Akshata Patel

• Generated a simple word problem dataset with single variable equations and implemented a Bidirectional GRU-LSTM Seq2Seq model with Attention to identify the corresponding math equation.
• Implemented a Transformer model to improve predictions and accuracy measured by the Corpus BLEU score.

• Performed temporal and spatial analysis of humidity, temperature and surface pressure and examined its effect on precipitation.
• Implemented and trained a Generative Adversarial Network (GAN) to produce a well-generalized distribution of precipitation in order to better capture the variability as opposed to a Dense Neural Network.

• Performed exploratory data analysis and created visualizations on the human trafficking and migration dataset collected from the CTDC collaborative.
• Created an interactive site to present the findings and patterns in human trafficking across different countries of the world using D3
• Link of the interactive Component: https://bl.ocks.org/akshatapatel/raw/97326ed42c9d5e29ca6986e87d656146/

• Implemented realignment and segmentation of brain fMRI images in MATLAB.
• Performed dimensionality reduction using Principal Component Analysis in Python.
• Classified the brain state images into one of the four states - rest, finger movement, lips movement or foot movement using Support Vector Machine and testing using K-fold cross validation using Scikit-learn in Python.

• Implemented Iris Localization, Image Normalization and Enhancement on the CASIA-Iris V1 dataset.
• Extracted the iris patterns/features and implemented Fisher Linear Discriminant for dimensionality reduction and iris matching.
• Output: Correct Recognition Rate – 75%

Contributed to https://edav.info/ by creating a walkthrough page for tidying the MASS:biopsy dataset in R using tidyr:gather() and dplyr functions such as : select, mutate, filter, arrange, summarize and group_by.

Made a Android application for Traffic Updates for Ahmedabad city. This application uses data from Google API's and traffic inputs from the users of the application to provide a picture of the traffic conditions in different parts of the city.

• Built a flask app for a system that recommends YouTube channels based on the user inputs such as product description, product category and video category for a product that is to be promoted.
• Used the YouTube Data API to extract features of the latest videos dynamically.
• Dockerized the training and prediction services, which used the word embeddings of the features and channel popularity to find the top 5 relevant channels.

• Analyzed restaurant reviews from the SemEval 2014-Task 4 data set and created normalized vectors from words using the word2vec function from the gensim library.
• Implemented a Recurrent Neural Network with the word vector input for aspect extraction in TensorFlow to predict output aspect vectors.

• Cleaned the restaurant reviews from the yelp data set by using Python libraries such as NumPy and BeautifulSoup.
• Analyzed the sentiment of the reviews (positive or negative) using lexicon-based classifier and the Bigram model of words using Scikit-learn library.
• Visualization of the accuracy of different classification techniques such as random forest, support vector machine and Naive Bayes in Orange tool.