Computational Intelligence - August 2017 - 9

natural language. Moreover, it can be successfully adapted to other target domains
as desired and proved experimentally
with several data sets.
The third paper "Learning to Generate Descriptions of V
isual Data Anchored
in Spatial Relations" was authored by
A. Muscat and A. Belz. Authors face the
challenge of automatically generating linguistic descriptions of images.They developed a novel method with three steps.
Firstly, objects in the image are identified.

Publication Spotlight

Secondly, spatial relations between objects
are discovered. Meaningful labels are
assigned paying attention to language and
visual features. Thirdly, labels are mapped
to natural language descriptions. The
proposal was thoroughly evaluated using
several NLG strategies and CI methods.
A random forest model performs best in
relation with learning the mapping
from features to spatial relations.
Finally, as Guest Editors of this special issue, we would like to thank authors

for their submissions, not only the
authors of the three accepted papers but
all authors who submitted a paper for
this special issue. In addition, we would
like to recognize the hard and great job
made by the anonymous referees who
supported the review process. We thank
them all for their kind, valuable and outstanding cooperation, as well as for
their constructive feedback.

(continued from page 7)

IEEE Transactions on Cognitive
and Developmental Systems

Learning From Explanations Using
-S entiment and Advice in RL, by
S. Krening, B. Harrison, K. M. Feigh,
C. L. Isbell, M. Riedl, and A. Thomaz,
IEEE Transactions on Cognitive and
Developmental Systems, Vol. 9, No. 1,
March 2017, pp. 44-55.
Digital Object Identifier: 10.1109/
TCDS.2016.2628365
"In order for robots to learn from
people with no machine learning expertise, robots should learn from natural
human instruction. Most machine learning techniques that incorporate explanations require people to use a limited
vocabulary and provide state information,
even if it is not intuitive. This paper discusses a software agent that learned to
play the Mario Bros. game using explanations.The authors' goals to improve learning from explanations were twofold: 1) to
filter explanations into advice and warnings and 2) to learn policies from sentences without state information. The authors
used sentiment analysis to filter explanations into advice of what to do and
warnings of what to avoid. The authors
developed object-focused advice to represent what actions the agent should take
when dealing with objects. A reinforcement learning agent used object-focused

advice to learn policies that maximized its
reward. After mitigating false negatives,
using sentiment as a filter was approximately 85% accurate. Object-focused
advice performed better than when no
advice was given, the agent learned where
to apply the advice, and the agent could
recover from adversarial advice.The authors
also found the method of interaction
should be designed to ease the cognitive
load of the human teacher or the advice
may be of poor quality."
IEEE Transactions on Emerging
Topics in Computational
Intelligence

Partitioning of Intelligent Buildings for
Distributed Contaminant Detection and
Isolation, by A. Kyriacou, S. Timotheou, M. P. Michaelides, C. Panayiotou, and M. Polycarpou, IEEE
Transactions on Emerging Topics in Computational Intelligence, Vol. 1, No. 2,
April 2017, pp. 72-86.
Digital Object Identifier: 10.1109/
TETCI.2017.2665119
"Intelligent buildings are responsible for
ensuring indoor air quality for their occupants under normal operation as well as
under possibly harmful contaminant
events. An emerging environmental application involves the monitoring of intelligent buildings against harmful events by

incorporating various sensing technologies
and using sophisticated algorithms to
detect and isolate such events. In this context, both centralized and distributed ap--
proaches have been proposed, with the
latter having significant benefits in terms of
complexity, scalability, reliability, and performance. This paper considers the automatic partitioning of the building into
subsystems, which enables the distributed
simulation, modeling, analysis, and management of the intelligent building while
ensuring the effective detection and isolation of contaminants in the building interior. Specifically, we develop both a
high-quality heuristic algorithm and an
optimal mixed integer linear programming
(MILP) formulation for the building partitioning problem. The MILP formulation is
based on graph partitioning techniques,
while the heuristic is based on matrix clustering techniques. Both ap--proaches partition the building into subsystems while
ensuring 1) maximum decoupling between
the various subsystems, 2) strong connectivity between the zones of each subsystem,
and 3) control of the size of the subsystems with respect to the number of allocated zones. A combination of the two
ap-proaches is also pro--posed for reconfiguring an initial partitioning composition in
real time in order to accommodate partitioning needs that arise from dynamic
system changes."

AUGUST 2017 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE

Table of Contents for the Digital Edition of Computational Intelligence - August 2017