Machine Learning at Khoury College of Computer Sciences
Enabling computers to create knowledge from data in any discipline and at vast scales
The development of algorithms that work in ways inspired by the human brain has led to a revolution in how computers work. Machine learning helps computers gain knowledge and the ability to solve problems for which they were not explicitly programmed — for instance, processing visual information and understanding natural language. These approaches are revolutionizing what computers can do — even what they are.
Research in this area also takes in the human connection, asking questions about the implications of using computer learning approaches in specific contexts, such as health care, scientific discovery, robotics, and addressing cybercrime.
Empowering intelligent computer systems to autonomously solve tasks
Machine learning approaches have impacts both everyday and profound, driving the work of Khoury College researchers. Many recommendation engines that users encounter online are powered by machine learning, as are social media platforms, search engines, and journalism — with news stories written by computers ever more prevalent. Understanding how these programs work and the bias and privacy issues they can raise can provide policy and practice guidance.
Machine learning is revolutionizing health care through medical imaging analysis in which algorithms for analyzing medical test results promise to improve the speed and accuracy of things like tumor recognition.
These everyday impacts of machine learning are driving the work of Khoury College researchers, work that also includes improved predictions for weather, better real-time GPS services, and more secure financial networks through improved fraud detection.
Sample research areas
- Machine learning theory
- Deep learning
- Graphical models
- Learning-to-rank
- Semi-supervised learning
- Data mining
- Statistical pattern recognition
- Open source software
- Computer vision
- Image processing
Meet researcher Rob Platt
Rob Platt talks about his work in exploring machine learning models that is advancing the development of self-learning robots.
Current project highlights
This research group combines statistics and machine learning to analyze data from mass spectrometry, a powerful technique used to study molecules in living organisms.
This research project tackles the challenge of analyzing massive datasets (big data) efficiently. It focuses on designing algorithms that can run on multiple computers working together, distributed computing in the cloud.
Khoury researchers, and colleagues, are engaged in a National Science Foundation-funded effort to shed light on the “black box” of complex AI systems. This is vital, because even as AI is changing society, scientists cannot explain its predictions–or ultimately how it works. This research focus has the potential to help build more trustworthy and beneficial AI applications in the future.
Recent research publications
Boosting Multitask Learning on Graphs through Higher-Order Task Affinities
Authors: Dongyue Li, Haotian Ju, Aneesh Sharma, Hongyang R. Zhang
Khoury researchers and colleagues are researching how to improve how AI models learn to perform tasks with multiple steps (e.g., identifying communities in a social network). Traditional methods may fail due to complex relationships between tasks. Grouping tasks may allow models to learn more effectively and achieve better results.
Boundary-Aware Uncertainty for Feature Attribution Explainers
Authors: Davin Hill, Aria Masoomi, Max Torop, Sandesh Ghimire, Jennifer Dy
This research addresses the issue of unreliable explanations generated by AI models, particularly for complex decisions. Khoury researchers and colleagues are developing new methods, potentially providing a way to make AI’s reasoning easier to assess, leading to better AI tools for users.
Deep Bayesian Active Learning for Accelerating Stochastic Simulation
Authors: Dongxia Wu, Ruijia Niu, Matteo Chinazzi, Alessandro Vespignani, Yi-An Ma, Rose Yu
Complex simulations, for instance those that track infectious disease, can be very slow and resource intensive. This research aims to create faster and more focused ways to run such simulations by using deep learning approaches.