Artificial Intelligence/Machine Learning, ASA(ALT), Direct to Phase II
Sensor Synthetic Data Generation
Objective
US Army requires large-scale, accurate and easily accessible training, test, and validation data to support AI model development for multiple security domains (e.g. SIPR, JWICS…). Sensor data is critical to develop AI/ML models. Unfortunately, there is not enough data yet to create highly performant models. Sensor Synthetic Data Generation will potentially serve to reduce bottleneck of training data supply that helps improve ML models by developing a synthetic data generation tool.
Description
Currently, nearly all of the AI/ML models are developed using actual or representative data. There is not enough unique defense/intel data available to create performant models (e.g. it takes roughly 50M pieces of data to create a 60-70% performant model). Additionally, this data must be labeled; synthetically generated data has the ability to be labeled as it is generated, reducing human data labeling effort for real-world data and data generated from an external (e.g., vendor) source.
Sensor Synthetic Data Generation topic encompasses the development of a synthetic data generation tool for sensors (e.g. radar, etc.) that can augment the limited, labeled, training data available to support Artificial Intelligence / Machine Learning model development. The purpose of this topic is to lead to the creation/integration of mission-focused synthetic data to include but not be limited to: Priority Needs: Commercial Satellites/Electro Optical (EO) – World View 1,2,3 (Imagery), Digital Globe, Blacksky // Synthetic Aperture Radar (SAR) – RADARSAT and Capella; Other Needs: 0903 Full Motion Video (FMV) // Electronic Intelligence (ELINT) spectrums/waveforms // Variable Message Format (VMF) and Chat; Desired synthetic data to be used in AI/ML model development: Surface to Surface Radars, Surface to Air Missile Launchers, Tanks, Etc. Please note: labeled data is a critical input to model training and model test & eval.”
Phase I
This is a Direct to Phase II effort. Please see Phase II Topic Description for further instruction.
Phase II
Develop and demonstrate a technically feasible software prototype that showcases how the solution addresses the challenges described in the DESCRIPTION of this topic and meets or exceeds the OBJECTIVE of this topic. The demonstration shall show the prototype as a proof-of-concept in a form-factor compatible with Army uniformed officer staffing and deployment decisions.
Phase III
This SBIR would integrate Artificial Intelligence and Machine Learning algorithms as a pathfinder initiative into developing new models to support sensor modalities across the Army and will significantly improve in performance; thus increasing the ability to identify high value targets.
For more information, and to submit your full proposal package, visit the DSIP Portal.
References:
- Alzantot, Moustafa, et al. “SenseGen: A Deep Learning Architecture for Synthetic Sensor Data Generation.” 2017 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops), 2017, https://doi.org/10.1109/percomw.2017.7917555;
- Nikolenko, Sergey. Synthetic Data for Deep Learning. https://arxiv.org/pdf/1909.11512.pdf.
Objective
US Army requires large-scale, accurate and easily accessible training, test, and validation data to support AI model development for multiple security domains (e.g. SIPR, JWICS…). Sensor data is critical to develop AI/ML models. Unfortunately, there is not enough data yet to create highly performant models. Sensor Synthetic Data Generation will potentially serve to reduce bottleneck of training data supply that helps improve ML models by developing a synthetic data generation tool.
Description
Currently, nearly all of the AI/ML models are developed using actual or representative data. There is not enough unique defense/intel data available to create performant models (e.g. it takes roughly 50M pieces of data to create a 60-70% performant model). Additionally, this data must be labeled; synthetically generated data has the ability to be labeled as it is generated, reducing human data labeling effort for real-world data and data generated from an external (e.g., vendor) source.
Sensor Synthetic Data Generation topic encompasses the development of a synthetic data generation tool for sensors (e.g. radar, etc.) that can augment the limited, labeled, training data available to support Artificial Intelligence / Machine Learning model development. The purpose of this topic is to lead to the creation/integration of mission-focused synthetic data to include but not be limited to: Priority Needs: Commercial Satellites/Electro Optical (EO) – World View 1,2,3 (Imagery), Digital Globe, Blacksky // Synthetic Aperture Radar (SAR) – RADARSAT and Capella; Other Needs: 0903 Full Motion Video (FMV) // Electronic Intelligence (ELINT) spectrums/waveforms // Variable Message Format (VMF) and Chat; Desired synthetic data to be used in AI/ML model development: Surface to Surface Radars, Surface to Air Missile Launchers, Tanks, Etc. Please note: labeled data is a critical input to model training and model test & eval.”
Phase I
This is a Direct to Phase II effort. Please see Phase II Topic Description for further instruction.
Phase II
Develop and demonstrate a technically feasible software prototype that showcases how the solution addresses the challenges described in the DESCRIPTION of this topic and meets or exceeds the OBJECTIVE of this topic. The demonstration shall show the prototype as a proof-of-concept in a form-factor compatible with Army uniformed officer staffing and deployment decisions.
Phase III
This SBIR would integrate Artificial Intelligence and Machine Learning algorithms as a pathfinder initiative into developing new models to support sensor modalities across the Army and will significantly improve in performance; thus increasing the ability to identify high value targets.
For more information, and to submit your full proposal package, visit the DSIP Portal.
References:
- Alzantot, Moustafa, et al. “SenseGen: A Deep Learning Architecture for Synthetic Sensor Data Generation.” 2017 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops), 2017, https://doi.org/10.1109/percomw.2017.7917555;
- Nikolenko, Sergey. Synthetic Data for Deep Learning. https://arxiv.org/pdf/1909.11512.pdf.