BEST Undergraduate Researcher Competes in the Amazing Race

Aparna Dhinakaran and her brother Eswar Dhinakaran were one of teams competing for the 2020 series of Amazing Race; first showing Wednesday night October 14, 2020. See: Bay Area siblings live out their childhood dream by competing on season 32 of ‘The Amazing Race’.

Excerpt: Eswar and Aparna attended UC Berkeley and became software engineers. But Aparna hadn’t given up on her dreams of competing on the show. She waited until her brother turned 21, the minimum age to participate, and filed the application.

Caption: Siblings from Fremont and Berkeley, Eswar Dhinakaran (L) and Aparna Dhinakaran (R), are seen participating in a challenge during the season 32 premiere of “The Amazing Race.” The show returns Wednesday, Oct. 14 on CBS.

After completing her B.S. degree at UC Berkeley along with several exciting jobs, Arpana joined a PhD program at Cornell focused on Computer Vision, advised by Serge Belongie. She is currently on leave to found Arize AI which is an ML Observability platform. Theyhelp make machine learning models successful in production! 

More at:

Aparna worked as an undergraduate researcher in the Berkeley Energy and Sustainable Technologies Lab as part of our Smart Lighting project. She works with two graduate students (Chandrayee Basu and Julien Caubel ) in our controls group to create a control scheme to take into account the level of natural light available and user preferences in order to control the lighting levels in a room so as to decrease power consumption and increase energy cost savings.

Aparna and her research team developed an indoor lighting inverse model as a piecewise linear function of the minimum number of sensed parameters: window light levels and adjoining dimmable lights’ statuses, discretized by sub-hourly sun angles at a given time of the day. Their inverse-model system was a combination of light sensors and software that allowed for reduced sensing in comparison to current light sensor systems while retaining prediction accuracy. Through this virtual sensor system at our test-beds, they have seen that our system retains prediction accuracy while using a smaller number of sensors than other available systems, effectively reducing deployment cost by 40% and energy use by 60%. If just 10% of buildings without smart lighting are retrofitted, we project that there will be a resultant California energy savings of 60,000 gigawatt-hours a year.

For the smart lighting project, Aparna worked on both the hardware and software components. For the hardware, she designed and monitored the NASA Ames Sustainability Base testbed to efficiently allow the network of wireless light sensors to send data via a multi-hop network to a local server to implement computations for light level predictions and optimal sensor placement. On the software end, she assisted with the inverse model function and developed a functional database using SQLite for linear regression models employing unsupervised machine learning clustering algorithms. Aparna was co-author of a paper that was a finalist for best student paper: Basu, C., Chu, B., Richards, J., Dhinakaran, A., A.M. Agogino, Martin, R., “Affordable and Personalized Using Inverse Modeling and Virtual Sensors”, Proceedings of the SPIE Smart Structures/NDE 2014, 9-13 March 2014.