Contributor
Akhilesh Raj
Biography
Having contributed to the field of control and estimation, I was keen to understand what machine learning can do to this branch of engineering aimed at control applications. I am currently pursuing my Ph.D. in the Department of Electrical and Computer Engineering at the Vanderbilt University with focus on:
• Composition of surrogate models using edges to reduce the computation/ communication load on cloud hosting.
• Using physics guided machine learning to increase the fidelity of the learned models distributed across a network.
• Performance and Power optimization in HPC nodes using Reinforcement Learning (Argonne National Lab).
I have 9 publications in various international conferences and journals, with over 33 citations. These first authored works span a wide range of research areas which include distributed estimation, control, machine learning, computer vision etc. I also have a book chapter related to learning and estimation titled as Computational Intelligence: Theories, Applications and Future Directions ‑ Volume II, based on the research conducted at IIT Kanpur.
Currently in my Ph.D., I am using my ideas from control system design and dynamics modeling to replicate a test bed for a complex system spatially hosted across a network. I work towards distributed hosting of surrogate models to accommodate a complex system with more reliability on the edge computations for the surrogate training, while keeping the cloud as a central controller or the host. Last summer, my internship with ANL, has provided me with an opportunity to collaborate with them in their future works. Therefore, a part of my PhD dissertation will be dealing with the HPC node power optimization algorithms. Using an RL agent,
trained using Proximal Policy Optimization Algorithm, we already achieved a significant reduction in the power consumption of HPC nodes at negligible cost on the time of execution and the related publication is under review for the ICCPS 2023 conference. We plan to carry forward this work with more complex scenarios under consideration.
• Composition of surrogate models using edges to reduce the computation/ communication load on cloud hosting.
• Using physics guided machine learning to increase the fidelity of the learned models distributed across a network.
• Performance and Power optimization in HPC nodes using Reinforcement Learning (Argonne National Lab).
I have 9 publications in various international conferences and journals, with over 33 citations. These first authored works span a wide range of research areas which include distributed estimation, control, machine learning, computer vision etc. I also have a book chapter related to learning and estimation titled as Computational Intelligence: Theories, Applications and Future Directions ‑ Volume II, based on the research conducted at IIT Kanpur.
Currently in my Ph.D., I am using my ideas from control system design and dynamics modeling to replicate a test bed for a complex system spatially hosted across a network. I work towards distributed hosting of surrogate models to accommodate a complex system with more reliability on the edge computations for the surrogate training, while keeping the cloud as a central controller or the host. Last summer, my internship with ANL, has provided me with an opportunity to collaborate with them in their future works. Therefore, a part of my PhD dissertation will be dealing with the HPC node power optimization algorithms. Using an RL agent,
trained using Proximal Policy Optimization Algorithm, we already achieved a significant reduction in the power consumption of HPC nodes at negligible cost on the time of execution and the related publication is under review for the ICCPS 2023 conference. We plan to carry forward this work with more complex scenarios under consideration.
Presentations
