Data Scientist with 10+ years of industry experience.
I am a data scientist with background in machine learning, network security and fraud detection. My current interests lie in the areas of cyber security, graph neural networks, graph databases, building machine learning pipelines and general machine learning model orchestration.
My expertise is a mix of machine learning and software skills. I have been responisble for delivering machine learning models as well as designing machine learning pipelines, PySpark pipelines, developing and deploying Python packages, desiging Jenkins CI/CD jobs for package and model release, developing MLFlow and Airflow pipelines.
Python, Julia, SQL, Java (beginner)
Scikit-Learn, PyTorch, TensorFlow
Amazon Web Services (Batch, Lambda, S3), Apache Spark/Pyspark, Pandas, Dask
Graph Convolutional Networks (GCN), Graph Attention Networks (GAT), GraphSage, Deep Graph Infomax (DGI), Graph Isomorphism Networks (GIN), Neo4j
Kubeflow, MLFlow, Airflow, NiFi, Papermill
2002 - 2007
University of Maryland
College Park
Thesis: Intrusion Detection for Defense at the MAC and Routing Layers of Wireless Networks
Advisor: John S. Baras
2000 - 2002
University of Maryland
College Park
Thesis: Detection and Classification of Network Intrusions using Hidden Markov Models
Advisor: John S. Baras
1999
University of Belgrade
SerbiaMajor: Telecommunications
July 2021
Senior Data Scientist
San Diego, CAPart of the team responsible for the development of ISS Cyber Risk Score
November 2020 - July 2021
Senior Data Scientist
San Diego, CAApplication fraud: Worked on projects involving Graph Neural Networks (GNN) for fraud detection on homogeneous and heterogeneous graphs. Explored Graph Convolutional Networks (GCN), Graph Attention Networks (GAT), GraphSage, Deep Graph Infomax (DGI), Graph Isomorphism Networks (GIN) and other state of the art Graph Neural networks.
Tools development: Part of the team that develops tools and pipelines for data analysis and end-to-end model development.
November 2017 - November 2020
Lead Data Scientist
San Diego, CAApplication fraud: Part of the team that developed algorithms and methods for detecting application fraud. Responsible for developing machine learning models for fraud detection: building fraud detection variables, risk models, building data analysis pipeline using Python and Spark and exploring new modeling approaches for improving the existing models. Project lead on several fast-paced PoC projects focused on developing state of the art fraud detection models.
Graphs: Developed a Neo4j model for detecting complex fraud patterns in application fraud and GNN model for detecting application fraud.
September 2014 - November 2017
Analytic Scientist II
San Diego, CACyber security: Part of the team that developed analytic model for detecting various levels of security threats, including DDoS attacks, beaconing, exfiltration and numerous other types of attacks. Analyzed large amounts of data to gather insights on normal and malicious behavior and reduce the number of false positives. Experience with analysis and threat detection in HTTP, DNS and NetFlow. Relevant links: FICO Takes Fraud-Detection Techniques to Cybersecurity, FICO Cyber Security, FICO's CyberSecurity Analytics Solution to Detect and Preclude Malicious Network Activity
Graph databeses for fraud detection: used Neo4j for detectiong credit card payment fraud, including detection of credit card testing sites. Used several months worth of credit card transaction data to create a model for detecting complex fraud networks.
Research Scientist
Palo Alto, CA
A. A. Cardenas, S. Radosavac and J. S. Baras, “Evaluation of Detection Algorithms for MAC Layer Misbehavior: Theory and Experiments”, IEEE/ACM Transactions on Networking (ToN), Pages 605-617, Vol. 17, Issue 2. April 2009.
S. Radosavac, G. V. Moustakides, J. S. Baras and I. Koutsopoulos, “An analytic framework for modeling and detecting access layer misbehavior in wireless networks”, ACM Transactions on Information and System Security (ACM TISSEC), Vol. 11, No. 4, July 2008.
Svetlana Radosavac, Alvaro A Cárdenas, John S Baras, George V Moustakides, Detecting IEEE 802.11 MAC layer misbehavior in ad hoc networks: Robust strategies against individual and colluding attackers, Journal of Computer Security, Volume 15, Issue 1, January 2007, pp 103–128
H. Sharara, C. Westphal, S. Radosavac and U. C. Kozat, “Utilizing Social Influence in Content Distribution Networks” in Proceedings of IEEE ICC-2011, Kyoto, Japan, 2011 (best paper award)
J. Grossklags, S. Radosavac, A. Cardenas, J. Chuang, “Nudge: Intermediaries’ Role in Interdependent Network Security” Proceedings of 3rd International Conference on Trust and Trustworthy Computing (Trust’10), June 2010.
A. Cardenas, S. Radosavac, J. Grossklags, J. Chuang and C. Hoofnagle, “An Economic Map of Cybercrime”, 37th Research Conference on Communication, Information and Internet Policy (TPRC) 2009, George Mason University Law School, Arlington, VA, September 25-27, 2009.
S. Radosavac and J. S. Baras, “Application of Sequential Detection Schemes for Obtaining Performance Bounds of Greedy Users in the IEEE 802.11 MAC”, IEEE Communications Magazine: special issue on Security in Mobile Ad Hoc and Sensor Networks, pages 148-154, Vol. 46, No. 2, February 2008.
S. Radosavac, J. Kempf and U. C. Kozat , “Using Insurance for Increasing Internet Security”, ACM SIGCOMM Workshop on the Economics of Networks, Systems and Computation (NetEcon ’08), August 22, Seattle, WA
A. A. Cardenas, S. Radosavac and J. S. Baras, “An Analytical Evaluation of MAC Layer Misbehavior Detection Schemes”, Proceedings of the 26th Annual IEEE Conference on Computer Communications, INFOCOM 2007.
S. Radosavac, J. S. Baras, I. Koutsopoulos, "A framework for MAC protocol misbehavior detection in wireless networks", Proceedings of the 4th ACM workshop on Wireless security, WiSE 2005