Deep Learning Generative models can learn and predict the turbulence wake behind a cylinder
Deep Leaning Generative model can learn and predict the turbulence wake behind a cylinder with different geometry and shape
Time-dependent generative models can learn and predict the time-dependent heat diffusion by just giving the boundary condition!

Machine Learning Physical and Engineering Phenomena

MAIL at CMU focuses on the problems at the interface of Mechanical Engineering and data science. We will use the state of the art deep learning and machine learning algorithms and tools to learn, infer and predict the physical phenomena pertinent to mechanical engineering.

Examples of such phenomena are multiphysics transport, fluid mechanics and heat transfer. Specifically, with the exponential growth of sensory data and internet of things, data driven modeling of complex physical phenomena is critical in order to engineer the resilient infrastructures. For more information, you can see our recent papers here:


Our lab is developing algorithm for learning from sparse sensory data. The applications of such algorithm is huge given the rise of internet of the things and cloud data. These machine learning algorithms can predict and fill the blanks in sparse spatio-temporal points.

Machine Learning Material Discovery

Another area of interest in MAIL is to accelerate material discovery using machine learning tools. Since it is prohibitively expensive to use experimental and computational tools to search for novel material for energy applications, the search process can be greatly accelerated by using deep learning technology. To this end, we will both develop and apply these tools to find optimal materials for energy and health applications.

Artificial Intelligence and Robotics

We are combining deep learning and reinforcement learning (RL) together and apply it to the robot intelligence. One of the applications of focus in MAIL is to bring intelligence to robots. Currently, robots are using traditional control techniques to perform tasks which in complex environment might not be robust. Recent observations we had with deep reinforcement learning shows promises to give robots intelligence and more human like behavior. Examples of these behaviors can be decision making under constraints, creativity, and persistence.

We believe future robots are able to acquire these characteristics of human. To this end, our lab explores AI algorithms which can add such characteristics and behaviors to next generation of robots.

We also apply AI to drones and UAVs. In recent years, Air-delivery and it's associated technologies are growing. The complexity and the uncertainty of the environment that the drones are flying require them to be smart, intelligent and solve challenges on the fly. Our lab is developing real time machine learning algorithms to give drones these capabilities.

Read our paper on creativity of robots and AI here:

Read our paper on effect of sparse reward:

Computational Bio-engineering and AI

At MAIL, we are combining molecular dynamic simulations, machine learning and statistical learning to understand and predict the properties and interactions of bio-molecules. To be more specific, we are focused on two types of problems:

1. Interactions and recognition of bio-molecules such as DNA with synthetic materials: We perform MD simulations to generate the trajectories . We use statistical learning techniques to learn the collective variables. This is an interesting area where statistical mechanic and machine learning coincides. You can read our recent paper here and more in publications:

2. To understand small molecule and protein interaction. We are dealing with super high dimensional time-series data coming from MD simulations. Reducing the dimension of the data, and learning the reaction coordinates help us understand many fundamental biophysical and physiological mechanisms for protein conformation and dynamics.