Bio

Parth Raghav is a curious and ambitious rising senior at KR Mangalam World School, Vikaspuri. He finds himself interested in data science and artificial intelligence, which makes him fascinated by pretty much everything ranging from robotics to functional genomics. For the past two summers, he has been giving CS101 and Introduction to Research and Development courses to more than 100 students in his native village. He has been a global Google Science Fair regional finalist (top 18 in the CS category worldwide) and IRIS National Fair national finalist (top 15 in the CS category PAN India). Parth has worked on numerous projects, like the following:

• Computer-Aided Discovery of a Novel Non-Invasive Synergic MicroRNA Vector Signature for Non-Small-Cell Carcinoma Diagnosis
• ULTRON: A Novel Self-Assembling Modular Robotic System
• PAERS: Post-Accident Emergency Response System
• Capturing Biomarker Synergy via Quantification of miRNA–mRNA Interactions
• Non-Parametric Statistical Test for Discovering Biomarkers That May Be Elevated in Only Specific Subsets of Data
• Early Diagnosis of Small-Cell Carcinoma Using PNN Monitoring microRNA–mRNA Interactions
• Detection of Synergetic Associations among Histone Modifications for Early Diagnosis of Pancreatic Adenocarcinoma Using Probabilistic Neural Network and Novel Kinase Vector Model
• Optimized Face Detection and Recognition via Computer Vision Using HSV Templates
• Violence Detection in Video Streams Using Spatiotemporal Features
• Classification Machine Learning Model for Identification of Linguistic Patterns in the Ancient Transcripts

Project: Handwritten Text Generator

My project is to make a handwritten text generator using the current Wolfram Language. I hypothesize that the service would require a person to provide sample handwritten text (in the form of a scan or image upload), and the service would identify the intrinsic patterns in the handwriting of a certain user. The service not only maps the characters but also identifies different kinds of connections between character tuples. What makes handwriting different from a usual font face are its irregularities and a certain amount of deformation. Services build a custom dataset for training the convolutional neural network, inputs as character features and outputs as elastically distorted characters. Now, as the neural network trains, the service iteratively tests the characters. This goes on until the neural network is fully trained. Essentially, by doing this, we have all different characters with all possible forward and backward projections in as many as possible mapped distortions. The process would take less than three minutes for the handwriting analysis (and is mandatory for each user to do at least once, because the service generates output based on persistent data, and is not based on a central dataset). Once done, it shall remain persistent and is not required again. Now the user could simply provide text on the microsite, and the service would return a corrected and concatenated text. Essentially, the service interpolates the characters as per the demand, and corrects or joins the projects to make the text continuous. The model could perform better than existing generative models in terms of special characters, number generation and sequence length.