Research Areas
The division conducts fundamental and applied research in the following topics:
Machine Learning and Applications
Focusing on fundamental research, including AI robustness, adversarial machine learning, anti-spoofing, domain adaptation and federated learning, and applied research in application areas such as biomedical sciences, biometric authentication, computational social science and cybersecurity.
People
Wael Abd-Almageed - Jose Luis Ambite - Aram Galstyan - Shrikanth Narayanan - Jay Pujara - Filip Ilievski - Muhao Chen - Xuezhe Ma - Greg Ver Steeg - Ke-Thia Yao - Mohammad Rostami - Mohamed Hussein - Leonidas Spinoulas - Hengameh M. Dastjerdi
Projects
- Commonsense reasoning
Multi-modal open world grounded learning and inference - CORAL
Combined Representations for Adept Learning (DARPA Learning with Less Labels) - QUASAR
QUantum Assisted Sampling for MAchine LeaRning (DARPA) - Secure Heterogeneous Learning Federation with Information‑Theoretic Guarantees (DARPA)
- BATL
Biometrics Authentication with a Timeless Learner (IARPA) - LR2
Learning Robust Representations (DARPA) - AI2AI
AI Investigating AI (Keston Award) - DARPA Cooperative Secure Learning SHELFI
Secure Heterogeneous Learning Federation with Information-Theoretic Guarantees
Natural Language Processing
Focusing on low resource machine translation, multilingual representation learning, transfer learning, dialogue, decision-making, question answering, summarization, ontologies, information retrieval, text decipherment.
Visit Website
People
Jon May - Muhao Chen - Xuezhe Ma - Xiang Ren - Ulf Hermjakob - Elizabeth Boschee - Marjorie Freedman - Jerry Hobbs - Scott Miller - Nanyun (Violet) Peng - Ralph Weischedel
Projects
- CLEAR
Cross-Lingual Event and Argument Retrieval (IARPA BETTER) - CORAL
Combined Representations for Adept Learning (DARPA Learning with Less Labels) - ELICIT
A System for Extracting and Organizing Causal Information (DARPA Causal Exploration) - ELISA
Exploiting Language Information for Situational Awareness (DARPA LORELEI) - EvidxExtraction
Evidence Extraction Systems for the Molecular Interaction Literature (NIH R01) - LESTAT
Learning Event Schema Temporally and Transmodally (DARPA KAIROS) - MICS
Machine Intelligence from Common Sense (DARPA MCS) - SARAL
Summarization and Domain-Adaptive Retrieval Across Languages (IARPA MATERIAL)
Knowledge Graphs
Using AI and machine learning techniques to construct and exploit large-scale knowledge bases and to induce taxonomies from data. Notable applications include probabilistic models for scientific reproducibility, incorporating extractions from scientific articles and scientific networks of citation and reference, and business knowledge graphs characterizing innovation and competition using web data and regulatory filings.
Visit Website
People
Jay Pujara - Pedro Szekely - Filip Ilievski - Ke-Thia Yao - Muhao Chen - Hans Chalupsky - Craig Knoblock - Jose Luis Ambite
Projects
- KGTK
The Knowledge Graph Toolkit (DARPA) - Table Linker
Linking tables to knowledge graphs (DARPA, Novartis) - Building a knowledge graph for food security (DARPA)
- Datamart
Creating the largest publicly available knowledge graph to power data-driven models (DARPA) - Karma
A data integration tool - Knowledge graphs for business (DARPA)
- Scoring scientific research
Developing automated techniques for evaluating scientific claims (DARPA) - Semantic modeling
Automatically building semantic models of sources (DARPA) - CSKG
The commonsense knowledge graph - Knowledge Graph Completion with Transferable Representation Learning (NSF)
- Novartis-USC
Scaling Up Data FAIR-ificaton
Scientific Data Analysis and Discovery
Using interactive knowledge capture, intelligent user interfaces, semantic workflows, provenance, and collaboration; large-scale data integration and analysis of biomedical data (including sensor, environmental, neuroimaging, clinical and genetic data) and (paleo)climate data.
Visit Website
People
Yolanda Gil - Deborah Khider - Jose Luis Ambite
Projects
- WINGS
A semantic workflow system that assists scientists with the design of computational experiments - MINT
Integrating scientific models - DISK
Automating the discovery of scientific models - LinkedEarth
Paleoclimate data and analysis - Scientific/geoscience paper of the future
Encouraging scientists to publish papers with the associated products of their research - P4ML
A phased performance-based pipeline planner for automated machine learning - ASSET
A sketching project to accelerate scientific workflows - OntoSoft
A software metadata registry to describe scientific software in a user-friendly manner - Organic data science
Resolving science processes through an open framework that facilitates participation - OPMW-Prov
Tracking the provenance of scientific experiments and their executions - NHGRI
National Human Genome Research Institute- PAGE
Population Architecture using Genomics and Epidemiology Coordinating Center
- PAGE
- NeuroBridge
Automating the discovery of scientific models - NIMH
National Institute of Mental Health- NRGR
Repository and Genomics Resource
- NRGR
Multi-modal Understanding
Including image and video understanding for deepfake detection, visual misinformation identification, identifying manipulated scientific literature and multimedia analysis, face recognition, biometric anti-spoofing, and robust AI; table understanding to automate exploitation of millions of tables on the web focusing on automatic layout detection, semantic modeling, table retrieval, table summarization, entity linking, and fact-checking.
People
Wael Abd-Almageed - Jay Pujara - Pedro Szekely - Muhao Chen - Mohamed Hussein
Projects
- Table Understanding
Understanding the structure and semantics of tables (DARPA) - DiSPARITY
Digital, Semantic and Physical Analysis of media integRITY (DARPA)
Common Sense Representation and Reasoning
Using cognitively-inspired computational paradigms for evaluating commonsense AI (including those based on large-scale language models) to create and solve new challenge tasks based on logical axioms and numeracy; human-centric dialog agents that maximize metrics of human utility alongside algorithmic utility in task-focused dialogs; game-theoretic simulators for poker, Monopoly, and wargames that enable refinement and evaluation of theories of novelty for general AI agents.
Visit Website
People
Mayank Kejriwal - Jay Pujara - Filip Ilievski - Muhao Chen - Xiang Ren
Projects
- Commonsense reasoning
Multi-modal open world grounded learning and inference - CSKG
The commonsense knowledge graph
Computational Social Science
With emphasis on structure detection and pattern matching in unusual complex systems with hidden information (e.g., human trafficking, dark money networks); large-scale, contextualized social media analysis (e.g., in the context of natural disasters) including analysis involving non-verbal tokens such as emojis; computational social science methods for quantifying socio-demographically segmented impacts of COVID-19 on wellbeing, technological inequity, and vaccine hesitancy; applied AI in industrial applications, such as e-commerce.
People
Emilio Ferrara - Kristina Lerman - Fred Morstatter - Goran Muric - Keith Burghardt
Projects
- Polarization in the context of COVID-19
The pandemic has exacerbated echo chambers, paving the way for the rampant spread of misinformation. - Science of growth
This project explores the impact of growth on various human behaviors, such as how the growth of cities impact infrastructure, how message boards increase interactions, or the growth of institutions impact collaborations. - DARPA Influence Campaign Awareness and Sensemaking (INCAS)
- Early Detection of Influence Indicators with Machine Intelligence (EDIFICE)
Project to detect influence campaigns in foreign social media - Universal Population Segmentation and Characterization Algorithms for OnLine Environments (UPSCALE)
Project to analyze influence campaigns in foreign social media
- Early Detection of Influence Indicators with Machine Intelligence (EDIFICE)
- VENICE
VErifyiNg Implicit Cultural modEls: Infer cultural causal relationships; store in queryable knowledge graph. Verify relationships via real-world interviews with people in foreign country.
AI Fairness
Detecting and mitigating bias, robustness against adversarial attacks, identifying cultural values, polarization and misinformation, forecasting, and crowdsourcing. Notable case studies include a study of gender bias in 19th century English literature using natural language processing (NLP) methods and a study of how state-of-the-art named entity recognition approaches systematically fail to identify female names.
People
Fred Morstatter - Kristina Lerman - Jay Pujara - Keith Burghardt
Projects
- Auditing fairness of AI systems
Many off-the-shelf AI tools have societal biases embedded within them. These projects focus on identifying and measuring the impact of these biases. - Improving fairness in data
Improving fairness can benefit from being model-agnostic, which allows for the best models to be applied to data of interest. This project explores what methods work best for various types of data. - Adversarial Attacks on Fairness
Adversarial attacks attempt to intentionally skew the impact of a machine learning system. We investigate the efficacy of such attacks in the context of fairness.