ML | SKKU IRIS Lab

Neuromorphic/In-Memory Computing

Tue, 22 Dec 2020 22:12:44 +0900

Our research focuses on memory-based deep learning techniques, including SRAM/ReRAM-based deep learning processing-in-memory (PIM) and Flash memory based robust deep learning inference techniques.

Sensor Systems with Integrated Deep Learning

Sun, 22 Dec 2019 22:12:44 +0900

To maximize the performance and efficiency of deep learning based data processing, we are currently exploring sensor platform design optimized for deep learning by leveraging the interactions between a sensor platform and a DNN.

Deep Learning Model Compression and Acceleration

Sat, 22 Dec 2018 22:12:44 +0900

Deep neural networks (DNNs) are widely adopted at the IoT edge devices to enable more intelligence. However, the biggest challenge is their storage demand and computational complexity. As DNNs contain a large number of synaptic weights, the memory demand is a key challenge for application of DNNs, especially for memory-constrained platforms such as mobile systems. Another bottleneck of DNNs is the computation demand, mostly due to their large number of multiplications in convolution layers. Therefore, reducing the storage and computation demand of neural networks is critical, specifically, to support in-field and on-chip training and inference. To enable deep learning based intelligent multimedia processing at the IoT edge devices with limited hardware resource, the research targets for neural network design with lower storage and computation demand.

Edge-Host Collaborative Deep Learning

Sat, 22 Dec 2018 22:12:44 +0900

Based on the trade-off study between energy, accuracy, and throughput of an edge platform with an embedded DNN, we are currently exploring collaborative DNN inference between the edge and the host to optimally utilize the available bandwidth and energy.

Variation-Robust Deep Learning

Sat, 22 Dec 2018 22:12:44 +0900

One of the challenges of deploying deep neural networks in sensor platforms is the variations in the input images; structural noise (adversarial images), inherent random noise (image perturbation), input scene variability, and weight value errors. To enhance the robustness of deep learning inference to the variability, we are designing variation-robust deep learning techniques.