Saran Ahluwalia

Saran Ahluwalia

My primary interests are in applying statistical learning, graphical models and unsupervised machine learning in service of - amongst others - healthcare services, quantitative finance and real estate economics. I hope to continue to marry my prior experience with my passion for the democratization of knowledge through open source development

Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Pages

Posts

Derivation of Support Vector Machines

less than 1 minute read

Published:

Recently, I am experimenting with different mechanisms for deriving decision functions for multiclass classification problems. I have spent some time reading and experiementing with examples from the following references:

Natural Language Processing with Deep Learning

less than 1 minute read

Published:

I recently completed Stanford’s course on natural language processing. This course opened my eyes to different representaions of speech and linguistics.

Sutton and Barto Part 3

less than 1 minute read

Published:

I am currently reading Sutton and Barto (reference below). Along the way I decided to recreate certain experiments cited in each book chapter. This particular example includes the Blackjack problem definition.

Sutton and Barto Part 2

less than 1 minute read

Published:

I am currently reading Sutton and Barto (reference below). Along the way I decided to recreate certain experiments cited in each book chapter. This particular example includes the example from Figure 6.2.

Sutton and Barto Part 1

less than 1 minute read

Published:

I am currently reading Sutton and Barto (reference below). Along the way I decided to recreate certain experiments cited in each book chapter. This particular example includes the Blackjack problem definition in Example 5.1. Along with it is a visual with varying numbers of steps taken.

Repucall - A new means to characterize phone number reputation

less than 1 minute read

Published:

Just wrapped up my second semester of post-graduate studies in mathematics and statistics. I decided to take a graduate network security course. My capstone research project was within the field of telephony. You can find the paper included in the publications section. In retrospect, I could have submitted this to a security conference, but, at the time, I also got swept up in a couple of other projects with Dr. Laber. In the end, I missed the opportunity to submit this paper to NDSS, CCS and other highly prestigious security conferences. This was certainly a learning experience in finding out what works for you and making sure you can do one thing well and execute it to its fullest extent. However, I learned much about academic writing, writing for a security-research audience and how to structure your argumentation to make sure your paper is contributing to a body of work.

portfolio

StackOverFlow K-Means Analysis

Published:

Implemented a distributed k-means algorithm which clusters posts on the popular question-answer platform StackOverflow according to their score. Moreover, this clustering was executed in parallel for different programming languages, and the results were compared.

publications

Multi-angled Statistical Approach to Human Trafficking Detection and Profiling

Published in Laboratory for Analytical Sciences Conference, 2017

This conference presentation highlights the use of deep learning applied to human trafficking.

Recommended citation: Saanchi, Y, Wang, M., Ahluwalia, S., Laber, E. & Caltagirone, S. (2017, December). Multi-angled Statistical Approach to Human Trafficking Detection and Profiling. Laboratory for Analytical Sciences, Raleigh, NC. http://ahlusar1989.github.io/files/las_2017_trafficking_detection_v2.pdf

Deep Learning within the Context of Doom

Published in North Carolina State University Symposium, 2019

There are two main approaches to reinforcement learning: value based methods that aim to find an optimal Q-function, and policy based ones that directly look for the optimal policy. However most of reinforcement learning problems (such as learning to play games) have large or even continuous states spaces, which makes constructing Q-values table impossible. Thus, there is a need for approximate reinforcement learning. In this paper the Deep reinforcement learning methods we’ve used to train an agent to play in a Doom environment.

Recommended citation: Ahluwalia, Saran and Laber, Eric. (2019, May). " Deep Learning within the Context of Doom" Unpublished Manucript http://ahlusar1989.github.io/files/st_498_independent_study_05_04_2019.pdf

talks

Predicting Gentrification in District of Columbia Wards

Published:

I was part of a interdisciplinary group - aptly named Gentrifuge - tasked with creating a solution during a three month sprint with the Department of Commerce and Small Business Administration. The problem was framed as such:

Enabling Search Using bi-LSTMs

Published:

Within large enterprises, disparate unstructured data sources contain valuable information that needs to be unified with other instrumentation in order to fulfill business objectives. In this talk, we present a deep learning implementation, using bidirectional - LSTMs that enabled Cisco S&TO to enable search for forensic evidence evaluation and unification. Our solution enabled Cisco to streamline processes for service request triaging - reducing time to address customer requests.

teaching

Capital Teaching Resident

Secondary Education, E.L Haynes Public Charter High School, 2012

Member of Selective (Acceptance Rate: 8%) Capital Teaching Residency Program (Partnership between The New Teacher Project (TNTP) and the KIPP Foundation’ KIPP DC Network)

AP Biology Teacher

High School, E.L Haynes Public Charter High School, 2013

Independently created science department for 11th grade in novel high school (at the time, its third year of existence), providing service to 250 students spanning grades 10 and 11. Utilizing CRM - based software, Google Analytics and Excel collaborated with and informed administration, utilizing quantitative and qualitative data, in order to streamline school graduation pathways and drive data-driven interventions for at-risk students (students who possessed greatest achievement gaps in literacy and numeracy)

Upper School Educational Technologist

High School, Stone Ridge School of the Sacred Heart, 2013

Under the guidance of the Director of Technology spearheading entrepreneurial Robotics and Web - Development program in the Upper School, leveraging 3-D printing technologies and in-house professional development curriculum for Upper School Faculty that utilizes Google Applications, Web 2.0 applications and peripheral hardware