Our Research

Systems for AI

  • Designing new frameworks for deep neural networks
  • Large scale AI/ML design with high utilization of multiple GPUs
  • Maximizing memory efficiency of AI models while remaining statistical efficiency
  • Accelerating training for deep neural network

AI for Systems

  • Learned index in short read alignment:
    Accelerate DNA sequencing by introducing learned index and solving the exact match search problem for efficient seeding
  • Microservice auto-scaling:
    Auto-scale microservices for optimal resource utilization while meeting the service level objective
  • 5G resource scheduling:
    Dynamic resource scheduling to optimize the use of network resource and wireless spectral efficiency under 5G, 6G environments

AI + Video

  • HTTP adaptive streaming + neural super-resolution[USENIX OSDI’18] (First paper from KAIST in the history of OSDI)
  • Live streaming + neural super-resolution [ACM SIGCOMM’20]
  • Mobile + neural super-resolution [ACM MobiCom’20’]
  • Codec + neural super-resolution [In-progress]
  • Scalable neural super-resolution [In-progress]

Cloud Systems

  • TEE-based network system security [IEEE/ACM ToN 2020]
  • Designing global QoS algorithm for distributed file system
  • Leveraging internal parallelism of Samsung key-value SSD