Dr. Erik Blasch
Air Force Office of Scientific Research
Talk Title: Autonomy in Use for Information Fusion Systems (Plenary)
Talk Title: Deep Learning Applicability Measures of Effectiveness (Deep Learning and AI Track)
Erik Blasch is a program officer at the United States Air Force Research Laboratory (AFRL) Air Force Office of Scientific Research (AFOSR). He has been with AFRL for 22 years, with assignments in Dayton, OH; Arlington, VA; Valcartier, QC; and Rome, NY. Additional positions include adjunct professor at Wright State University, IEEE Aerospace and Electronics Systems Society (AES) Board of governors, President of the International Society of Information Fusion (ISIF), and associate editor of 3 journals. He has focused on information fusion, target tracking, pattern recognition and robotics research compiling 750+ scientific papers, 22 patents, and 32 tutorials. His books include High-Level Information Fusion Management and Systems Design (Artech House, 2012), Context-enhanced Information Fusion (Springer, 2016), Multispectral Image Fusion and Colorization (SPIE, 2018), and Handbook of Dynamic Data Driven Applications Systems (Springer, 2019). He is an Associate Fellow AIAA, Fellow of SPIE, and Fellow of IEEE.
Current trends in autonomy result from many technical advancements including artificial intelligence, information processing, and systems design. Key to future developments includes instrumentation, modeling, and computational architectures; respectively. Examples of architectures are autonomy in motion for dynamic data assessment systems (e.g., robotics) and autonomy at rest for static data collection systems (e.g., security). However, using the analysis from streaming data architectures, there is a need for autonomy in use (AIU). AIU requires pragmatic use of message passing and data flow architectures, contextual and theoretic modeling, and user and information fusion. Information fusion provides methods for data aggregation, correlation, and temporal analysis. AIU accesses the data required to make real-time decisions; especially for systems that require information fusion, operator infusion, and control diffusion. An example is presented for command-guided swarms which brings together autonomy at rest, in motion, and in use.
Abstract (Deep learning and AI):
The rapid resurgence of interest in artificial intelligence (AI) results from impressive performance in deep learning. Many deep learning algorithms utilize hierarchical supervised training such as the convolutional Neural Network (CNN). Current AI needs focus on contextual reasoning, explainable results, and repeatable understanding. The deployment of many deep learning (DL) techniques raises concerns akin to that of standard pattern recognition techniques in the last 20 years. The talk will focus on measures of performance (MOP) and measures of effectiveness (MOE) for DL techniques.
The MOP concerns include (1) Timeliness: Computational Efficiency, (2) Accuracy: Operational Robustness, and (3) Confidence: Semi-supervised Representation. The MOE concerns include (1) Throughput: Data Efficiency, (2) Security: Adversarial Robustness, and (3) Completeness: Problem representation. Other ways to measure the quality of DL techniques should align with known efforts in verification and validation testing.