Qiong Yang

Qiong Yang

By |

My research program at the University of Michigan (UM) integrates the fields of biophysics, quantitative systems biology, and bottom-up synthetic biology to understand complex stochastic cellular and developmental processes in early embryos.
We have developed innovative computational and experimental techniques in microfluidics and imaging to allow high-throughput quantitative manipulation and single-cell lineage tracking of cellular spatiotemporal dynamical processes in various powerful in vitro and in vivo systems we established in my lab. These systems range from cell-free extracts, synthetic cells reconstituted in microemulsion droplets, presomitic mesoderm (PSM) and progenitor zone (PZ) cells dissociated from the zebrafish tail buds, their re-aggregated 2D and 3D cell-cell communications, ex vivo live tissue explants, and live embryos.
Our current research questions center around the understanding of the design-function relation of robust biological timing, growth, and patterning, how individual molecules and cells communicate to generate collective patterns, and how biochemical, biophysical, and biomechanical signals work together to shape morphogenesis during early embryo development.

Mohammed Ombadi

Mohammed Ombadi

By |

My research focuses on understanding and quantifying climate change impacts on hydroclimatic extremes. From heavy storms and floods to extreme heatwaves and droughts, I study the changing characteristics of these events and their impacts on our daily lives. I use a wide range of data-driven methods such as causal inference, information theory, nonlinear dynamics and machine learning in the analysis of environmental systems. I am primarily interested in using causal inference to obtain new mechanistic insights on the impact of global warming on extreme weather events. My recent work has explored global warming impact on extreme events such as rainfall extremes, using a combination of observations and model simulations. Additionally, recent work has focused on developing new techniques to measure resilience of environmental systems to extreme events, with the ultimate goal of informing mitigation and adaptation strategies to climate change.Map of the Northern Hemisphere showing the projected increase in extreme daily rainfall by 2100 relative to 1950–1979 (the risk ratio). Darker areas are predicted to be more prone to increased rainfall extremes with global warming.

Map of the Northern Hemisphere showing the projected increase in extreme daily rainfall by 2100 relative to 1950–1979 (the risk ratio). Darker areas are predicted to be more prone to increased rainfall extremes with global warming.

What are some of your most interesting projects?

Resilience of Watersheds to Extreme Weather and Climate Events:

Hydrologic watersheds are the fundamental units of the land surface used in the analysis and management of water resources systems. The response of watersheds to extreme events is highly complex and determined by a multitude of factors, including the presence of dams and reservoirs, snowpack, groundwater-surface water interaction, and vegetation cover, among others. One significant knowledge gap in this field of research is how to objectively and unambiguously quantify the resilience and resistance of watersheds to extreme events, such as droughts and floods. Developing metrics to quantify resilience is of utmost importance, particularly in light of the changing characteristics of extreme events due to global warming.

In this project, I employ a wide range of statistical methods to quantify resilience. I then apply methods of machine learning, causal inference, and graph-based techniques to explore patterns of resilience across watersheds worldwide.

How did you end up where you are today?

I originally hail from Sudan, nestled at the heart of Africa. Growing up along the banks of the Nile River, I developed an early fascination with water and its profound connection to humanity. On one hand, it provides people with their needs for drinking, agriculture, transportation, and recreation. On the other hand, an excess or shortage of water often results in devastating natural disasters, such as floods, droughts, and famines. Throughout antiquity, humans have endeavored to regulate rivers by building dams, canals, and various other structures. This deep-seated interest in water sciences and related engineering disciplines led me to pursue a degree in Civil Engineering.

Following my college graduation, I chose to embark on graduate studies to delve deeper into the intricate relationship between climate change and the water cycle. Embarking on an arduous 18-hour flight, I traversed the vast Atlantic Ocean and the continental expanse of the United States to arrive in California, the Golden State. There, I successfully completed my MSc and PhD degrees at the University of California, Irvine, in just under five years. Subsequently, I relocated northward to Berkeley, where I conducted my postdoctoral research, focusing on the impact of global warming on climate extreme events. I come to University of Michigan with a deep research interest in exploring the impact of global warming on extreme events and the resilience of ecosystems to such events. My research group harness recent advances in data science (machine learning, causal inference and information theory) to obtain new mechanistic and predictive insights on these questions with the ultimate goal of informing climate change adaptation strategies.

Kamran Diba

Kamran Diba

By |

My lab is primarily interested in how the brain represents, coordinates, and stores memories. Hippocampal neuronal networks generate an assortment of firing patterns that vary depending on the behavior and state of an animal, from active exploration to resting and different stages of sleep. In our lab’s extracellular recordings from large populations of spiking neurons in rodents, we observe state-dependent temporal relationships between activities at multiple timescales. Recent work in my lab is aimed at understanding what role these unique spike patterns play and what they tell us about the function and limitations of different brain states for memory in healthy and compromised animals. To answer these and related questions, we combine behavioral studies of freely moving, learning and exploring rats, multi-channel recordings of the simultaneous electrical (spiking) activity from hundreds of neurons during behavior, sleep and sleep-deprivation, statistical and machine learning tools to uncover deep structure within high-dimensional spike trains, and chemogenetics and optogenetics to manipulate protein signaling and action potentials in specific neural populations in precise time windows.

Spike times recorded from a population of hippocampal neurons during running on a maze.

Spike times recorded from a population of hippocampal neurons during running on a maze.

What are some of your most interesting projects?

Evaluating the impact of sleep loss on hippocampal replay.
Using unsupervised machine learning to evaluate the temporal structure of hippocampal firing patterns during sleep.

What is the most significant scientific contribution you would like to make?

Understand how the hippocampus serves memory and what role sleep plays in this process.

What makes you excited about your data science and AI research?

The potential for AI models to help explain how the brain works.


Research Highlights

jjpark

Jeong Joon Park

By |

3D reconstruction and generative models. I use neural and physical 3D representations to generate realistic 3D objects and scenes. The current focus is large-scale, dynamic, and interactable 3D scene generations. These generative models will be greatly useful for content creators, like games or movies, or for autonomous agent training in virtual environments. For my research, I frequently use and adopt generative modeling techniques such as auto-decoders, GANs, or Diffusion Models.

In my project “DeepSDF,” I suggested a new representation for a 3D generative model that made a breakthrough in the field. The question I answered is: “what should the 3D model be generating? Points, meshes, or voxels?” In DeepSDF paper, I proposed that we should generate a “function,” that takes input as a 3D coordinate and outputs a field value corresponding to that coordinate, where the “function” is represented as a neural network. This neural coordinate-based representation is memory-efficient, differentiable, and expressive, and is at the core of huge progress our community has made for 3D generative modeling and reconstruction.

3D faces with apperance and geometry generated by our AI model

Two contributions I would like to make. First, I would like to enable AI generation of large-scale, dynamic, and interactable 3D world, which will benefit entertainment, autonomous agent training (robotics and self-driving) and various other scientific fields such as 3D medical imaging. Second, I would like to devise a new and more efficient neural network architecture that mimics our brains better. The current AI models are highly inefficient in terms of how they learn from data (requires a huge number of labels), difficult to train continuously and with verbal/visual instructions. I would like to develop a new architecture and learning methods that address these current limitations.

Cam McLeman

Cam McLeman

By |

My research interests lie in the application of mathematical tools to machine learning models, e.g., using tools from graph theory and stochastic processes to study graphical neural networks, and conversely, the application of artificial intelligence to mathematical proofs, e.g., automated theorem-proving and theorem-generation. With IDEAS, I also work to use more standard applications of machine learning models to solve problems for groups who traditionally lack access to data science expertise.

My doctoral training was in algebraic number theory, and one of the boasts of number theory is that you are required to use tools from every discipline in mathematics to understand all of its facets. This brought me in contact with graph theory both in the abstract and in the applied setting of Markov chains and stochastic processes, and using these ideas to model evolutions of systems in natural settings. Most recently, the dynamic updating of stochastic processes on graphs is very similar in spirit to the training of many models of neural networks, and exploring the symbiosis between these two sets of ideas has been a driver of my recent research.

In IDEAS, we are excited about taking the reams of student and faculty expertise and research at the University of Michigan and using it “for the people” — finding ways of furthering the goals of small businesses or local community groups that do not have the resources to have a data scientist on staff. On the research front, I personally am very excited to see how the study of mathematics evolves as generative AI models meet formal theorem-proving systems.

Stella Yu

Stella Yu

By |

My research lies at the intersection of computer vision, human vision, and machine learning. Visual perception presents not just a fascinating computational problem, but more importantly an intelligent solution for large-scale data mining and pattern recognition applications.
My research has thus three themes.
1. Actionable Representation Learning Driven by Natural Data. I attribute our fast effortless vision to actionable representation learning driven by natural data, where mid-level visual pieces can be reassembled and adapted for seeing the new.
2. Efficient Structure-Aware Machine Learning Models. I view a computational model as dual to the data it takes in; since visual data are full of structures, models reflective of such structures can achieve maximum efficiency.
3. Application to Science, Medicine, and Engineering. I am interested in applying computer vision and machine learning to capture and exceed human expertise, enabling automatic data-driven discoveries in science, medicine, and engineering.


Accomplishments and Awards

Nishil Talati

Nishil Talati

By |

I am a research faculty at the Computer Science and Engineering department at University of Michigan. I work with a group of talented PhD students on computer architecture, compiler techniques, and software engineering. My group focuses on developing novel software and hardware solutions to optimize large-scale data intensive worklods (e.g., graph traversals).

I earned my PhD degree in CSE from University of Michigan, Ann Arbor, USA, master’s degree in EE from Technion – Israel Institute of Technology, Haifa, Israel, and an undergraduate degree in EEE from BITS Pilani, Goa Campus, Goa, India.

Minji Kim

Minji Kim

By |

Communication mechanisms exist inside our cells to regulate gene expression and function. In cells, communication between a gene and sequences that regulate the gene occurs through physical contact. These physical contacts between regulatory element enhancers and the gene promoters may activate gene expression. Our goal is to understand how this communication occurs within the nucleus. In particular, we leverage 3D genome mapping technologies to capture the dynamic chromatin interactions and study gene regulation in the context of developmental biology. Algorithms on graphs allow us to interpret these data generated from high-throughput sequencing assays.

How did you end up where you are today?

I studied electrical engineering, hoping to advance the field of information and coding theory. Instead, I aim to understand the communication mechanisms governing gene transcription.

What makes you excited about your data science and AI research?

Interpreting biological data is often a creative process. I enjoy browsing through the data, connecting pieces, and finding patterns.

Siddhartha Srivastava

Siddhartha Srivastava

By |

My research broadly revolves around extending, specializing, and developing novel ML/AI methods for computational mechanics. My primary focus is data-driven physics-based modeling that utilizes approaches like Variational System Identification and PDE-constrained optimization. I apply these methods for inferring PDE models for complex physical phenomena, for instance, foldings during brain growth, deformation mechanics in soft matter (human tissue and ligaments), and migration and proliferation in biological cells. I also develop graph-based approaches for Machine Learning and NISQ (Noisy Intermediate Scale Quantum) computing. These methods are rooted in classical physics and mathematical analysis but simultaneously developed in concert with real-life experimental data.

Max Li

Max Li

By |

Max’s research interests lies in the design, management, and optimization of large-scale infrastructure systems, focusing on the air transportation system and emerging aerial mobility systems. He is interested in the application of methods applicable to networked systems, especially with resource constraints (e.g., airspace and airport capacity), diverse stakeholders (e.g., passenger-centric, airline-centric), and complex dynamics (e.g., changing temporal behaviors). Max has worked on a variety of data-driven problems related to analyzing flight delays across airport networks, strategic/tactical air traffic management and delay assignments, privacy and routing in drone-based applications, and uncertainty-aware traffic management. He is interested in methods such as graph signal processing and signal processing over non-Euclidean domains, data-driven optimization, mixed-integer/integer/combinatorial programs, resilient network design, and stochastic optimization. Broadly, Max hopes to contribute to a safe, resilient, and efficient air transportation system (inclusive of intra- and inter-city modalities) within the context of a passenger’s (or cargo’s) door-to-door journey.

A flight delay assignment mechanism that does not rely on knowing the exact value of an airline's private per-flight valuation of flight delays. The mechanism requires the system capacities/demand per round, and computes an initial solution. This solution is adjusted by a Coordinating Airline using privatized information from Participating Airlines. The resultant solution is proposed, and any negative public delays incurred is recorded to a ledger. The role of Coordianting/Participating Airlines then rotate depending on the current ledger balance.

Image: A flight delay assignment mechanism that does not rely on knowing the exact value of an airline’s private per-flight valuation of flight delays. The mechanism requires the system capacities/demand per round, and computes an initial solution. This solution is adjusted by a Coordinating Airline using privatized information from Participating Airlines. The resultant solution is proposed, and any negative public delays incurred is recorded to a ledger. The role of Coordinating/Participating Airlines then rotate depending on the current ledger balance.

What is the most significant scientific contribution you would like to make?

Working with air traffic controllers and traffic flow managers to deploy/prototype a congestion management solution/model!

What makes you excited about your data science and AI research?

The potential positive impact on the air transportation system, and helping to make it more efficient and equitable in terms of access.


Accomplishments and Awards