User Stories

Lockheed Martin Assesses 5G Network Vulnerabilities with Reinforcement Learning Toolbox

“5G is a critical infrastructure that we must protect from adversarial attacks. Reinforcement Learning Toolbox allows us to quickly assess 5G vulnerabilities and identify mitigation methods.”

Ambrose Kam, Lockheed Martin
Learn more about Liquid Instruments

Key Outcomes

  • Instrument space reduced by up to 9X.  The MokuPro can perform the same functions as up to nine standalone and fixed function instruments which represents a 9-to-1 improvement in shelf space. 
  • FPGA compilation time reduced from hours to seconds.  Users can leverage a simple, iPad-based graphical interface to change instrument configuration and this step is accelerated by Vivado compiling VHDL code in the cloud rather than on a local compute resource. 
  • Connect, mix, and match an expanded set of instruments.  The MokuPro architecture enables users to simplify and compress their workflow while staying powered-on.  Manufacturing and Industrial professionals can dial-up software modules easily and dynamically to maximize production output and improve operations over time. 
FPGA Design and Codesign - Xilinx System Generator and HDL Coder

With approximately 35,000 employees working in 16 countries, Lockheed Martin Rotary and Mission Systems (RMS) develops a broad range of products and technologies. Recently, the RMS cybersecurity team has been working on assessing vulnerabilities in 5G infrastructure, including those related to data security, user privacy, confidentiality, integrity, and availability. These complex problems cannot be solved by traditional methods or approaches.

To assess these gaps, the team used MATLAB® and Reinforcement Learning Toolbox™ to discover attack scenarios. It began by building 5G models using EXata Cyber emulation software and defining a set of threat vectors based on security frameworks from the 3rd Generation Partnership Project (3GPP), National Security Agency (NSA), and other organizations. They then used Reinforcement Learning Toolbox to train an adversarial deep Q-network agent to optimize attack patterns and find worst-case scenarios. Based on these results, the team identified mitigation techniques to address the discovered vulnerabilities.

The team is now exploring opportunities to increase the fidelity of their simulations using 5G Toolbox™ and achieve better realism with a multi-agent reinforcement learning framework enabled by Simulink®.

Learn more about Lockheed Martin Rotary and Mission Systems

Key Outcomes/Advantages:

  • Shortened development and analysis time compared to traditional methods
  • Enabled cybersecurity engineers with little previous AI experience to implement the solution
  • Identified worst-case attack scenarios and developed mitigation methods
  • Quickly evaluated and compared multiple reinforcement learning algorithms

Products Used

Learn more about Liquid Instruments

Key Outcomes

  • Instrument space reduced by up to 9X.  The MokuPro can perform the same functions as up to nine standalone and fixed function instruments which represents a 9-to-1 improvement in shelf space. 
  • FPGA compilation time reduced from hours to seconds.  Users can leverage a simple, iPad-based graphical interface to change instrument configuration and this step is accelerated by Vivado compiling VHDL code in the cloud rather than on a local compute resource. 
  • Connect, mix, and match an expanded set of instruments.  The MokuPro architecture enables users to simplify and compress their workflow while staying powered-on.  Manufacturing and Industrial professionals can dial-up software modules easily and dynamically to maximize production output and improve operations over time. 
FPGA Design and Codesign - Xilinx System Generator and HDL Coder

Products Used