Education

PhD in Computer Science; RIET Lab under Professor Shiri Dori-Hacohen

University of Connecticut, 2021 - Current

  • Built pipeline for on-premises twitter data-lake that is used for ad-hoc investigations into disinformation.
  • Devised a language model optimization scheme for Claim-Matching that assists our non-profit fact-checking partner Meeden that also placed first in CLEF CheckThat! 2022 Competition [1]. Open sourced pytorch codebase for freeuse and reproducibility,
  • Currently researching argumentation-based tasks like Claim Extraction and Relation Classification to further enhance automated campaign classification and detection.

INDUSTRY

Amazon.com; Alexa org - Software Developer Engineer II (L5)

Summer 2020 - Summer 2021

  • Designed data pipelines and service architectures while articulating them to peers and managers in open forum reviews
  • Led project to rewrite an open source BI tool to handle Red/Critical data in a secure and scalable fashion
  • Worked with customers to determine and build dashboarding functionalities that would save them dev time
  • Developed and maintained internal services needed for research scientists to efficiently access and query Alexa data

Systems & Technology Research - Machine Learning Researcher

Spring 2018 - Summer 2020

    KEY PROJECT: Group Dynamics

  • Designed a language model to classify stubbornness and suspiciousness from dialogue
  • Wrote a context dependent text-generator using both MLE and adversarial training regimes
  • Developed an RNN with a GPT2 backbone that works on threads and dynamically learns author embeddings
  • Presented research results to government stakeholders at DARPA

  • KEY PROJECT: Physically Realizable Attacks

  • Developed a pipeline to perform Expectation-of-Transformation (EoT) attack experiments in a variety of domains
  • Demonstrated that Black Box settings can still work in an EoT setup that is physically realizable
  • Created internal tool to perform feature and model visualizations that helped understand experiment results
  • Fooled both state-of-the-art Object Detectors and Classifiers with physical attacks

  • KEY PROJECT: Middle East Event Detection

  • Applied computer vision techniques to learn representations of geolocations based on recorded event data
  • Used both probabilistic and neural NLP techniques to draw and extract relevant information from tweets
  • Implemented Graph based deep learning approaches to handle non-uniform quantization of positions
  • Presented research results to government stakeholders at IARPA


  • Skills

    Nothing better to describe this section than a hodgepodge of concepts and tools I've used

    Docker
    Git
    Computer Vision
    Natural Language Processing
    Reinforcement Learning
    Bash
    C
    C++
    Fortran 90/95
    Julia
    MATLAB
    R
    Scheme
    HTML + CSS
    JS/TS
    PyTorch
    RLlib
    React
    React Native
    Spark
    Tensorflow 1/2 + Keras
    MongoDB
    Neo4j
    SQL
    Flask
    Java/Kotlin
    Kubernetes
    Python
    AWS API Gateway
    AWS CDK
    AWS Lambda
    Amplify
    AppSync
    CloudFormation
    DynamoDB
    EC2
    ECS + Fargate
    ElasticSearch
    Glue
    Neptune
    RDS
    Redis
    Redshift
    Route 53
    S3
    SNS
    SQS
    SageMaker
    Secrets Manager

    Selected Courses

    • Natural Language Processing

    • Bayesian Statistics

    • Statistical Machine Learning

    • Statistical Decision Theory

    • Coding Theory

    • Probability & Stochastic Processes

    • Data Mining

    • Algorithms for Big Data

    • Information Ecosystem Threats

    • Partial Differential Equations

    • Software Intensive Engineering

    • Applied Scientific Computing

    • Data Structures & Algorithms

    • Discrete Structures

    • Abstract Algebra

    • Digital Signal Processing


    UNDERGRAD

    Single Cell RNA Sequencing Analysis

    Independant Study - Spring 2018

    • Participated in an Independent Study to use Hierarchical Poisson Factorization to learn latent embedding of cells
    • Used clustering techniques to classify the weakly labeled data. Simulated in splatter, we evaluated these methods

    Deep Learning for Hyperspectral Data Processing Research

    REU - Summer 2017

    • Participated in an Research Experience for Undergraduates (REU), learning and implementing Deep Learning algorithms in the fields of Computer Vision and Hyperspectral data processing using frameworks such as Tensorflow, Keras, and Caffe
    • Studied effects of transfer learning on source and targets to try to analyze what allowed for best results

    Numerical Eigensolver FEAST Research

    REU - Summer 2016

    • Participated in an REU working with the FEAST algorithm, a reduced system Eigensolver
    • Learned new numerical iterative and direct techniques in solving eigenvalue problems
    • Developed applications for the FEAST user to “tune” a specific interval and get a density of states for the eigenvalues

    Hardware Verification and Logic Debugging Research

    REU - Summer 2015

    • Participated in an REU doing hardware verification and logic debugging of arithmetic circuits
    • Transformed Boolean equations into pseudo-algebraic equations to analyze potentially bugged circuits
    • Brainstormed new ideas to find the origin of Boolean 0-equivalent residuals

    Text Detoxifier

    Project - Fall 2017

    • Finds all versions of “f***” and replaces it with a word that is less explicit but retains sentences original meaning
    • Programmed different models such as a Bag of Words approach, Multiclass Logistic Regression, clustering based off feature vectors, and a skip-gram with different length spans
    • Annotated scraped data, and used simple active learning approach to maximize annotations impacts

    Carvana Car-Masking Competition

    Project - Summer 2017

    • Participated in competition for creating a pixel mask determining where a car is in a picture
    • Used models that combined concepts from Inception-Resnet and U-nets
    • Pretrained on errors from previous model to transfer features on structured components not found in basic training

    Base Run Application

    Project - Fall 2016

    • Developed an application in which users locally play a capture the flag type game using their mobile geolocations.
    • Implemented the backend server side in C++ utilizing RPC’s to communicate with the java client
    • Described deadlines and requirements for team as Development lead

    References

    [1] S. D.-H. Michael Shliselberg, RIET Lab at CheckThat! 2022: improving decoder based re-ranking for claim matching, in: Working Notes of CLEF 2022 - Conference and Labs of the Evaluation Forum, CLEF ’2022, Bologna, Italy, 2022.