Priya Panda and Marynel Vazquez Win Amazon Awards


SEAS faculty members Priya Panda and Marynel Vazquez have been named winners of the 2019 Amazon Research Awards.

The awards are part of a grant program that provides up to $80,000 in cash and $20,000 in promotional credits to academic researchers investigating topics across 11 focus areas: computer vision; fairness in artificial intelligence; knowledge management and data quality; machine learning algorithms and theory; natural-language processing; online advertising; operations research and optimization; personalization; robotics; search and information retrieval; and security, privacy, and abuse prevention.

Vazquez, assistant professor of computer science, received an award for her project, “Improving Social Robot Navigation via Group Interaction Awareness,” described as follows: 

The project is about robot navigation in human environments. For many years, social navigation systems relied on explicit modeling of costs for path planning, including social costs for enabling robots to respect pedestrian’s personal space. These systems often became hard to tune due to the interplay of different kinds of objective functions, and the challenge of mathematically formalizing social conventions for navigation. More recently, alternative mobile navigation systems have relied on machine learning methods to model relevant navigation costs from data. This project will focus on advancing the latter type of approaches by making them more interpretable than traditional black-box end-to-end navigation methods, and making them more explicitly reason about robots' social context, including pedestrian interactions.

Panda, assistant professor of electrical engineering, won for “Adversarial Robustness with Efficiency-Driven Optimization of Deep Neural Networks,” described as follows: 

Despite achieving super-human performance on a variety of perception tasks, Deep Neural Networks (DNNs) have been shown to be 'adversarially' vulnerable. A DNN can be easily fooled into misclassifying an input with slight changes of pixel intensities. This vulnerability severely limits the deployment and potential safe-use of DNNs for real world applications such as self-driving cars, healthcare monitoring systems, etc. Our goal in this research is to use efficiency-driven techniques, such as quantization, pruning etc. for leveraging adversarial robustness. Such efficiency driven techniques have been conventionally used for simplifying the hardware requirements to yield low-power, energy-efficient and memory optimized DNNs. Using energy-aware techniques to harness adversarial robustness in deep learning is an under-explored area of research. Our goal is to imbibe such hardware-aware characteristics to design robust, yet, efficient neural algorithms. 

Recipients of the award can use more than 150 Amazon public data sets. They are also encouraged to publish their research results, give presentations at Amazon offices worldwide, and release related code under open-source licenses.