DEML 2016 - 2016 Workshop on Data-Efficient Machine Learning

1. Call for Papers
We invite researchers to submit their recent work on the development and
analysis of methods leading towards more data-efficient machine
learning. A submission should take the form of an extended abstract of
*2 pages* in PDF format using the ICML style. Author names do not need
to be anonymized, and references may extend as far as needed beyond the
2 page upper limit. If authors' research has previously appeared in a
journal, workshop, or conference (including the ICML 2016 conference),
their workshop submission should extend that previous work.
Submissions will be accepted as contributed talks or poster
presentations. Extended abstracts should be submitted by 1 May; see
website for submission details. Final versions will be posted on the
workshop website (and are archival but do not constitute a proceedings).
2. Workshop Abstract
Recent efforts in machine learning have addressed the problem of
learning from massive amounts data. We now have highly scalable
solutions for problems in object detection and recognition, machine
translation, text-to-speech, recommender systems, and information
retrieval, all of which attain state-of-the-art performance when trained
with large amounts of data. In these domains, the question that is
increasingly asked, is how we can design machine learning systems that
achieve the same performance, but that learn efficiently using less
data. Other problem domains, such as personalized healthcare, robot
reinforcement learning, sentiment analysis, and community detection, are
characterized as either small-data problems for which data will always
be scarce, or big-data problems that are a collection of small-data
problems. The ability to learn in a sample-efficient manner is a
necessity in these data-limited domains.
Collectively, these problems highlight the increasing need for
data-efficient machine learning: the ability to learn in complex domains
without requiring large quantities of data. The types of inductive
biases and generalization properties we use use, the degree of prior
knowledge and side information that is incorporated, and the specifics
of the models and inference approaches, all have significant impact in
this regime.
Data-efficiency has become an increasingly important requirement for
modern machine learning and AI systems. This workshop will discuss the
diversity of approaches that exist and the practical challenges that we
face. There are many approaches that demonstrate that data-efficient
machine learning is possible, including methods that
- Consider trade-offs between incorporating explicit domain knowledge
and more general-purpose approaches,
- Exploit structural knowledge of our data, such as symmetry and other
invariance properties,
- Apply bootstrapping and data augmentation techniques that make
statistically efficient reuse of available data,
- Use semi-supervised learning techniques, e.g., where we can use
generative models to better guide the training of discriminative models,
- Generalize knowledge across domains (transfer learning),
- Use active learning and Bayesian optimization for experimental design
and data-efficient black-box optimization,
-Apply non-parametric methods, one-shot learning and Bayesian deep learning.
The objective of this interdisciplinary workshop is to provide a
platform for researchers from a variety of areas, spanning transfer
learning, Bayesian optimization, bandits, deep learning, approximate
inference, robot learning, healthcare, computational neuroscience,
active learning, reinforcement learning, and social network analysis, to
share insights and perspectives on the problem, discuss challenges and
to debate the roadmap towards more data-efficient machine learning.
3. Key Dates
Paper submission: 1 May 2016
Acceptance notification: 13 May 2016
Final paper submission: 13 June 2016
Workshop: 23 or 24 June 2016
4. Workshop Organizers
Marc Deisenroth (Imperial College London)
Shakir Mohamed (Google DeepMind)
Finale Doshi-Velez (Harvard University)
Andreas Krause (ETH Zürich)
Max Welling (University of Amsterdam)