1. Cognitive Information Agents:
Effective Learning in the Wild
Eric Nyberg, Professor &
Director, Master of Computational Data Science Program
“Architecture and applications to support intelligent, natural interaction
with all kinds of information in support of complex human tasks.”
• Extended Configuration Description (ECD):
Specification language to describe space of
analytic configurations for a task [1]
• Configuration Space Exploration (CSE):
Evaluation and selection of best-performing
analytic configuration(s) for a task [1,2,3]
• Phased Ranking Models: Rank outputs of
any multi-phase, multi-strategy system
based on the features of the derivation
paths that produced them [4]
• Automatic Source Expansion: Multi-faceted
machine reading to improve in-task
performance on a specific topic [5];
pioneered in Watson [6]; trained on
human-labeled relevance judgments
Architecture
Automatically build and
execute analytic solutions
Perform
1
Specification of required
analytic input/output types,
desired information sources,
example dataset.
Learn Reflect
Sample Applications
2
Train
Measure
Proactively evaluate
task performance,
analyze errors, propose
learning tasks
Bioinformatics Question Answering (BioQA): Document and passage retrieval which can be automatically
optimized for new datasets (applied to TREC Genomics, CLEF and and corporate sponsor datasets)[2,3]
Question Answering for Decision Support (QUADS): Automatically learn how to leverage QA systems to support
complex human decision-making with multiple decision factors (for gene target prediction and product ranking)[7]
Team
1. Garduno, E., Yang, Z., Maiberg, A., McCormack, C., Fang, Y. and E. Nyberg (2013). “CSE Framework: A UIMA-based Distributed System for Configuration Space Exploration”, Proceedings
of the 3rd Workshop on Unstructured Information Management Architecture, International Conference of the German Society for Computational Linguistics and Language Technology.
2. Yang, Z., Garduno, E., Fang, Y., Maiberg, A., McCormack, C. and Nyberg, E. (2013). “Building Optimal Information Systems Automatically: Configuration Space Exploration
for Biomedical Information Systems”, Proceedings of the ACM CIKM Conference.
3. A. Patel, Z. Yang, E. Nyberg, and T. Mitamura (2013). “Building an Optimal QA System Automatically Using Configuration Space Exploration for QA4MRE”, Proceedings of CLEF 2013.
4. Liu, R. and Nyberg, E. (2013). “A Phased Ranking Model for Question Answering”, Proceedings of the ACM Conference on Information and Knowledge Management.
5. Schlaefer, N. (2012). Statistical Source Expansion for Question Answering, Ph.D. Thesis, Language Technologies Institute, School of Computer Science, Carnegie Mellon University.
6. N. Schlaefer, J. Chu-Carroll, E. Nyberg, J. Fan, W. Zadrozny, D. Ferrucci (2011). “Statistical Source Expansion for Question Answering”, Proceedings of the ACM CIKM Conference.
7. Z. Yang, Y. Li, J. Cai, and E. Nyberg (2014). “QUADS: Question Answering for Decision Support”, Proceedings of the ACM SIGIR Conference on Information Retrieval, 2014.
3
Subject Matter Experts (SMEs)
Analyst’s
Information
Need
Configure
Optimize
Automatically execute
learning tasks, update
models, KBs, etc.
Machine Learning Agents
• Targeted Machine
Reading
• E-R Extraction
• Set Extension
• Clarification Dialogs
• Type/instance
knowledge
• Concept learning
Crowdsourcing
• Type instance labeling
• New feature extraction
• Relevance judgments
Rui Liu
Ph.D. Candidate
Phased Ranking Models
Leo Boytsov
Ph.D. Candidate
BioQA, Semantic Retrieval
Hugo Rodriguez
Ph.D. Candidate
Question Generation
Di Wang
Ph.D. Candidate
Source Expansion
Zi Yang
Ph.D. Candidate
CSE, BioQA, QUADS
Avner Maiberg
MLT Candidate
ECD, CSE, BioQA
Eric Nyberg
Team Leader
http://oaqa.github.io/