SURAJ JHA

Thermal Foresight Engine: (Paper: IJCNN ‘21) to predict & mitigate device overheating issues. Trained an LSTM to predict a smartphone's future skin temperature with avg. error of <0.7 degrees. Deployed it on Android devices using tf-lite, as an AI-powered Android service. Commercialized in Samsung smartphones, used by Samsung's gaming & 5G teams. Ambient Temperature Estimation: (Patent submitted). Trained & feature engineered a deep neural network with avg error <1.2 degrees. Deployed it on android devices as a service. Performance Boost Engine supporting profiles for multiple use-case scenarios for heavy tasks. Served as a platform for boosting in-house services. Commercialized in Samsung smartphones. Controlled android kernel tunables to reduce file encryption & decryption time by 28%. Gyroscope bias reduction for camera stability. Devised an algorithm using exponential decaying with sliding-window technique to remove noise & bias in gyroscope readings. Deployed it on Android devices as a library. Developed an api for sensitivity controlled camera stability. Samsung Adaptive Battery (SAB): classifies application into 3 buckets: “active”, “working set” & “restricted”. Resolved app’s bucket persistence issue by penalizing false positives for active buckets & changing optimizer from SGD to Ada-delta so that learning rate adapts more effectively to shifts in user preferences. Commercialized in Samsung smartphones. Mentored a team of interns in research projects as part of Samsung's industry-academia collaboration program, PRISM.

Developed an AI-powered user behavior modeling system using LSTM based architecture to optimize device performance by accurately predicting the next app to be launched by the user.

Developed virtual drums played using gesture recognition.

Course work: OOP, OS, Neural Networks & Fuzzy Logic, Linear Algebra, Calculus, Optimization.