimal QoS-aware services composition. The algorithm alternatively solves MDP with dynamic programming and learns the preferences via direct queries to the user. An important feature of the proposed algorithm is that it is able to get the optimal composition and, at the same time, limits the number of interactions with the user. Experiments on a real-world large size dataset with more than 3500 web services show that our algorithm finds the optimal composite services with around 50 interactions with the user.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Christophe Rodrigues; Pegah Alizadeh; Dmitry Bondarenko
Industrie 4.0 : Prédiction de données réelles par fine-tuning à partir de simulations Conférence
Extraction et Gestion de Connaissances, Bruxelles, Belgique, 2020.
@conference{rodrigues_1077,
title = {Industrie 4.0 : Prédiction de données réelles par fine-tuning à partir de simulations},
author = {Christophe Rodrigues and Pegah Alizadeh and Dmitry Bondarenko},
url = {https://egc2020.sciencesconf.org/resource/page/id/28},
year = {2020},
date = {2020-01-01},
booktitle = {Extraction et Gestion de Connaissances},
address = {Bruxelles, Belgique},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Héctor Murrieta; Ivan Vladimir Meza Ruiz; Pegah Alizadeh; Jorge Garcia Flores
Towards Identifying for Evidence of Drain Brain from Web Search Results using Reinforcement Learning Conférence
LatinX in AI Research at NeurIPS 2019, Vancouver, Canada, 2019.
@conference{murrieta_1221,
title = {Towards Identifying for Evidence of Drain Brain from Web Search Results using Reinforcement Learning},
author = {Héctor Murrieta and Ivan Vladimir Meza Ruiz and Pegah Alizadeh and Jorge Garcia Flores},
url = {https://www.latinxinai.org/neurips-2019},
year = {2019},
date = {2019-12-01},
booktitle = {LatinX in AI Research at NeurIPS 2019},
address = {Vancouver, Canada},
abstract = {Brain drain is the phenomenon in which experts on a field abandon their origin country to practise their profession in a different country. It forms part of the migration patterns around the world. However brain drain can have damaging effects on the source of origin when it happens at large scale. This has been problem in several countries of latinamerica, particularly at the postgraduate level. The correct characterisation of this phenomena is vital to outline polices that keep or attract the talent needed in these countries. In this research, we propose a methodology to identify evidence of drain brain through results of web search
engines which commonly contains links to career information pages given a seed name, however it could be very time consuming explore and analyse all resulting pages. For this reason, in this research we propose to exploit a Reinforcement Learning setting to learn to navigate and extract significant information from the snippet results. In this work we outline the main architecture based on the Dopamine RL framework.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Pegah Alizadeh; Peggy Cellier; Thierry Charnois; Bruno Cremilleux; Albrecht Zimmermann
Étude expérimentale d'extraction d'information dans des retranscriptions de réunions Conférence
CORIA TALN 2018, Rennes, France, 2018.
@conference{alizadeh_1500,
title = {Étude expérimentale d'extraction d'information dans des retranscriptions de réunions},
author = {Pegah Alizadeh and Peggy Cellier and Thierry Charnois and Bruno Cremilleux and Albrecht Zimmermann},
url = {https://project.inria.fr/coriataln2018/fr/},
year = {2018},
date = {2018-05-01},
booktitle = {CORIA TALN 2018},
address = {Rennes, France},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Christophe Haikal; Pegah Alizadeh; Christophe Rodrigues; Bi Chongke
Place Embedding across Cities in Location-based Social Networks Inproceedings
Dans: ACM Symposium on Applied Computing, Brno, Czech Republic, 2022.
@inproceedings{haikal_1767,
title = {Place Embedding across Cities in Location-based Social Networks},
author = {Christophe Haikal and Pegah Alizadeh and Christophe Rodrigues and Bi Chongke},
url = {The proceeding is not published yet:
https://www.sigapp.org/sac/sac2022/file2022/TOC-Jan-23-2022.pdf},
year = {2022},
date = {2022-01-01},
booktitle = {ACM Symposium on Applied Computing},
address = {Brno, Czech Republic},
abstract = {In the urban computing field, analysing human mobility patterns are used to understand tourists and residents behaviours, or urban areas functionalities. Transferring the information from one city to another also assists urban strategists in comparing two cities
and applying their knowledge from one to another, such as plan- ning new touristic attractions or implementing some successful urban strategies in new cities. For this goal, in this work, we propose a twofold semi-supervised approach: we first propose a place embedding method based on users mobility trajectories in each city individually, second we learn a place translation function between two cities using the set of reviews written for each place. The translation function is based on measuring textual and semantic
similarities of the reviews. Extensive experiments on two french metropolitan areas (Lille and Bordeaux) based on the extracted data from Trip-Advisor demonstrate the effectiveness of our approach.},
keywords = {},
pubstate = {accepted},
tppubtype = {inproceedings}
}
Aomar Osmani; Massinissa Hamidi; Pegah Alizadeh
Clustering Approach to Solve Hierarchical Classification Problem Complexity Inproceedings
Dans: AAAI Conferece on Artificial Intelligence, Vancouver, Canada, 2022, ISBN: will be completed later.
@inproceedings{osmani_1768,
title = {Clustering Approach to Solve Hierarchical Classification Problem Complexity},
author = {Aomar Osmani and Massinissa Hamidi and Pegah Alizadeh},
url = {The proceeding is not ready yet.
Even the list of accepted papers is not published yet. I will modify this part as soon as I have them.},
issn = {will be completed later},
year = {2022},
date = {2022-01-01},
booktitle = {AAAI Conferece on Artificial Intelligence},
address = {Vancouver, Canada},
abstract = {In a large domain of classification problems for real applications, like human activity recognition, separable spaces between groups of concepts are easier to learn than each concept alone. This is because the search space biases required to separate groups of classes (or concepts) are more relevant than the ones needed to separate classes individually. For example, it is easier to learn the activities related to the body
movements group (running, walking) versus ?on-wheels? activities group (bicycling, driving a car), before learning more specific classes inside each of these groups. Despite the ob-
vious interest of this approach, our theoretical analysis shows a high complexity for finding an exact solution. We propose in this paper an original approach based on the association of clustering and classification approaches to overcome this limitation. We propose a better approach to learn the concepts by grouping classes recursively rather than learning them class by class. We introduce an effective greedy algorithm and two theoretical measures (namely cohesion and dispersion) to evaluate the connection between the clusters and the classes. Extensive experiments on the SHL dataset show that our approach improves classification performances while reducing the number of instances used to learn each concept.},
keywords = {},
pubstate = {accepted},
tppubtype = {inproceedings}
}
Louis Zigrand; Pegah Alizadeh; Emiliano Traversi; Roberto Wolfer Calvo
Machine Learning Guided Optimization for Demand Responsive Transport Systems Inproceedings
Dans: European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Database, Bilbao Spain (online), 2021.
@inproceedings{zigrand_1542,
title = {Machine Learning Guided Optimization for Demand Responsive Transport Systems},
author = {Louis Zigrand and Pegah Alizadeh and Emiliano Traversi and Roberto Wolfer Calvo},
url = {https://2021.ecmlpkdd.org/wp-content/uploads/2021/07/sub_383.pdf},
year = {2021},
date = {2021-06-01},
booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Database},
address = {Bilbao Spain (online)},
abstract = {Most of the time, objective functions used for solving staticcombinatorial optimization problems cannot deal efficiently with theirreal-time counterparts. It is notably the case of Shared Mobility Systemswhere the dispatching framework must adapt itself dynamically to thedemand. More precisely, in the context of Demand Responsive Transport(DRT) services, various objective functions have been proposed in theliterature to optimize the vehicles routes. However, these objective func-tions are limited in practice because they discard the dynamic evolutionof the demand. To overcome such a limitation, we propose a MachineLearning Guided Optimization methodology to build a new objectivefunction based on simulations and historical data. This way, we are ableto take the demand's dynamic evolution into account. We also presenthow to design the main components of the proposed framework to fit aDRT application: data generation and evaluation, training process andmodel optimization. We show the efficiency of our proposed method-ology on real-world instances, obtained in a collaboration with PadamMobility, an international company developing Shared Mobility Systems.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Aomar Osmani; Pegah Alizadeh; Christophe Rodrigues
Equational Model Guided by Real-time Sensor Data to Monitor Industrial Robots Inproceedings
Dans: International Conference on Automation Science and Engineering, Lyon, France, 2021.
@inproceedings{osmani_1543,
title = {Equational Model Guided by Real-time Sensor Data to Monitor Industrial Robots},
author = {Aomar Osmani and Pegah Alizadeh and Christophe Rodrigues},
url = {https://case2021.sciencesconf.org/},
year = {2021},
date = {2021-06-01},
booktitle = {International Conference on Automation Science and Engineering},
address = {Lyon, France},
abstract = {The monitoring of industrial robots is often en-
sured by generic simulators which model the equational aspect
of the target machines. We propose an original approach to
complete the equational simulator of a milling machine using
the accumulated data from the used sensors. This approach cre-
ates a specific simulator for each machining situation by taking
the triplet (material, cutting tool, workpiece) into account. This
improvement brings great added value to the industrial experts
and improves the efficiency of industrial robots. It allows them
to better follow and interpret the behavior of machines during
the milling process. In addition to correct the simulator using
real data, our method detects also the anomalies during the
real manufacturing performance and fixes the minor bugs
along the observed real data during its continuous simulation
mimicry. The additional interest of our model remains the
precise definition of the complementary model between the real
system and the equational simulator. This makes it possible, by
using an inductive approach to search for regularities in the
model in order to better interpret the structural differences
between the model and the system and to better understand
the situations linked to their functionalities or undesirable
situations. The intensive experiments on real data validate our
model and open up many perspectives for future works.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Aomar Osmani; Massinissa Hamidi; Pegah Alizadeh
Hierarchical Learning of Dependent Concepts for Human Activity Recognition Inproceedings
Dans: K., Karlapalem (Ed.): Pacific-Asia Conference on Knowledge Discovery and Data Mining, p. 79-92, Springer, Cham, Dehli, India, 2021, ISBN: 978-3-030-75764-9.
@inproceedings{osmani_1499,
title = {Hierarchical Learning of Dependent Concepts for Human Activity Recognition},
author = {Aomar Osmani and Massinissa Hamidi and Pegah Alizadeh},
editor = {Karlapalem K. et al.},
url = {https://link.springer.com/chapter/10.1007/978-3-030-75765-6_7},
issn = {978-3-030-75764-9},
year = {2021},
date = {2021-04-01},
booktitle = {Pacific-Asia Conference on Knowledge Discovery and Data Mining},
volume = {12713},
pages = {79-92},
publisher = {Springer, Cham},
address = {Dehli, India},
abstract = {In multi-class classification tasks, like human activity recog-
nition, it is often assumed that classes are separable. In real applications,
this assumption becomes strong and generates inconsistencies. Besides,
the most commonly used approach is to learn classes one-by-one against
the others. This computational simplification principle introduces strong
inductive biases on the learned theories. In fact, the natural connections
among some classes, and not others, deserve to be taken into account.
In this paper, we show that the organization of overlapping classes (mul-
tiple inheritances) into hierarchies considerably improves classification
performances. This is particularly true in the case of activity recognition
tasks featured in the SHL dataset. After theoretically showing the expo-
nential complexity of possible class hierarchies, we propose an approach
based on transfer affinity among the classes to determine an optimal hi-
erarchy for the learning process. Extensive experiments show improved
performances and a reduction in the number of examples needed to learn.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Massinissa Hamidi; Aomar Osmani; Pegah Alizadeh
A Multi-View Architecture for the SHL Challenge Inproceedings
Dans: Adjunct Proceedings of the 2020 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2020 ACM International Symposium on Wearable Computers, p. 317-322, virtual event, 2020, ISBN: 978-1-4503-8076-8.
@inproceedings{hamidi_1275,
title = {A Multi-View Architecture for the SHL Challenge},
author = {Massinissa Hamidi and Aomar Osmani and Pegah Alizadeh},
url = {https://dl.acm.org/doi/10.1145/3410530.3414351},
issn = {978-1-4503-8076-8},
year = {2020},
date = {2020-09-01},
booktitle = {Adjunct Proceedings of the 2020 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2020 ACM International Symposium on Wearable Computers},
pages = {317-322},
address = {virtual event},
abstract = {To recognize locomotion and transportation modes in a user-independent manner with an unknown target phone position, we (team Eagles) propose an approach based on two main steps: reduction of the impact of regular effects that stem from each phone position, followed by the recognition of the appropriate activity. The general architecture is composed of three groups of neural networks organized in the following order. The first group allows the recognition of the source, the second group allows the normalization of data to neutralize the impact of the source on the activity learning process, and the last group allows the recognition of the activity itself. We perform extensive experiments and the preliminary results encourage us to follow this direction, including the source learning to reduce the phone position's biases and activity separately.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Pegah Alizadeh; Jorge Garcia Flores; Ivan Vladimir Meza Ruiz
Apprentissage par renforcement pour la recherche d'experts sur le web Inproceedings
Dans: Extraction et Gestion de Connaissances, Bruxelles, Belgique, 2020.
@inproceedings{alizadeh_1078,
title = {Apprentissage par renforcement pour la recherche d'experts sur le web},
author = {Pegah Alizadeh and Jorge Garcia Flores and Ivan Vladimir Meza Ruiz},
url = {https://egc2020.sciencesconf.org/resource/page/id/28},
year = {2020},
date = {2020-01-01},
booktitle = {Extraction et Gestion de Connaissances},
address = {Bruxelles, Belgique},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Pegah Alizadeh; Aomar Osmani; Emiliano Traversi
Calcul d'une politique de?terministe dans un MDP avec re?compenses impre?cises Inproceedings
Dans: EGC 2019, p. 45-56, Metz, France, 2019.
@inproceedings{alizadeh_614,
title = {Calcul d'une politique de?terministe dans un MDP avec re?compenses impre?cises},
author = {Pegah Alizadeh and Aomar Osmani and Emiliano Traversi},
url = {https://egc2019.sciencesconf.org/},
year = {2019},
date = {2019-01-01},
booktitle = {EGC 2019},
volume = {RNTI-E-35},
pages = {45-56},
address = {Metz, France},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
No posts by this author.
N'hésitez pas à contacter le service des admissions pour tout renseignement complémentaire :