Pushpendra Parmar

Project

•     Stochastic Neural Networks for Cryptocurrency Price Prediction
  Published on IEEE [PAPER] Inspired by the Random-Walk model on financial markets, we induced layer-wise randomness into the observed feature activations of neural networks to simulate market volatility. We trained the Multi-Layer Perceptron and LSTM models for Bitcoin, Ethereum, and Litecoin.
  
•     Enhanced Neural Arithmetic Logic Unit Modified Neural Arithmetic Logic Unit to learn basic functions like addition and multiplication including advanced functions like exponents and logarithms to generalize it over very large value.
  
•     Pattern Recognition in Computer Language Extracted patterns in computer languages such as C, Java and Assembly and make model sample the code intangible for humans at first glance, nevertheless unable to compile or execute.
  Extracted pattern includes knowledge to close a bracket when opened, not to use return statement when function declared void.
  
•     WhatsTrack Exploited WhatsApp's feature of providing when one is online to track their WhatsApp activity even if one has hidden their last seen.
  
•     Image Reconstruction Application of Graph Network, CNN and RNN to reconstruct image from incomplete input.
  Plausible accuracy was expected and achieved for grayscale images. Further extension similar model on speech synthesis from incomplete parts produces unreal output for human ears.
  
•     16 Bit ProcessorDesigned a 16 Bit processor having minimal functionalities of RISC V architecture.