Areas of Interest: Units
Areas of Interest: Units
Areas of Interest: Units
• Deep learning | Computer Vision | Natural Language Processing | Machine Learning | Data Science
TECHNICAL SKILLS
• Tools: MATLAB, Simulink,Version, ROS, LaTeX
• Languages: C++, Python,R,SQL
MAJOR PROJECT:
• Objective :Visual Question Answering:
Guide:Prof Biplab Baneerjee(Centre of Studies in Resource Engineering(CSRE) Department),IIT Bombay
◦ Objective:Develop a deep learning based model capable of extracting Semantic discriminative features for
image and question and combine Visual and textual image features to guide answer prediction
◦ OngoingWork: Training a deep learning model using CNN(UNET) architecture to get semantic features of
significant image region and using RNN(LSTM) for extracting textual features and answer prediction.
◦ Future work: Introduce feature wise attention mechanism to learn cross feature wise attention between
image and question modalities
◦ Determination of ARX,ARMAX Model and ARMAX using recursive least square estimation for Single
Board Multi-Heater System using MATLAB System Identification ToolBox.
◦ Implementation of Decentralized PI Controller, Normalized Decoupling,Inverted Decoupling(Internal Model
Control) by analysing Relative Gain Array(RGA),Condition Number(CN) and Niederlinski Index(NI).
◦ Design of linear Quadratic Optimal Controller(LQOC) for perfect set point tracking and Regulation with
specified Control Limits
• Car Object Detection
◦ Objective:
Implemented Car object detection using pretrained MobileNet Architecture with additional Convolu-
tional layers and achieved an accuracy of 99.8 with MAE of 2.72 and loss of 0.02
• Semantic Image Segmentation for CityScape Image Paris Kaggle Dataset-
◦ Objective: Implemented RNN and LSTM Models on Google Stock Price Prediction Kaggle dataset.
* ’tanh’ activations and ’Adam’ Optimizer was used in the design with ’METRICS’ accuracy
* Applying LSTM and RNN model to improve the accuracy of training
• LSTM Text Generation-
◦ Objective: Designed and trained LSTM RNN on Newyork Times Comments and headlines dataset to train
a text generation language model which can be used to generate news headlines.
* Performed tokenization of every text document in the dataset to create sequence of tokens for predic-
tion of sequence of words using LSTM Model which includes ’softmax’ activations with categorical
crossentropy loss and ’adam’ optimizer
RELEVANT COURSES
• Machine Learning for Remote Sensing-I • Introduction to Probability and Random Processes
• Applied Predictive Analytics(Data Analytics,Deep
Learning,Predictive Maintenance) • Optimization
ONLINE COURSES
• Deep learning Specialisation | Deeplearning.ai(Coursera)
Neural Networks,Hyperparameter Tuning,Regularization,CNNs,Sequence Models
• Machine Learning | Standford University(Coursera)
Classification,Linear Regression,Decision Trees,KNN,SVM
• Google Data Analytics | Google(Coursera)
Data Analytics,SQL,R,Tableau