Research Projects

Stopping Criteria for Active Learning

• Description: We are researching principled methods for detecting when to stop the collection of additional training data for building machine learning systems.
• Related Publications: [1], [2], [3], [4], [5], [6], [7]

Forecasting the Performance of Machine Learning Systems

• Description: We are researching how to forecast the performance of machine learning systems.
• Related Publications: [1], [2]

Active Learning to Improve Cost-Effectiveness of Annotation

• Description: We are researching how to selectively sample data for annotation with the aim of maximizing the benefits received from annotation effort.
• Related Publications: [1], [2], [3]

Computer-Assisted Translation Technologies

• Description: We are researching how to improve CAT (computer-assisted translation) technologies so that professional translation can be performed faster and with higher quality.
• Related Publications: [1]
• Related News Articles: New CASL Technology Makes Human Translators More Efficient, Algorithm Set to Take Translation Memory Systems by Storm

Computational Deep Understanding of Language

• Description: We are researching how to computationally model nuanced semantics of language such as modality and impacts on machine translation.
• Related Publications: [1], [2], [3]

Data Cleaning

• Description: We are researching how to better automatically or semi-automatically detect and repair errors in data.
• Related Publications: [1], [2], [3], [4]

Machine Learning for the Life Sciences

• Description: We are researching how to use machine learning to help make new discoveries in the life sciences.
• Related Publications: [1]
• Related News Articles: UMD Researchers Discover a Way That Animals Keep Their Cells Identical

Filtering Social Media Content for Relevance to Social Unrest

• Description: We are researching how to use machine learning and natural language processing techniques to filter social media content for relevance to social unrest.
• Related Publications: [1]

Translation Lexicon Induction for Historically Unwritten Languages

• Description: We are researching how to induce translation lexicons for historically unwritten languages through bridging loanwords.
• Related Publications: [1]

Using global constraints, reranking, and fast approximations of the Hungarian algorithm to improve cognates detection

• Description: We are researching how to use global constraints, reranking, and fast approximations of the Hungarian algorithm to improve cognates detection.
• Related Publications: [1]