Machine Learning Engineer
Resume:
Mohammad Umair
WORK EXPERIENCE
Ebiquity UMBC
Research Assistant SUMMER 2020 – PRESENT
Built a language modeling approach that can be ap- plied to any given language encoder where we further train pretrained Transformers like BERT by incorpo- rating semantic meaning representations during the pre-training step, adding additional task-agnostic se- mantic knowledge to the given model and improving
downstream semantic inference task performance by
5-10% on more than 10 NLP tasks for a number of
baseline models, notably BERT.
UMBC
Teaching Assistant FALL 19 - SPRING 20
Teaching Assistant for Data Structures and Algo-
rithms Undergrad course, helping students with C++ programming and algorithms related issues.
SAFARIFONE Inc
Software Support Engineer JUNE 18 – JULY 19
Provide support for complete Safarifone solution net- work which involves working on CentOS Linux distri- butions, MySQL databases and Log Management Sys-
tems in Python and Bash. Responsible for handling
customer-related issues, assigning tickets to the rele- vant teams, and keeping up the communication until the issue is resolved.
CODEMAX
App/Game Developer JAN 16 - AUG 17
Design, develop and upload basic android utility apps in Android Studio, and write C# scripts for unity pack- ages and user-optimized games in Unity3D for both
Android and iOS platforms.
Bahria University
NLP Research FALL 17 - SPRING 18
Research work on Sentiment Analysis tasks in the
area of Digital Text Forensics where I learned and coded supervised classification algorithms and devel- oped a python desktop application that allows auto- mated training, evaluation and report-generation for text classification models.
KEY SKILLS
LANGUAGES Python, C++, C#, Java,
JavaScript, SQL
TECHNOLOGIES Pytorch, ASP.NET, Django,
Linux, Android, Git, Pandas
TECH SKILLS Natural Language Processing,
Machine Learning, Agile
Workflows, Software
Management, Algorithms
COMMUNICATION Active Listening and Speaking,
Work Ethics, Teamwork,
Mentorship
b 590A Yale Ave, Baltimore, MD, 21229
B ***********@*****.***
E www.linkedin.com/in/umairwork50/
EDUCATION
2021 Masters of Science
COMPUTER SCIENCE, 3.62 CGPA
University of Maryland, Baltimore County
Graduation Date: May 2021
2018 Bachelors of Science
COMPUTER SCIENCE, 3.46 CGPA
Bahria University Islamabad, PK
KEY PROJECTS
2020 Triplet loss with Siamese BERT
Further taking the research work on lan-
guage encoders, implemented triplet loss
objective function with Siamese BERT-
networks, significantly improving general
knowledge of pretrained models and gen-
erated embeddings, with a 5-10% improve-
ment on more than 20 NLP downstream and
probing tasks, for a number of baseline mod-
els including BERT, XLNET and GPT-2.
2019 Tweet Sentiment Analysis
Applied novel approach of contextual polar-
ity in phrase-level sentiment analysis to a
large Twitter corpus, by training two maxi-
mum entropy models that disambiguate neu-
tral and polar tweets and then further cate-
gorizes the polar tweets as positive or nega-
tive. Over the unigram-only MaxEnt model,
it improves by an absolute 0.1 increase in
macro F1 score.
2018 Authorship Attribution capstone
Given a labeled corpus of newspaper arti-
cles, train a model by extracting stylome-
try features from text with TF-IDF weighing
and stemming techniques, and classify the
correct author using cosine similarity, KNN
and SVM classifiers. For our datasets with
around 50-100k sentences, it takes about 5
minutes to train the model on a local CPU
while reaching an accuracy of about 70-80%
for test data.
2017 Multiplayer Game
A globally-connected 3D Simulation game for
an Upwork client, developed in Unity using
C# and optimized for older versions of An-
droid and iOS devices.
PUBLICATIONS (UNDER APPROVAL)
2020 Jointly representing semantic meaning
representations improves downstream
inference tasks
Dr. Frank Ferraro, M. Umair
Contact this candidate