provide the community with a set of rules. In a surveillance scenario, the data can come from location
and user reports augmented by historical models of
object behaviors in different scenarios using latent
analysis for categorical assessment (Kashoob et al.
2009; Blasch et al. 2014), while the context includes
known cultural norms of the desired area.
Another aspect is cooperative intelligence, which
is the ability of multiple systems to coordinate actions.
Many AI systems must develop joint training methods
to determine if coordinated action enhances performance. Recently, efforts have been focused on cooperative deep neural networks such as multimodal
information fusion (Ngiam et al. 2011), signals fusion (Shen et al, 2018), and image fusion (Zheng
et al. 2018). The multimodal analysis can serve not
only with assessment, but also in developing models for coordinated action. In a surveillance scenario,
cooperative sensing can position sensors collecting
data at the correct locations using contextual information to facilitate relevant data collection for object
detection as in swarm behavior (Cruise et al. 2018).
Elements of AI Systems
for Contextual Reasoning
Physical, social, and cooperative intelligence includes
similar aspects of model building, entity extraction,
relationship linking, and event assessment. Emerg-
ing concepts include graphical models, Markov
logic networks, statistical relational learning, and
DL networks. Together these approaches support
contemporary efforts in AI/ML toward contextual
reasoning for machines to be self-aware in explain-
ing choices for actions.
The three waves of AI include the first phase
(1960–1980) for handcrafted knowledge and rules
(Fogg 2017; Cruise et al. 2018), as shown in figure 2.
The second phase (starting in 1990) includes popular
methods in ML using statistical analysis such as natural language processing and computer vision. The
third wave (current) seeks to develop explainable
methods for scenarios and situations. However, there
is still a gap in machine–human teaming, machines
that think, and machines that rival humans with
common sense. To address the issues of human–
machine teaming, it is important to understand
the different types of AI and ML methods as well as
combining data such as deep multimodal image fusion
(Liu, S. et al. 2018).
ML attempts to build models in support of AI goals,
for which a variety of methods exist (Domingo 2015;
Cruise et al. 2018). The types of ML approaches foster
multiple opportunities for analysis, many of which are
common in information fusion methods and typically
designed for the specific scenarios, data, and reasoning
desired (table 5).
The various ML approaches stem from the types of
data that are collected (figure 3), either unlabeled or
labeled. If a world model is known with prior information
Data Statistical Features Decision Machine engineering (sensor)
Rules Logic Symbol Command Knowledge engineering (human)
Table 4. Machine-Knowledge Coordination.