Projects
The models have been commercialized for Samsung's Korean Smart AC lineup in the first phase, after which they will be deployed for all ACs and will probably control your AC as well!
SHiFT Analytics Framework
SAMSUNG RESEARCH INSTITUTE
My team and I are developing a distributed cloud based framework for analytics and for predicting operations/actions for various non-mobile, smart home devices such as Air Conditioner and Washer to achieve home automation.
We have developed various distributed data processing and filtering modules to process Samsung’s proprietary SHP data. Intra-module parallelization and was achieved using Spark.
I have primarily worked on developing and implementing a tailored procedure for clustering Air Conditioners on the basis of their similarity in usage and on developing and implementing a desired temperature predictor for Air Conditioners using random forest. The sparsity and skewed nature of the data was handled using Joachim's Transformation. A model for predicting mode of the AC was designed using Multilayer perceptron .
Indoor Localization using WiFi Fingerprinting
SAMSUNG RESEARCH INSTITUTE
Here, my colleagues and I developed a method to locate a WiFi enabled smartphone, using stray WiFi signals, using a custom data processing technique with an ensemble of classifiers to determine location. Most of the data was collected using a self developed android application.
IEEE International Conference on Consumer Electronics (ICCE) 2018
Investigation of Information Density Models in Hindi and English Language Production
Indian Institute of Technology
Here, we tried determining the extent to which word order choices are governed in accordance with the Uniform Information Density hypothesis (UID) in Hindi language. We created several metrics for capturing the uniformity in information. We performed various statistical analyses to validate these metrics. Their representative capabilities were measured through ML models trained to distinguish naturally produced sentences from their grammatically correct variants.
Cognitive Modeling and Computational Linguistics (CMCL) 2018
Deutschen Gesellschaft fur Sprachwissenschaft (2017)
Road Irregularity Detector
SAMSUNG RESEARCH INSTITUTE
Here my team and I developed a model, which detects road irregularities using the motion of a vehicle, captured using an IMU of a smartphone. I primarily worked on developing algorithms to use crowd-sourced data to further reduce errors in predictions of the above mentioned model.