Publications

• Papers
  1. Vijay Daultani, Subhajit Chaudhury, Ishizaka Kazuhisa, Convolutional Neural Network Layer Reordering for Acceleration, Synthesis And System Integration of Mixed Information, SASIMI 2016.
     Abstract: here
  2. Vijay Daultani, Ohno Yoshiyuki, Ishizaka Kazuhisa, Sparse Direct Convolutional Neural Network, International Symposium on Neural Networks, ISNN 2017.
     Abstract: here
     Download PDF: here
• Patents
  1. Vijay Daultani (NEC), Information processing apparatus, information processing method and storage medium storing program.
     Filed: June 2016, Application No: PCT/JP2016/002686.
     Documents: here.
  2. Vijay Daultani (NEC), Information processing method and device for neural network.
     Filed: June 2016, Application No: PCT/JP2016/068741.
     Documents: here.
  3. Vijay Daultani (NEC), Improved sparse convolutional neural network.
     Filed: Nov 2016, Application No: PCT/JP2016/081973.
     Documents: here.
• Presentations
  1. Vijay Daultani, Association Rules.
     Download Presentation: here
     Note: Presentation is made using Sozi and Inkscape.
     Date: March 2017.
  2. Vijay Daultani (Rakuten), Recurrent Neural Networks.
     Download Presentation: here
     Video: here
     Date: Oct 2017.

Research Projects

• Professional
  1. Convolutional Neural Networks
     Experimenting convolutional neural network on NEC’s SX super computer architecture, inventing ways for optimal hardware utilisation for acceleration. Researching new algorithms for convolutional operation of CNN.
  2. Instruction Selection in Compiler
     Invented a new technique for optimal instruction selection of a compiler and filed a patent for it.
  3. Biometrics
     Profiled, analysed biometrics application, modified data structures and proposed algorithms to suite and accelerate on underlying hardware architecture.
  4. Elevator Advisory
     Giving a optimal advice to the users whether to choose the conventional elevator system or stair case or escalator, by calculating and comparing the service time for each transportation medium like elevator, and staircase.
• Personal
  1. Data Visualization using Seaborn
     I created a Notebook for exploration, visualisation and analysis of abalone datasets using python, pandas, matplotlib and seaborn.
     Code: here
  2. Parallel Computation using MPI
     As a course project for CSV 880: Special Module in Parallel Computation, I implemented following two assignments.
     • Assignment 1: Averaging based computations on a matrix using block-cyclic distribution.
     • Assignment 2: Broadcasting on a 2D Torus using a single spanning tree.
     Code: here
  3. Pintos
     Pintos is a simple instructional operating system framework for the 80x86 processor architecture from Stanford University. As a course project for Resource management in Computer Science I implemented following modules in the framework of Pintos
     • Loading and running user programs.
     • Virtual Memory management.
     • File system.
  4. CGRA
     Use LLVM to compile the computation intensive kernels in C into assembly code. A parser to identify data dependencies then parses the intermediate representation of code generated by LLVM and a DAG representing the data dependency is constructed. After which Instruction Legalisation, Scheduling and Mapping on CGRA is performed.
     Code: here
  5. NUCA
     Analysed energy breakdown of Static and Dynamic Non-Uniform Cache Access (NUCA). I measured the power consumption and collected heat generation patterns of NUCA architecture. Also interpreted inter banks link data access patterns.
     Presentation: here