Mohammad Afshar

IBL Education

Machine Intelligence Engineer • Jan, 2018 — Present

Building weakly supervised deep generative models for IBL Education's predictive analytics pipeline.

Tapad

Software Engineering Intern • Jun, 2017 — Aug, 2017

• Utilize the scalatest and mockito frameworks to increase test coverage on the data platform service improving the stability of the platform engine’s ability to generate an accurate graph of all connected electronic devices
• Connect a web of consumer and client APIs using kafka streams

Criteo

Machine Intelligence Engineering Intern • Sep, 2015 — Aug, 2016

• Leveraged Caffe and Torch to implement learning models (Convolutional Neural Nets) to identify similar products based on image and description
• Implemented API to leverage those models at scale, serving 10 million requests per day
• Project contributed to 14% improvement in overall product performance

Criteo

Data Engineering Intern • Jun, 2015 — Aug, 2015

• Leveraged Graphite and Grafana to record better KPIs
• Aided in creating a real-time computation to increase throughput and maintain latency
• Increased visibility into the data pipeline by 16%

New York University - Courant Institute of Mathematical Sciences

Master of Science • 2016 — 2018

Computer Science, Machine Learning

New York University - Courant Institute of Mathematical Sciences

Bachelor of Arts • 2013 — 2017

Majors - Mathematics, Computer Science | Minor - Electrical Engineering

Facial Recognition

Developer • 2016

Designed and implemented using Lua, Torch, and Torchnet a Convolutional Neural Network (CNN) to recognize faces. The dataset was the Labelled Faces in The Wild (LFTW) dataset; the design of the CNN was based on Facebook's initial facial recognition pipeline, called DeepFace. Pipeline focused on initial detection followed by training. Detection was based on 2-D and 3-D alignment. Accuracy on test set was 85%.

Single-Shot MultiBox Detector with Pose Estimation

Developer • 2017

Designed and implemented a pose-estimation platform using Python, OpenCV, and TensorFlow to detect humans, estimate their pose using a Single Shot Multi-box Detector, and follow their pose in realtime. Leveraged TensorFlow to design Convolutional Neural Networks (CNNs) for human detection. Utilized an open source library written in Keras for pose-estimation pipeline.

Speaker Verification

Developer • 2015

Designed and implemented a speaker classification system based on the Simpson's dataset. The pipeline worked to obtain low-level features (MFCCs) and uses a bag-of-words method to feature engineer a vector that represents the audio. The classification model used an SVM with an RBF Kernel, which led to a 75% accuracy.

Cat&Dog Classification

Developer • 2015

Survey of various techniques in the classification of cat and dog images (total of 37 classes)

1)
• A bag-of-words model using both SIFT and SURF keypoints
• Feature engineer using these keypoints and use a multiclass Support Vector Machine (SVM) with an RBF Kernel
• Max accuracy after parameter tuning was 30%.

2)
• Using pretrained AlexNet Convolutional Neural Network (CNN) to train on current data, with a SoftMax layer at the end for a probability distribution.
• Top 1 accuracy was 75%
• Top 5 accuracy was 85%

3)
• Using pretrained AlexNet Convolutional Neural Network (CNN) to train on current data, with a Random Forest at the end for classification.
• Top 1 accuracy was 85%
• Top 5 accuracy was 93%

Bump & Texture Mapping

Developer • 2017

Utilized the OpenGL pipeline to provide texture and bump mapping for .obj files. Developed the entire project from scratch, including both GPU and CPU programming. Obtained realistic results on objects rendered with both bump mapping and texture mapping.

Weakly Supervised Disease Detection and Classification

Developer • 2018

Coming soon.

Languages

Python, Scala, Java, Lua

Tools & Frameworks (Machine Learning)

Scikit-Learn, TensorFlow, PyTorch, OpenCV, Caffe

Tools & Frameworks (Big Data)

Spark, Hadoop, Hive

Tools & Frameworks (Cloud Services, CI/CD)

AWS, Docker, Travis, Jenkins, Kubernetes