Page 32 - Artificial Intelligence for the Internet of Everything

P. 32

CHAPTER 2

Uncertainty Quantification in

Internet of Battlefield Things

Brian Jalaian, Stephen Russell
US Army Research Laboratory, Adelphi, MD, United States

2.1 INTRODUCTION

The battlefield of the future will comprise a vast array of heterogeneous com-
putational capable sensors, actuators, devices, information sources, analytics,
humans, and infrastructure, with varying intelligence, capabilities, and con-
straints on energy, power, computing, and communication resources. To
maintain information dominance, the future army must far surpass the ability
of future opposing forces, kinetic and nonkinetic alike, to ensure mission suc-
cess and minimize risks. This vision can only be achieved if systems progress
from a state of total dependence on human control to autonomous and, ulti-
mately, autonomic behavior. A highly adaptive and autonomous future will
dependuponfoundational algorithms,theories,andmethodsthatdonotfully
exist today. There are many challenges in ensuring that the delegated mission
goalsareaccomplishedregardlessoftheimmediateavailabilityofhumanpres-
ence in the control loop or the partial attrition of mission assets.
Duringthe pastdecade, therehas beena tremendousgrowth in the field of
machine learning (ML). Large datasets combined with complex algorithms,
such as deep neural networks, have allowed for huge advances across a variety
ofdisciplines.However,despitethesuccessofthesemodelstherehasnotbeen
as much focus on uncertainty quantification (UQ); that is, quantifying a
model’s confidence in its predictions. In some situations, UQ may not be a
huge priority (e.g., Netflix recommending movie titles). But in situations
where the wrong prediction is a matter of life or death, UQ is crucial. For
instance, if army personnel in combat are using an ML algorithm to make
decisions, it is vital to know how confident the given algorithm is in its pre-
dictions. Personnel may observe a data point in the field that is quite different
from the data the algorithm was trained on, yet the algorithm will just supply a
(likely poor) prediction, potentially resulting in a catastrophe.

27 28 29 30 31 32 33 34 35 36 37