MLD Labs and Research Groups
Our faculty leads research efforts on the most important issues facing machine learning today. We have included explanations of many of the efforts below, but new initiatives are always underway in our department.
You can also check out Core Faculty Directory to check out your PhD to get detailed ideas for research initiatives on the ML@CMU blog. Thesis Archives.
AI Social Decision Making Research Institute (AI-SDM)
Aarti Singh, Bryan Wilder, Hoda Heidari, Roni Rosenfeld, Jeff Schneider, Aaditya Ramdas
AI Societal Decision Making (AI-SDM) develops AI to enhance human decision-making in social areas such as public health and disaster management that require complex and rescuing decisions, taking place under uncertain, dynamic, resource-constrained circumstances, and occupying perceptions of people of risk, trust, and equity. It uses data-driven recommendations through bandits and reinforcement learning algorithms, adaptive control trials, counterfactual inference, and personalized interventions to enable public health and emergency management organizations to identify socially accepted effective and ethical policy decisions. He also trains and skills the workforce at the intersection of AI and social sciences through targeted engagement with partners in high schools, community colleges, universities, businesses and governments. For more information, please visit the AI-SDM website.
Database Groups
Christos Farautus
CMU's database groups focus on high-performance database architectures, multimedia and data mining. We participate in many interdisciplinary efforts and work closely with many other groups at CMU. Find out more on our website.
Delphi Research Group
Roni Rosenfeld, Ryan Tibshirani, Larry Wasserman, Valerie Ventura, Alex Reinhardt, Bryan Wilder
Epidemiological predictions are highly needed for decision-making by public health authorities, commercial and non-profit agencies, and the general public. Delphi Group has developed several award-winning prediction technologies based on statistical machine learning and other technologies. Our long-term vision is that weather forecasts are universally accepted and useful epidemiological predictions as they do today. We have participated in and successfully participated in all epidemiological prediction challenges that have been organized by the US government to date. For more information, please visit the Delphi Group website.
Sailing Lab
Eric Singh
Our main research interests are to solve machine learning and statistical methodologies, and large-scale computing systems and architectures to solve problems including automation, inference and decision-making in artificial, biological and social systems, involving high-dimensional, multimodal, dynamic worlds. Find out more on the Sailing Lab website.
Select a lab
Jeff Gordon
Our long-term research goal is to develop smarter ways of learning and decision making. Towards this goal, we will explore ways to design, analyze, understand and control complex real-world systems. Our research spans the overall spectrum, from the theoretical foundations to real-world applications.
Orton Lab
Artur Dubrawski, Barnabas Poczos, Jeff Schneider
Our main research focuses on investigating useful data structures and algorithms, as well as creating interesting statistical and learning methods that can be applied to large amounts of data. We are interested in computer science, mathematics, statistics, and practical applications of our work that underlie these structures. We work closely with food safety analysts, public health agencies, nuclear safety experts, equipment fleet managers, social network personnel, astrophysicists, biologists, chemists, pharmaceutical companies, exploration companies and robotics scientists. Find out more on our website.
Neurostat
Robkas
Researchers around the world who investigate neural networks in the brain are trying to answer detailed questions using large but noisy data sets, creating new challenges in statistics and machine learning. Neurostat groups contribute, particularly by focusing on how to reliably identify collaborative neural activity in multiple brain regions. Find out more on our website.
Machine learning fundamentals, computing theory, algorithm game theory
Nina Vulcan
This group develops the fundamentals and practical algorithms of important modern learning paradigms. These include interactive learning, distributed learning, learning representation, lifelong learning, metallairing, noise-resistant or robust learning, and neural architecture search. We also explore the fundamentals and applications of data-driven algorithm design, as well as the design and analysis of algorithms on realistic instances (also known as the worst case scenario). This group explores computational and data-driven approaches in game theory and economics, as well as the theoretical aspects of calculation, learning theory and games in multi-agent systems.
Next-generation Statistical Machine Learning
Pradeep Ravikumar
It focuses on two fundamental aspects of next-generation statistical machine learning. With graceful AI, we want to learn more about “graceful” models beyond average case performance. And Scrappy AI, who wants to learn models under resource constraints. In elegant AI, the group is engaged in researching explainable AI (XAI), robust ML, hostile ML, reliable/resilient ML in an undistributed (OOD) test environment, and statistical game theory. Under Scrappy AI, the group is engaged in research on structural causal models and directed graphical models, incorporating domain knowledge (what DARPA calls “Third Wave AI”) and self-monitoring learning.
Learn from people
Nihal Shah
Many applications include many items ratings by people for many items, but each item is evaluated only by a subset of people, and each person only evaluates a subset of items. This distributed nature of assessment brings many problems of bias and inequity. We emphasize addressing and affecting these issues through basic theoretical analysis, algorithm design, and real-world experiments. Our focus application is peer review, the backbone of scientific research. Our work also applies to other applications such as employment, admission, crowdsourcing, healthcare, online assessment and recommendations, and peer grading.
Brain ML
Leila Wave
How does the human brain organize information while achieving complex tasks such as understanding sentences and visual scenes? How is brain activity from humans performing everyday tasks related to the activation of AI algorithms that process the same information? It helps you understand how information is processed in different brain regions and how these regions communicate with each other, along with AI representations. This alignment also helps to improve your understanding of AI algorithms and suggest ways to improve them. By studying how individual brains differ from the perspective of the information they represent, we can also predict behavioral differences and propose tools that can help us understand neurological and mental illness.
Machine learning, the fundamentals of deep learning theory
Yuanzhi Li
Our work develops the mathematical foundations of current deep learning models and explores why it is more effective than traditional learners. We also create a more efficient principles approach to learning multilayer neural networks and try to understand new phenomena related to inviscous optimization landscapes in deep learning, particularly algorithmic regularization.
catalyst
Tianqi Chen
The rapid advances in ML models and ML-specific hardware make it increasingly difficult to build efficient and scalable learning systems that can take full advantage of the performance features of modern hardware and runtime environments. Today's ML systems rely heavily on human efforts to optimize the training and deployment of ML models on a specific target platform. Unlike traditional application domains, learning systems must address the ever-growing complexity and diversity of machine learning models, hardware backends, and runtime-time environments. Our response to this unique challenge in ML systems is Catalyst (CMU Automated Learning Systems Group), a collaborative research group across the fields of machine learning, systems, programming languages, and computer architecture. Our mission is to build ML algorithms and learning systems that leverage the mathematical and statistical properties of ML calculations and automate cross-stack optimization with co-design systems, hardware, and ML algorithms. Find out more about working on our website.
Almost Correct Machine Intelligence (ACMI) Lab
Zach Lipton
Building an intelligent system that can be applied in the real world requires more than predictable. Driving decisions require causal insights. Reliability requires a reliable robust model under clear assumptions. Deploying data-driven technology in society requires accounting for the complex dynamics and feedback loops that mediate this interaction. Cooperating with social design such as equity requires philosophically consistent treatment. ACMI Labs researches core machine learning methods, applications in healthcare, and social impact. We try to address these outer loop questions, but we leverage the breakthroughs of representational learning to address the diverse raw data sources that deep learning can access.
Cell Organizer
Robert Murphy
The team, led by Bob Murphy, faculty honorary in the Computational Biology Department, combines image-derived modeling methods with active learning to build a continuous, comprehensive model of protein localization. To fully grasp the localization of all proteins within a cell and how they change under various conditions is a challenging task given the human body has 100 cell types, and the tens of thousands of proteins expressed in each cell type, as well as the one million conditions (including the presence of potential drugs or disease-causing mutations). Automated microscopy is useful by enabling rapid acquisition of large numbers of images, but with automation it is not possible to directly determine the localization of all proteins of all cell types under all conditions. For more information, please visit the program website.
