72 AI MAGAZINE
This workshop follows the success of previous
health-related AAAI workshops, including those
focused on personalized and population health-
care, and the two subsequent joint workshops held
at AAAI- 17 and AAAI- 18. This year’s two-day work-
shop brought together a wide range of participants
(roughly 70 registrants) from the multidisciplinary
field of medical and health informatics. Partici-
pants were interested in the theory and practice of
computational models of web-based public health
intelligence as well as personalized healthcare
delivery. The full and short papers and the posters
presented at the workshop covered a broad range
of disciplines within AI, including knowledge rep-
resentation, machine learning, natural language
processing, prediction, mobile technology, inference,
and dialogue systems. From an application perspec-
tive, presentations addressed topics in epidemiology,
environmental and public health informatics, dis-
ease surveillance and diagnosis, medication dosing,
health behavior monitoring, and human-computer
interaction.
The workshop included an invited talk by Barry
O’Sullivan (University College Cork), who gave a
presentation on case studies in improving healthcare delivery. To further promote the work presented
at the workshop, the authors of mature research were
given the opportunity to submit revised and significantly extended manuscripts for review to appear in
a special issue of the Journal of Artificial Intelligence in
Medicine on precision digital medicine and health.
Martin Michalowski, Arash Shaban-Nejad, David L.
Buckeridge (McGill University), John S. Brownstein
(Harvard University), and Niels Peek (University
of Manchester) will serve as guest editors for this
collection.
Martin Michalowski and Arash Shaban-Nejad
served as cochairs of this workshop and submitted
this report. The workshop papers were published
by Springer in the Studies in Computational Intelligence series.
Knowledge
Extraction from Games
The second workshop on Knowledge Extraction from
Games again focused on mechanically extracting
knowledge from games — including but not limited
to game rules, character graphics or audio, environ-
ments, high-level goals or heuristic strategies, trans-
ferrable skills, aesthetic standards and conventions,
or abstracted models of games. Games enjoyed by
human players have been an area of interest for
AI from the days of fraudulent chess automata to
today’s superhuman play of Go and Atari games. But
games are more than just planning problems: where-
as deep Q-learning and other efforts yield successful
policies for playing specific games, we might want to
ask different questions of a game system besides how
does one win.
Games provide useful structuring information
for many reasoning tasks and are therefore an ideal
environment for this work. For example, games in
which nonplayer characters (or environment design)
offer hints to solving problems might be useful stepping stones toward contextual query answering; it
is not enough to find the right solution, but one
must identify the relationship between the textual
or visual hints and the correct embodied actions.
Games often share genre conventions and other similarities, or continually force a player to learn new
skills or exercise their existing competencies in novel
contexts; therefore, it seems especially interesting to
explore transfer learning and analogical reasoning
within and between games.
This workshop brought together practitioners
from these communities and others whose goals
overlap but whose approaches are developed in
parallel — search, general (video) game playing,
reinforcement learning, design support, human
factors, sequence analysis, and others. We had a
significant increase in submissions, accepted papers,
and attendance from the previous year. Our authors
presented 11 papers, ranging from applications of
local search for game exploration to an analysis of
how humans learn game rules. We also hosted two
invited talks. Alexander Zook explained some ways
in which Blizzard Entertainment’s data science
team analyzes players’ behavioral data to achieve
design goals, and Stanford University’s Srijan Kumar
explored the automated analysis of player gaze in
games of deception.
The key strength of this workshop was, as hoped,
the integration of multiple communities of AI and
automated reasoning researchers and game designers.
The questions posed after talks were stimulating,
and at least one of the papers has been cited by
upcoming work in automated game state-space
exploration. We look forward to hosting the workshop
again and seeing the new syntheses that emerge in
the next round of submissions!
The cochairs of the workshop were Joseph Osborn
(Pomona College), Matthew Guzdial (Georgia Tech),
and Samuel Snodgrass (Drexel University). The
proceedings were published as CEUR Workshop
Proceedings, Volume 2313. Joseph C. Osborn wrote
this report.
Network
Interpretability for Deep Learning
The AAAI- 19 workshop on Network Interpretability
for Deep Learning brought together scientists, engi-
neers, and students in both academic and industrial