AISTATS*2014 Talks and Papers
The AISTATS 2014 proceedings-track papers are available in the
H01 Gaussian Processes for Data-Efficient Learning in Robotics and
Marc Deisenroth, Dieter Fox, Carl Rasmussen
Autonomous reinforcement learning (RL) approaches typically
require many interactions with the system to learn controllers,
which is a practical limitation in real systems, such as robots,
where many interactions can be impractical and time consuming. To
address this problem, current learning approaches typically
require task-specific knowledge in form of expert demonstrations,
realistic simulators, pre-shaped policies, or specific knowledge
about the underlying dynamics. We follow a different approach and
speed up learning by extracting more information from data. In
particular, we learn a probabilistic, non-parametric Gaussian
process transition model of the system. By explicitly
incorporating model uncertainty into long-term planning and
controller learning our approach reduces the effects of model
errors, a key problem in model-based learning. Compared to
state-of-the art RL our model-based policy search method achieves
an unprecedented speed of learning. We demonstrate its
applicability to autonomous learning in challenging real robot
and control tasks.
MP Deisenroth, D Fox, and CE Rasmussen. Gaussian Processes for
Data-Efficient Learning in Robotics and Control.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014.
Accepted for publication.
H02 Bayesian Monitoring for the Comprehensive Nuclear-Test-Ban Treaty
Stuart Russell, Erik Sudderth, Nimar Arora
Verification for the Comprehensive Nuclear-Test-Ban Treaty
requires detecting and characterizing all seismic events
above a minimum magnitude occurring anywhere on Earth. The
treaty defines a network of sensors, the International
Monitoring System (IMS), managed by the United Nations
CTBTO. NET-VISA, a Bayesian monitoring system applied to
IMS data, exhibits a 2x-3x reduction in detection failures
compared to the current CTBTO system; the UN has
recommended its deployment for treaty verification,
subject to approval by member states. NET-VISA's prior
is a complex, open-universe generative probability model
(written originally in the Bayesian Logic formal language
and trained on historical data) describing event
occurrence, signal propagation, signal detection, and
noise processes; the evidence consists of "blips"
(above-threshold signals, 90% of which are noise)
extracted from raw IMS waveform data. More recent work
extends the generative model all the way to the raw
waveforms, promising greater sensitivity but requiring
new modeling and inference techniques.
Nimar S. Arora, Stuart Russell, and Erik Sudderth,
``NET-VISA: Network Processing Vertically Integrated
Seismic Analysis.'' In Bulletin of the Seismological
Society of America, 103(2A), 709-729, 2013.
H03 Representation Learning: A Review and New Perspectives
The success of machine learning algorithms generally depends on
data representation, and we hypothesize that this is because
different representations can entangle and hide more or less the
different explanatory factors of variation behind the data.
Although specific domain knowledge can be used to help design
representations, learning with generic priors can also be used,
and the quest for AI is motivating the design of more powerful
representation-learning algorithms implementing such priors. This
paper reviews recent work in the area of unsupervised feature
learning and deep learning, covering advances in probabilistic
models, auto-encoders, manifold learning, and deep networks.
This motivates longer-term unanswered questions about the
appropriate objectives for learning good representations, for
computing representations (i.e., inference), and the geometrical
connections between representation learning, density estimation
and manifold learning.
Yoshua Bengio, Aaron Courville, Pascal Vincent,
"Representation Learning: A Review and New Perspectives,"
IEEE Transactions on Pattern Analysis and Machine Intelligence,
vol. 35, no. 8, pp. 1798-1828, Aug. 2013, doi:10.1109/TPAMI.2013.50
H04 Spatiotemporal point process models of conflicts
Andrew Zammit-Mangion, Michael Dewar,
Visakan Kadirkamanathan, Guido Sanguinetti
Modern conflicts are characterised by an ever increasing use of
information and sensing technology, resulting in vast amounts of
high resolution data. Modelling and prediction of conflict,
however, remains a challenging task due to the heterogeneous and
dynamic nature of the data typically available. Here we propose
the use of dynamic spatiotemporal modelling tools for the
identification of complex underlying processes in conflict, such
as diffusion, relocation, heterogeneous escalation and
volatility. Using ideas from statistics, signal processing and
ecology, we provide a predictive framework able to assimilate
data and give confidence estimates on the predictions. We
demonstrate our methods on the Wikileaks Afghan War Diary. Our
results show that the approach allows deeper insights into
conflict dynamics and allows a strikingly accurate (in a
statistical sense) forward prediction of armed opposition group
activity in 2010, based solely on data from previous years.
Andrew Zammit-Mangion, Michael Dewar, Visakan Kadirkamanathan,
and Guido Sanguinetti, Point process modelling of the Afghan
War Diary, Proc Natl Acad Sci U S A. 2012 July 31; 109(31):