Ayşegül Dündar

About

Ayşegül Dündar’s research lies at the intersection of efficient deep learning hardware, domain adaptation, and robotic vision. Her most impactful work, “Embedded Streaming Deep Neural Networks Accelerator With Applications” (122 citations), pioneered high-throughput, power-efficient architectures for deploying deep convolutional neural networks (DCNNs) on resource-constrained platforms like autonomous robots and mobile devices. She further advanced this direction with a hardware-accelerated real-time DCNN implementation on a mobile coprocessor (43 citations), demonstrating practical feasibility for embedded systems. In domain adaptation, Dündar introduced “Domain Stylization: A Fast Covariance Matching Framework” (43 citations), a method that bridges the gap between synthetic and real-world data—critical for training models in robotics and autonomous driving without costly manual labeling. Her earlier work on clustering learning for robotic vision (2013) proposed an unsupervised technique to efficiently compute network filters, reducing training time and parameters. Across these contributions, Dündar has demonstrated a consistent focus on making deep learning practical, efficient, and deployable in real-world visual perception systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Embedded Streaming Deep Neural Networks Accelerator With Applications

   122 citations · 2016
2. 2
   An efficient implementation of deep convolutional neural networks on a mobile coprocessor

   43 citations · 2014
3. 3
   Domain Stylization: A Fast Covariance Matching Framework Towards Domain Adaptation

   43 citations · 2020
4. 4
   Clustering Learning for Robotic Vision

   3 citations · 2013

Key Collaborators

Contact & Links