Research

Our research focuses on developing and analyzing machine learning algorithms for various applications, including computer vision, cybersecurity, transportation, communications, energy, climate change, and healthcare. Specifically, in machine learning, our research covers three technical areas: anomaly detection, reinforcement learning (RL), and multimodal data fusion.

Detecting unexpected behaviors (e.g., anomalies, novelties) or in general changes in data patterns with respect to a baseline is an important problem in machine learning and statistics. There are some prominent applications in cyber and physical security, surveillance, and computer vision. Emerging applications such as IoT networks and video analytics pose new challenges (e.g., high dimensionality and heterogeneity) which prohibit the use of tractable probabilistic models and require data-driven approaches. To this end, we have been working on data-driven sequential change detection methods for real-time anomaly detection in various real-world problems. For instance, we developed the first real-time anomaly detection algorithms in video anomaly detection and cybersecurity with performance guarantees and closed-form expressions for detection threshold in the literature. Those results have been published in high-impact venues, such as the top machine learning journal (IEEE Transactions on Pattern Analysis and Machine Intelligence) and the top security journal (IEEE Transactions on Information Forensics and Security). Our anomaly detection research has been funded by NSF, DOT, Cyber Florida, and SCEEE.

Problem formulation for online detection and offline localization of video anomalies.

One of the proposed video anomaly detection systems.

The proposed video anomaly detection system provides interpretable results, in addition to achieving state-of-the-art detection and localization performance while controlling the false alarm rate with theoretical guidance.

Grants:

• NSF ECCS-2040572, PI, “Collaborative Research: Real-Time Data-Driven Anomaly Detection for Complex Networks”, $450,000 (Our portion $225,000), 8/21-7/24
• NSF ECCS-2029875, co-PI, “Collaborative Research: SWIFT: SMALL: Understanding and Combating Adversarial Spectrum Learning towards Spectrum-Efficient Wireless Networking”, $450,000 (USF part $270,000, Our portion $135,000), 9/20-8/23
• DOT-FRA, PI, “Autonomous Track Inspection System based on Passive Sensing and Anomaly Detection - Phase 1”, $177,672 (Our portion $127,000), 5/20-7/21
• Florida Center for Cybersecurity, PI, “Online Learning and Visualization for Intrusion Detection and Prevention in Privacy-Enhanced IoT Networks”, $75,000 (Our portion $37,500), 7/20-6/21
• SCEEE Research Initiation Grant 17-03, Sole PI, “Online Nonparametric Cyber-Attack Detection”, $25,000 (plus $25,000 cost sharing), 7/17-12/18

• M. Kurt, Y. Yilmaz and X. Wang, "Real-Time Nonparametric Anomaly Detection in High-Dimensional Settings", IEEE Transactions on Pattern Analysis and Machine Intelligence [pdf]
• Doshi, K. and Yilmaz, Y., 2021. Online anomaly detection in surveillance videos with asymptotic bound on false alarm rate. Pattern Recognition. [pdf]
• Aktukmak, M., Yilmaz, Y. and Uysal, I., 2021. Sequential Attack Detection in Recommender Systems. IEEE Transactions on Information Forensics and Security. [pdf]
• M. Kurt, Y. Yilmaz and X. Wang, "Real-Time Detection of Hybrid and Stealthy Cyber-Attacks in Smart Grid", IEEE Transactions on Information Forensics and Security, Feb. 2019 [pdf]
• E. Hou, Y. Yilmaz and A. Hero, "Anomaly Detection in Partially Observed Traffic Networks", IEEE Transactions on Signal Processing, 2019 [pdf]
• M. Kurt, Y. Yilmaz and X. Wang, "Distributed Quickest Detection of Cyber-Attacks in Smart Grid", IEEE Transactions on Information Forensics and Security, Aug. 2018 [pdf]

• IEEE BigData Cup Second Rank Award, Global Road Damage Detection Challenge, IEEE International Conference on Big Data (IEEE BigData) (2020)
• AI City Challenge Second Rank Award, Traffic Video Anomaly Detection Track, NVIDIA, IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020)
• Young Investigator Award, Southeastern Center for Electrical Engineering Education (SCEEE) (20% success rate)

• M. Necip Kurt, PhD, 9/2016-6/2020, Columbia University, co-advised with Dr. Xiaodong Wang, currently Research Scientist at Samsung
• Mahsa Mozaffari, 8/2017-8/2020, USF, currently PhD student at Rochester Institute of Technology
  
  
  

  ---

  ---

  
  
  Reinforcement Learning:
  
  

  RL provides a powerful tool for efficiently allocating limited resources while achieving high performance. We developed novel Markov decision process (MDP) formulations and RL algorithms for complex real-world problems in climate change, transportation, energy, communications, and healthcare. Our RL research has been funded by federal grants from NSF and NOAA, and research outcomes have been published in high-impact venues, such as the top machine learning conference (ICML) and the top transportation journal (IEEE Transactions on Intelligent Transportation Systems).

  Proposed MDP model for maximum power point tracking (MPPT) in photovoltaics.

  Power output of the proposed deep RL method successfully adapts to varying resistance and temperature in the system.

  13 minutes video presentation - Predictive Maintenance for Increasing EV ChargingLoad in Distribution Power System [pdf].

  15 minutes video presentation - Deep Reinforcement Learning Based Cost Benefit Analysis for Hospital Capacity Planning [pdf].

  15 minutes video presentation - A Markov Decision Process Model for Socio-Economic Systems Impacted by Climate Change [pdf].

  Grants:

  • NSF 2137891, Senior Personnel, “NSF Convergence Accelerator Track E: Linking the Green Economy to the Blue Economy at the Coast”, $750,000 (Our portion $30,000), 9/21-8/22
  • NOAA-GCOOS, co-PI, “Decision Support Tools for Maritime Transportation in the Gulf of Mexico", accepted, $428,764 (Our portion $117,897), 2/22-1/27
  • NSF CNS-1737598, PI, “SCC-Planning: Agent-Based Scenario Planning for a Smart & Connected Community Against Sea Level Rise in Tampa Bay”, $100,000 (Our portion $74,000), 9/17-8/19 [Project Website]
  
  Selected Publications:
  • Shuvo, S.S., Yilmaz, Y., Bush, A. and Hafen, M., 2020. "A Markov Decision Process Model for Socio-Economic Systems Impacted by Climate Change". International Conference on Machine Learning (ICML).[pdf]
  • Nassar, A. and Yilmaz, Y., 2021. "Deep Reinforcement Learning for Adaptive Network Slicing in 5G for Intelligent Vehicular Systems and Smart Cities." IEEE Internet of Things Journal.[pdf]
  • Haydari, A. and Yilmaz, Y., 2020. "Deep Reinforcement Learning for Intelligent Transportation Systems: A Survey." IEEE Transactions on Intelligent Transportation Systems. [pdf]
  • Shuvo, S.S., Ahmed, M.R., Symum, H. and Yilmaz, Y., 2021. Deep Reinforcement Learning Based Cost-Benefit Analysis for Hospital Capacity Planning. International Joint Conference on Neural Networks (IJCNN). [pdf]
  • Shuvo, S.S. and Yilmaz, Y., 2020. "Predictive Maintenance for Increasing EV Charging Load in Distribution Power System". IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm).[pdf]
  Students Graduated:
  • Almuthanna Nassar, PhD, 9/2016-6/2020, USF, currently Postdoctoral Researcher at Moffitt Cancer Center, Machine Learning Department
  • Ammar Haydari, MSc, 8/2017-7/2019, USF, currently PhD student at University of California, Davis
  Awards:
  • First Place Award in the 2nd Workshop on Continual Learning in Computer Vision, Reinforcement Learning Track, CVPR 2021
  
  
  

  ---

  ---

  
  
  Multimodal Data Fusion:
  
  

  Integrating heterogeneous data observed from a complex system can yield information diversity and in turn improved decision making. As opposed to the traditional decision fusion techniques which separately analyze different data modalities and integrate processed information, this modern data science (Big Data) challenge requires fusion methods that jointly analyze heterogeneous data to leverage cross-information between data modalities. We developed the first generic data integration technique for disparate data types (categorical and numerical) from the exponential family of distributions, and successfully applied it through deep neural networks to various problems in recommender systems, cybersecurity, genomics, social networks, and public safety. Our research in this domain is funded by federal grants from NIST and DOD. Representative papers have appeared in the top security (IEEE Transactions on Information Forensics and Security) and machine learning (Neurocomputing) journals. In the top recommender systems conference (RecSys), our paper was selected for oral presentation (3.6% acceptance rate) and received the Best Paper Honorable Mention Award.

  33 minutes video presentation - Scalable Data Fusion for Real-Time Event Detection, NIST ASAPS Contest 1 Winner: Team USF-EE.

  
  

  Our joint variational learning algorithm is capable of extracting shared communities across all graph layers as well as identifying communities unique to each layer. This figure shows for each chromosome the number of shared clusters, the number of private clusters for RNA gene expression levels and the number of private clusters for HiC contact maps between genes. This analysis suggests that the genes in chromosomes 2, 5, 15, 16, 20, 21, 22 are strongly co-expressed (high connectivity in RNA graph) and strongly connected (high connectivity HiC graph) since the number of shared clusters is dominant compared to the number of private clusters, while the genes in the other chromosomes are either strongly co-expressed or strongly connected.

  Grants:

  • DOD-USSOCOM, Senior Personnel, “Data-Driven Intelligence for Active Identification and Characterization”, $2,000,000 (Our portion $127,463), 8/19-12/21
  • NIST, PI, “Scalable Multimodal Data Representation from Video, Text, Audio for Real-Time Event Detection”, $30,000 (Our portion $30,000), 9/20-12/21
  
  Selected Publications:
  • Y. Yilmaz, M. Aktukmak and A. Hero, 2021. ``Multimodal Data Fusion in High-Dimensional Heterogeneous Datasets via Generative Models", IEEE Transactions on Signal Processing.[pdf]
  • Aktukmak, M., Yilmaz, Y. and Uysal, I., 2021. Sequential Attack Detection in Recommender Systems. IEEE Transactions on Information Forensics and Security. [pdf]
  • M. Aktukmak, Y. Yilmaz and I. Uysal, "Quick and Accurate Attack Detection in Recommender Systems through User Attributes", ACM Conference on Recommender Systems, 2019 [pdf] *Best Paper Award Runner-up
  • H. Ali, S. Liu, Y. Yilmaz, A. Hero, R. Couillet and I. Rajapakse, "Latent Heterogeneous Multilayer Community Detection", IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2019 [pdf]
  • M. Aktukmak, Y. Yilmaz and I. Uysal, "Matrix Factorization with Multimodal Side Information”, Neurocomputing, 2019
  • Y. Yilmaz and A. Hero, "Multimodal Event Detection in Twitter Hashtag Networks", Journal of Signal Processing Systems, 2018 [pdf]
  Students Graduated:
  • Mehmet Aktukmak, PhD, 8/2017-12/2020, USF, co-advised with Dr. Ismail Uysal, currently Postdoctoral Research Fellow at University of Michigan, Ann Arbor
  Awards:
  • Winner, NIST Automated Streams Analysis for Public Safety Challenge, 2020
  • Best Paper Honorable Mention Award, ACM Conference on Recommender Systems (RecSys), 2019