Ismet Dagli

I am a computer science PhD candidate at Colorado School of Mines advised by Prof. Mehmet Belviranlı. Before joining Mines, I completed my BS in Computer Engineering at Boğaziçi University and received guidanced from Prof. Levent Akın

My research focuses on the field of diversely heterogeneous accelerators. Particularly, I’ve worked on increasing performance, energy efficiency, and utilization of heterogeneous systems for autonomous vehicles and drones. Throughout my research, I have quite worked on edge SoCs (NVIDIA Jetson Family, Qualcomm Development Kits, Raspberry PIs). I’ve also been working other system architecture related research topics, such as analytical modeling to predict the performance of heterogeneous systems, memory contention to minimize the performance degredation.

Please refer to my CV for more details about my education and experiences.

Selected News

  • (Sep’24) Our abstract is accepted at CWIDCA (a workshop colocated with MICRO’24).
  • (May’24) Selected as MLCommons ML and Systems Rising Star.
  • (Apr’24) Our work is accepted to ICS’24. The paper proposes a novel scheduling technique for cyber-physical systems with heterogeneous processing units under real-world constraints.
  • (Apr’24) Our work is accepted to ARMA’24. The paper analyzes cutting forces of mining machines through several ML algorithms in terms of execution time, energy consumption, and accuracy. This work is done with collaborators from Mining Engineering at Mines.
  • (Mar’24) Our work is accepted to HCW’24, colocated with IPDPS’24. The title of the work is “A Runtime Manager Integrated Emulation Environment for Heterogeneous SoC Design with RISC-V Cores”
  • (Mar’24) I have selected as finalist and won the third place in CGO’24 SRC. The title of the work is “H-EYE: Holistic Performance Modeling for Diversely Scaled Systems”.
  • (Feb’24) Our work is accepted to DATE’24. The title of the work is “Constraint-aware resource management for cyber-physical systems”.
  • (Nov’23) Our work is accepted to PPoPP’24. The title of the work is “Shared Memory-contention-aware Concurrent DNN Execution for Diversely Heterogeneous System-on-Chips”.
  • (Sep’23) Our work is accepted to 2024 SME Annual Conference Expo. The title of the work is Enhancing Safety Using Energy-Efficient Machine Learning Algorithms Through Prediction of Rock Type and Cutter Wear. This work is done with collaborators from Mining Engineering at Mines.
  • (May’23) Our work is accepted to FRAME’23, colocated with HPDC’23. The title of the work is “Contention-aware Performance Modeling for Heterogeneous Edge and Cloud Systems”.
  • (Feb’23) I’ve accepted to Google ExploreCSR program as a mentor. I’ve been mentored two great students, Jason Lee and Cesar Lopez Hernandez.
  • (Nov’22) I’ve won the third place in MICRO’22 SRC. The title of the work is “HaX-CoNN: Heterogeneity-aware Execution of Concurrent Deep Neural Networks”.
  • (July’22) Our work is accepted to DAC’22. The title of the work is “AxoNN: Energy-Aware Execution of Neural Network Inference on Multi-Accelerator Heterogeneous SoCs”.
  • (August’21) Our work is accepted to RSDHA 2021. The title of the work is “Exploration of Multi-Accelerator Neural Network Inference in Diversely Heterogeneous Embedded Systems”.
  • (August’21) Our work is accepted to ICCAD’21. The title of the work is “Automated generation of integrated digital and spiking neuromorphic machine learning accelerators”.

Selected publications

  • Justin Mcgowen, Ismet Dagli, Neil Dantam, Mehmet Belviranli. Title: Scheduling for Cyber-Physical Systems with Heterogeneous Processing Units under Real-World Constraints. in 38th ACM International Conference on Supercomputing (ICS), 2024
  • Ismet Dagli, Mehmet E. Belviranli, Title: Shared Memory-contention-aware Concurrent DNN Execution for Diversely Heterogeneous System-on-Chips, Proceedings of the 29th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming, PPoPP’24. [Paper Link]
  • Justin Mcgowen, Ismet Dagli, Neil Dantam, Mehmet Belviranli. Title: Constraint-aware resource management for cyber-physical systems. in Design, Automation and Test in Europe Conference (DATE), 2024. [Paper Link]
  • Ismet Dagli, Andrew Depke, Andrew Mueller, Sahil Hassan, Ali Akoglu, Mehmet Belviranli. Title: Contention-aware Performance Modeling for Heterogeneous Edge and Cloud Systems. in 3rd workshop on Flexible Resource and Application Management on the Edge (FRAME) colocated with HPDC’23. [Paper Link]
  • Ismet Dagli, Alexander Cieslewicz, Jedidiah McClurg, Mehmet E. Belviranli, Title: AxoNN: Energy-Aware Execution of Neural Network Inference on Multi-Accelerator Heterogeneous SoCs, 59th ACM/IEEE Design Automation Conference(DAC), 2022. [Paper Link]
  • Ismet Dagli, Mehmet E. Belviranli, Title: Exploration of Multi-Accelerator Neural Network Inference in Diversely Heterogeneous Embedded Systems; 2021 IEEE/ACM Redefining Scalability for Diversely Heterogeneous Architectures Workshop (RSDHA) colocated with SC’21 [Paper Link]