CS Faculty Receive Google Faculty Research Awards

4/11/2019 By Allie Arp, Coordinated Science Laboratory

Google Faculty Research Awards are given to a select group conducting computer science and engineering research. This year two Illinois CS faculty were chosen, as well as three ECE faculty who are CS affiliates.

Written by By Allie Arp, Coordinated Science Laboratory

Each year through Google Faculty Research Awards, Google funds a select number of researchers working in computer science and engineering. This year, two Illinois Computer Science faculty received the honor, along with three Electrical and Computer Engineering faculty who are Illinois CS affiliates. We spoke with all five about their research and the expected inpact of the Google Faculty Research Awards.


Sarita AdveRichard T. Cheng Professor of Computer Sciences

What does it mean to you to win a Google Research Faculty Award?

Sarita Adve, CS Professor and Director of IMMERSE
Sarita Adve, CS Professor and Director of IMMERSE

 

It means a lot! More than the financial support, I value the connections with Google researchers. It is important to me that my research makes an impact on real products and influences long-term thinking on how to design systems. I appreciate that Google appoints specific technical liaisons for each awardee, providing direct contact for discussion and amplification of research ideas. It is also very exciting for my students.

What effect will this have on your research?

The general problem we are addressing is that of efficient memory hierarchies for emerging heterogeneous systems. There are two ways in which we expect this award to have a unique impact on our research agenda. (1) We would like to influence emerging standards for coherence and consistency interfaces and we hope deeper connections with Google will help with that. (2) Our most recent work has focused on memory hierarchies for system-on-chip designs. We would like to extend it into the arena of distributed systems and the cloud where we hope again to get insights from Google’s dominance of this space.

Can you give me a brief description of your current research?

With the end of Moore's Law, parallelism and application specialization will be the key enablers of high performance and energy efficiency. As the number of accelerators increases, there will be an increasing need to integrate the accelerator memories more tightly with each other and with the rest of the cache-based memory hierarchy and communication interface. Current systems are ill-suited for such integration, resulting in significant inefficiencies. We are working on a memory hierarchy and communication architecture that is specialized for workloads and platforms of interest to Google (e.g., machine learning in the cloud), providing far more energy-efficient data communication than the state of the art.


Svetlana Lazebnik, Associate Professor

What does it mean to you to win a Google Research Faculty Award?

Associate Professor Svetlana Lazebnik
Associate Professor Svetlana Lazebnik

 

I'm very grateful to Google for their continued support (I was co-PI on a 2015 Google Award, with Julia Hockenmaier as PI), and excited to have the opportunity to collaborate with my Google technical contact. 

What effect will this have on your research?

This award will give my group the resources to develop novel multi-task learning methods that can progressively grow neural networks to accommodate an increasing number of tasks. With the development of efficient neural network architectures and engines that can even run on mobile phones, it is becoming easier than ever to deploy neural networks in everyday applications. Consumer devices such as the Amazon Alexa and Google Pixel already feature virtual assistants that can process an image to produce class labels for objects within it, caption the image, and read text within the image. As such devices and applications proliferate, there is a need to develop methods that can augment the knowledge and skills of existing neural networks while keeping the overhead low. As the device learns to perform new tasks, it should retain its previously learned models, ideally without needing to store large and separate networks for each specialized task. Developing such capabilities is the goal of my project.

Can you give me a brief description of your current research?

In addition to the multi-task learning research described above, the most active research area in my group is learning of models for joint image-language understanding. From robotics to human-computer interaction, there are numerous real-world tasks that would benefit from practical, large-scale systems that can identify objects in scenes based on language and understand language based on visual context. We develop neural models for specific image-language scenarios, from grounding of natural language phrases in image regions to automatic image captioning, visual question answering, and visual dialogue.


Haitham Al-HassaniehAssistant Professor of Electrical and Computer Engineering, CS affiliate


What does it mean to you to win a Google Research Faculty Award?

Assistant Professor Haitham Al-Hassanieh
Assistant Professor Haitham Al-Hassanieh


It is an honor to receive this award. I am grateful for Google and my research collaborators at Google.

 

What effect will this have on your research?
The award provides support and funds as well as an opportunity to collaborate with Google on addressing the wireless connectivity challenges in VR. This can help support the research and make sure it heads in the right direction toward a real world impact.

Can you give me a brief description of your current research?
My current research focuses on designing wireless links that can replace these cables. Unfortunately, the problem is not as easy. Current wireless technologies like WiFi do not have sufficient bandwidth to support the very high data rates required for VR.  Our approach is to use millimeter wave wireless technology which operates at a very high frequency (around 60 GHz) and can provide us with wireless links that support multi-Giga bits/sec data rates. Millimeter waves, however, bring in a whole slew of challenges when it comes to signal attenuation, blockage, mobility, scalability, etc. Our goal is to address these challenges in order to enable multi-user untethered VR.


Jian HuangAssistant Professor of ECE, CS affiliate
 

What does it mean to you to win a Google Research Faculty Award?

Assistant Professor Jian Huang
Assistant Professor Jian Huang


The award provides my research group the opportunity to work closely with Google and its platform and storage team on memory and storage systems. It also indicates that our constant effort on memory/storage research is increasingly being recognized by our industry partners.

 

What effect will this have on your research?
My research group enjoys building practical systems for real-world applications and hardware. The collaboration with Google will not only help us identify more interesting research problems but also facilitate our technology transfer to the industry for larger impact.

Can you give me a brief description of your current research?
In my research group, we conduct systems and architecture research for a variety of computing platforms spanning from IoT devices to mobiles, and to large-scale data centers. Currently, we are especially interested in developing non-volatile memory technologies for these platforms.


 

Associate Professor Shobha Vasudevan
Associate Professor Shobha Vasudevan

Shobha VasudevanAssociate Professor of ECE, CS affiliate

 


What does it mean to you to win a Google Research Faculty Award?
Winning the Google FRA is a good validation that the problems we are working on are relevant and impactful. I am thankful that I was selected to receive the award.

What effect will this have on your research?/ Can you give me a brief description of your current research?
This award was granted for a proposal to verify large scale machine learning systems, including deep neural networks. Machine learning systems are ubiquitous in the world we live in. Many of them are being used in mission and safety critical systems like autonomous vehicles. Erroneous outputs or misclassification errors can be very costly to life and property in these situations. Our research is to provide an efficient, scalable means to check machine learning systems. We are excited to continue this line of work with this award.


Share this story

This story was published April 11, 2019.