Feroz Ahmad's Picture

Feroz Ahmad

Software | NLP @ Adobe || INTP

LinkedIn

About Me

I'm a Software Development Engineer at Adobe Systems, India.

Technology drives me since my childhood. Technology improves quality of life, facilitates an organisation to accomplish loads of work and generate great amount of revenue, & does tasks which humans cannot. I'm extremely passionate about intelligent systems powered by fundamental concepts of computer science and electrical engineering.

My research interests include Deep Learning, Natural Language Processing, Computer Vision, Evolutionary Computing, and Internet of Things.

Publications

Research Projects

Surveillance and Tracking  Computer Vision | Distributed Systems.

Object detection and tracking research has largely focussed on improving performance over contrained subspaces, largely restricted to single camera scope. Motivated by enhanced tracking needs in recent times, we worked on near-real time path retrieval over distributed geospaces.Used unsupervised deep learning techniques to generate compact component based feature descriptors.Deployed the same over a multinode Apache Spark cluster.Worked with the Spark Streaming module to obtain real-time processing.Obtained processing time of 10 ms for 300 queries using 6 component feature vectors over a multinode cluster.

Indoor Positioning SystemResearch Project

Developing effective Solution for Indoor Positioning of a node in a predefined indoor geo‐subspace. Comparing localization techniques used for positioning indoors and generalising their areas of usability by simulating in an indoor geo‐fenced environment.
Project Demo Video

Content Based Image Retrieval  Computer Vision

Minor course project for the semester of fall 2016(Our Team Top Scored with 180/200). I worked on hashcodes based generation of features for image classification. Fine tuned an alexnet after adding a latent layer for feature descriptor characterisation that was then fed into a supervised multiclass classifier. Trained the model on VOC 2012 dataset and studied the system performance by tweaking various parameters including hashcode vector lengths to analyse tradeoff between accuracy and computational complexity
Obtained a 72.8% mAP precision.

Dappa Computer Vision

Image Processing | Deep Learning Inspired by a research paper by Google Inc.,designed our own DCNN model to detect objects in video streams. Employed morphological operations and transformations for feed stabilization and preprocessing. Trained the model specifically for cars by using a popular dataset hosted by UT-Austin and picking off CCTV feeds from the internet.Obtained Compared the model performance with more traditional cascading techniques.

Sentiment Analysis on Hindi Text NLP

This project employs natural language processing for tracking the mood of the public about a particular product or topic, which is written in Hindi Text. I have used Bag of Centroids model after KMean Clustering to obtain a advanced version of Bag of Words Model. Employed Google's revolutionary Word2Vec word embedding model to obtain vectors for Hindi words. The biggest challenge faced was generation of Hindi Labelled Dataset for supervised learning. Libraries used : Numpy, Scipy, Sklearn

Other Projects

SHAKTI - Safety from Harassment, Kidnapping and eve Teasing Interface
Android & Django Hackathon First Prize [email protected]

A user application focussed on sending SOS messages without any cellular network at the point of distress. The applications employs Connectionless Datagram Packets to communicate distress/SOS messages to the near pears.
Project Demo Video

MIST

Competitive Quizzing Platform MIST
PHP SQL Used in production

A Web Application developed for Online Quizzing Event - MIST - hosted by IEEE-DTU, focused on optimized multimedia delivery for a flicker‐free experience. Enjoyed by 1,500 users, producing 65,900 page views . The application was in production for more than 1 week.

View Google Analytics Report
Source Code

Online Stock Market Simulator
Finance & Django Development Used in production

A Web Application Simulator developed for Virtual Online Stock Market - Bulls n Bears - hosted by IEEE-DTU. The application was in production for a span of 3 days.

Technologies used : Python, Django, SQLite3, AngularJS

View Google Analytics Report
Django Source Code and Backend

RaspMe-Home Automator
Raspberry Arduino ESP8266 Internet of Things

Minor course project for semester of Fall 2015. Obtained the Best Project Award. Developed a prototype to facilitate wireless mobile enabled control of appliances in domestic setting.Employed a multi-server architecture, where a global control server interacted with local servers hosted on a raspian system.A client side application was used to issue commands for manipulation of various devices characterised by binary signals.

Checkout GitHub Page

DressSocial

DressSocial
Android & Django Hackathon First Prize @HackEnvision2016

Wardrobe manager and recommender consumer-end application to support smarter buying among customers with a social platform. Used a supervised classifier to characterize similarity of clothing in a wardrobe. Designed and trained our own Deep Convolution Neural Network model to classify clothes into 15 different categories to aid automated recognition. Grand prize winner at HackEnvision 2016.

Checkout GitHub Page

Assignment Submission and Evaluation System
Database Management System (SQL)

A PHP web application that allows ease of sharing tutorials and assignments by the instructor, notifies all the students enrolled for that particular course and allow them to upload their submissions online. The application focuses on ease of evaluation system for instructors. Application also caters the ease for admin, for tasks such as addition of more entities like Students, Teachers and Courses. Submitted as semester Database Systems Project. Technologies used : PHP, MySQL, Bootstrap, JQuery

Checkout GitHub Page
Project Report

Wireless Controlled Bot
RF Transmission Wireless Control

A hobby project performed to control a simple bot wirelessly using RF module (Tx/Rx at 434 MHz). Technologies used : Motor IC : L293D , RF Transmitter and Reciever

Demo Video