publications

2022

1. ICML

   DNS: Determinantal Point Process Based Neural Network Sampler for Ensemble Reinforcement Learning

   Hassam Sheikh, Kory Frisbee, and Mariano Phielipp

   In International Conference on Machine Learning, 2022

   Abs Bib

   Ensemble methods improve the stability and performance of deep RL agents, but training large ensembles is computationally expensive. We propose DNS, a Determinantal Point Process-based neural network sampler that reduces computation by 50% while outperforming full-ensemble baselines by selecting maximally diverse subsets of networks for backpropagation at each training step.

   @inproceedings{sheikh2022dns,
     title = {{DNS}: {D}eterminantal Point Process Based Neural Network Sampler for Ensemble Reinforcement Learning},
     author = {Sheikh, Hassam and Frisbee, Kory and Phielipp, Mariano},
     booktitle = {International Conference on Machine Learning},
     year = {2022},
   }

2. ICLR

   Maximizing Ensemble Diversity in Deep Reinforcement Learning

   Hassam Sheikh, Mariano Phielipp, and Ladislau Bölöni

   In International Conference on Learning Representations, 2022

   Abs Bib

   We propose five regularization methods for ensemble RL that prevent value-function collapse and maximize representation diversity in parameter space, leading to significantly more stable and performant agents.

   @inproceedings{sheikh2022medrl,
     title = {Maximizing Ensemble Diversity in Deep Reinforcement Learning},
     author = {Sheikh, Hassam and Phielipp, Mariano and B{\"o}l{\"o}ni, Ladislau},
     booktitle = {International Conference on Learning Representations},
     year = {2022},
   }

3. IJCNN

   Learning Intrinsic Symbolic Rewards in Reinforcement Learning

   Hassam Sheikh, Shauharda Khadka, Santiago Miret, and 2 more authors

   In International Joint Conference on Neural Networks, 2022

   Abs Bib

   We propose an intrinsic reward generator based on learned symbolic trees, producing interpretable intrinsic rewards via arithmetic and logical operators that improve sample efficiency in sparse-reward settings.

   @inproceedings{sheikh2022lisr,
     title = {Learning Intrinsic Symbolic Rewards in Reinforcement Learning},
     author = {Sheikh, Hassam and Khadka, Shauharda and Miret, Santiago and Majumdar, Somdeb and Phielipp, Mariano},
     booktitle = {International Joint Conference on Neural Networks},
     year = {2022},
   }

2021

1. MSWiM

   Interaction and Behaviour Evaluation for Smart Homes: Data Collection and Analytics in the ScaledHome Project

   Matteo Mendula, Siavash Khodadadeh, Salih Safa Bacanli, and 5 more authors

   In International Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems, 2021

   Bib

   @inproceedings{mendula2021smarthomes,
     title = {Interaction and Behaviour Evaluation for Smart Homes: {D}ata Collection and Analytics in the {ScaledHome} Project},
     author = {Mendula, Matteo and Khodadadeh, Siavash and Bacanli, Salih Safa and Zehtabian, Sharare and Sheikh, Hassam and B{\"o}l{\"o}ni, Ladislau and Turgut, Damla and Bellavista, Paolo},
     booktitle = {International Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems},
     year = {2021},
   }

2020

1. IJCNN

   Multi-Agent Reinforcement Learning for Problems with Combined Individual and Team Reward

   Hassam Sheikh and Ladislau Bölöni

   In International Joint Conference on Neural Networks, 2020

   Abs Bib

   We propose DE-MADDPG, a multi-critic cooperative MARL framework that simultaneously maximizes global team rewards and local agent rewards, reducing parametric growth from exponential to linear and improving performance by 97% over baselines.

   @inproceedings{sheikh2020demaddpg,
     title = {Multi-Agent Reinforcement Learning for Problems with Combined Individual and Team Reward},
     author = {Sheikh, Hassam and B{\"o}l{\"o}ni, Ladislau},
     booktitle = {International Joint Conference on Neural Networks},
     year = {2020},
   }

2019

1. AAMAS

   Emergence of Scenario-Appropriate Collaborative Behaviors for Teams of Robotic Bodyguards

   Hassam Sheikh and Ladislau Bölöni

   In International Conference on Autonomous Agents and Multiagent Systems, 2019

   Abs Bib

   We demonstrate how multi-agent policy gradient algorithms can be adapted to learn collaborative robot behaviors for VIP protection in crowded public spaces, producing complex emergent bodyguard formations.

   @inproceedings{sheikh2019aamas,
     title = {Emergence of Scenario-Appropriate Collaborative Behaviors for Teams of Robotic Bodyguards},
     author = {Sheikh, Hassam and B{\"o}l{\"o}ni, Ladislau},
     booktitle = {International Conference on Autonomous Agents and Multiagent Systems},
     year = {2019},
   }

2. COMPSAC

   Learning Distributed Cooperative Policies for Security Games via Deep Reinforcement Learning

   Hassam Sheikh, Mina Razghandi, and Ladislau Bölöni

   In IEEE International Conference on Computer Software and Applications, 2019

   Abs Bib

   We find optimal defender policies using policy gradient in a partially observable cooperative RL framework for security game scenarios.

   @inproceedings{sheikh2019compsac,
     title = {Learning Distributed Cooperative Policies for Security Games via Deep Reinforcement Learning},
     author = {Sheikh, Hassam and Razghandi, Mina and B{\"o}l{\"o}ni, Ladislau},
     booktitle = {IEEE International Conference on Computer Software and Applications},
     year = {2019},
   }

2014

1. IJINS

   Automatic Feature Extraction, Categorization and Detection of Malicious Code in Android Applications

   Muhammad Zuhair Qadir, Atif Nisar Jilani, and Hassam Sheikh

   International Journal of Information & Network Security, 2014

   Bib

   @article{qadir2014malware,
     title = {Automatic Feature Extraction, Categorization and Detection of Malicious Code in {Android} Applications},
     author = {Qadir, Muhammad Zuhair and Jilani, Atif Nisar and Sheikh, Hassam},
     journal = {International Journal of Information \& Network Security},
     volume = {3},
     number = {1},
     pages = {12--17},
     year = {2014},
   }