Making Research Tools Accessible for All AI Students

1
Making Research Tools Accessible for All AI
Students
Zach Dodds, Christine Alvarado, and Sara Sood
A Foundation in Pen-based Computing
Emotional Reasoning through Search
Though a compelling area of research with many
applications, emotional reasoning lacks
well-documented open-source tools. As a result,
undergraduate research tasks involving emotional
classification quickly become overwhelmingly
complex. To mitigate this problem, we make our
Reasoning through Search (RTS) system, with
training data and documentation, available to
students both for research projects and AI
courses.
With the rise of pen-based computing (PBC),
fundamental algorithms for working with ink and
for recognizing hand-drawn input, such as text,
gestures or sketches, have found their way into
many courses. For students to focus on the
algorithms, they need a platform for handling
mundane tasks, such as collecting and displaying
users hand-drawn strokes. To this end, we have
used WPF, the Windows Presentation Foundation.
Briefly, RTS uses a combination of Bayesian,
case-based Reasoning, and information retrieval
approaches to classify text with an accuracy of
78. Trained on 106,000 movie and product
reviews, RTS quantifies the emotional
classification of input text as a valence score
between -1 and 1.
Shared Results
In all three of these cases,
(1) Students could complete more ambitious
projects at terms end early exposure to these
tools fostered comfort and creativity.
A Pomona College Senior leveraged the RTS system
in a project that examines blogs, tracks their
topics and bloggers sentiments, and displays
this information. Above is EmoMeters, a
visualization that uses the RTS system to analyze
a bloggers sentiment toward different topics
WPFs environment is unfamiliar to many students
it runs in the .NET platform and uses both XAML
and C. Students also need time to learn the data
structures and functions for storing and
manipulating ink. To provide this time, our first
HW asks students to build a simple drawing
application similar to Windows Journal. Creative
extensions are encouraged some students included
complex interface controls (above right) others
had recognition correction interfaces (above
left). Yet all students found WPF a powerful
basis for the open-ended projects that were their
capstone work in pen-based computing.
(2) Pushing development details to HW made more
classroom time available for course content.
topics that the system discovered on its own. A
second display, EmoCloud (right), leveraged RTSs
underlying data to create a tag cloud in which
highly emotional terms, judged by the ratio of
negative and positive document appearances, orbit
their topic.
(3) Students interacted with large, real-world
data sets from the very beginning of their AI
experiences.
Getting Set with OpenCV
OpenCV is a powerful C-based library of visual
routines. Its role in Stanleys vision systems is
one of its claims to fame. Because it is
researchware, however, OpenCV is not easy for
students to simply pick up on their own. Thus,
to motivate several facets of the library, we ask
students to implement an agent that plays Set,
the game of visual perception.
Set motivates in part because it highlights the
differences between artificial and human
intelligence human players find it hard to ID
appropriate triples (sets), but the primary
computational challenge lies in correctly
identifying the four attributes of each card it
requires color segmentation, connected
components, and shape identification.
Starting from Set, students have extended their
code to recognize landmarks (left) and to segment
paths from surrounding green areas (below),
routines used to guide indoor and outdoor robots.
matching pixels
original image
overlay
Naïve mapping
FastSLAM 1.0