Current Research Projects
As we near the end of Dennard scaling, traditional performance and power scaling benefits primarily derived from semiconductor process technology improvements no longer exist. At the same time, transistor density improvements continue; the result is the “dark silicon” problem in which chips now have more transistors than a system can fully power at any point in time. Constrained by power consumption, hardware acceleration in the form of datapath and control circuitry customized to particular algorithms or applications offers a promising approach to deliver orders of magnitude performance and energy benefits compared to general purpose computing.
To learn more about our ongoing projects in hardware accelerators, click here.
Our group adopts a full stack approach to deep learning. We explore deep learning optimization algorithms, deep neural network design, benchmarking of deep learning systems, hardware acceleration of DNNs, and more. Our targeted problem domains span image classification to speech recognition and machine translation.
To learn more about our ongoing projects in deep learning, click here.
Personalized recommendation is the task of recommending content to users based on their personal preferences. Recommendation algorithms are used pervasively to improve user experience through personalization. Search engines employ recommendation algorithms to order results; social networks, to suggest posts; e-commerce websites, to suggest purchases; and video streaming services, to recommend movies.As their sophistication increases, modern recommendation algorithms use deep learning approaches. Our group adopts a full-stack approach to studying neural personalized recommendation. We characterize the workload, develop open-source infrastructure to study datacenter scale inference, design runtime solutions for inference, and propose novel hardware accelerators.
To learn more about our ongoing project in personalized recommendation, click here.
Over the last couple of decades, the world has seen dramatic advancement in computing technology causing a Cambrian explosion in consumer devices, communication technology, and data centers. These advances enabled a plethora of applications including AI, robotics, e-commerce, social media, cloud storage, and scientific computing. However, the increasing demand for ubiquitous compute has resulted in increasing environmental impacts.
To learn more about our ongoing project in sustainable computing, click here.