Jinen Setpal

Aspiring academic. Research in Machine Vision, NLP & Interpretability.

Hi there 馃憢馃徏 I'm Jinen

Welcome to my portfolio website!

Work

  • DagsHub 路 Jun 2022 - Present
    Machine Learning Engineer
  • ARKaNLU @ Purdue University 路 Jan 2023 - Present
    Student Researcher
  • CS390-WAP @ Purdue University 路 Aug 2022 - Dec 2022
    Course Instructor
  • The Data Mine @ Purdue University 路 Feb 2022 - May 2022
    Undergraduate Teaching Assistant
  • Teachiq AB / exam.net 路 Sep 2020 - Jul 2021
    System Developer

Education

  • Purdue University 路 Aug 2021 - 2024
    B.Sc, Data Science
  • R.N. Podar School 路 2019 - 2021
    Computer Science (CBSE)

Publications

CutLang V2: Advances in a runtime-interpreted analysis description language for HEP data

CERN 路 Frontiers in Big Data 路 Prof. G枚khan 脺nel, et al.

Github Logo PDF Logo

We will present the latest developments in CutLang, the runtime interpreter of a recently-developed analysis description language (ADL) for collider data analysis. ADL is a domain-specific, declarative language that describes the contents of an analysis in a standard and unambiguous way, independent of any computing framework. In ADL, analyses are written in human-readable plain text files, separating object, variable and event selection definitions in blocks, with a syntax that includes mathematical and logical operations, comparison and optimisation operators, reducers, four-vector algebra and commonly used functions. Adopting ADLs would bring numerous benefits to the LHC experimental and phenomenological communities, ranging from analysis preservation beyond the lifetimes of experiments or analysis software to facilitating the abstraction, design, visualization, validation, combination, reproduction, interpretation and overall communication of the analysis contents. Since their initial release, ADL and CutLang have been used for implementing and running numerous LHC analyses. In this process, the original syntax from CutLang v1 has been modified for better ADL compatibility, and the interpreter has been adapted to work with that syntax, resulting in the current release v2. Furthermore, CutLang has been enhanced to handle object combinatorics, to include tables and weights, to save events at any analysis stage, to benefit from multi-core/multi-CPU hardware among other improvements. In this contribution, these and other enhancements are discussed in details. In addition, real life examples from LHC analyses are presented.







ArchiMeDe: A New Model Architecture for Meme Detection

Seventh Evaluation Campaign of Natural Language Processing and Speech Tools (EVALITA 2020) 路 Jinen Setpal, Gabriele Sarti

Github Logo PDF Logo

We introduce ArchiMeDe, a multimodal neural network-based architecture used to solve the DANKMEMES meme detections subtask at the 2020 EVALITA campaign. The system incorporates information from visual and textual sources through a multimodal neural ensemble to predict if input images and their respective metadata are memes or not. Each pre-trained neural network in the ensemble is first fine-tuned individually on the training dataset to perform domain adaptation. Learned text and visual representations are then concatenated to obtain a single multimodal embedding, and the final prediction is performed through majority voting by all networks in the ensemble.

Patents

Semi-Supervised Class Activation Mappings for Target Localization & Super-Resolution

TE Corporate (UK) 路 Patent Pending 路 Jinen Setpal, et al.

Our proposed invention incorporates techniques within Machine Interpretability to improve classification within electrical connectors manufactured under TE Connectivity's Global Application Tooling Division, using Class Activation Mapping based target localization, super resolution and subsequent classification.







Leveraging Latent Features for Modular Multiclass Classification

TE Corporate (UK) 路 Patent Pending 路 Jinen Setpal, et al.

We propose a system for multiclass classification enabling scalable, modular class modification with minimal training. Assigning each classification target as a binary model unto itself, we extract the latent features from the binary classifiers to train an aggregator network that performs the actual multiclass classification.

Highlighted Projects

Key Skillset


Industrial Knowledge

  • Machine Learning
  • Reverse Engineering
  • Binary Analysis
  • Shell Scripting
  • Lexical Analysis
  • Web Development
  • Technical Writing
  • Application Development
  • Cryptography
  • Data Mining
  • Web Exploitation

Tools & Technologies

  • Python
  • C++
  • Assembly
  • Java
  • Bash
  • R
  • Git
  • DVC
  • MLFlow
  • Node.js
  • LaTeX