02 — 3 peer-reviewed · 5 preprints & reports

Published coordinates

Work on algorithmic fairness, explainability, and the dynamics of online platforms — from NeurIPS workshops to Springer proceedings. Google Scholar ↗

arXiv preprint · cs.CY, cs.DB, cs.LG

Auditing Discriminatory Patterns in Mortgage Lending Through Association Rules and Fair Binning

Archit Rathod, Dhwani Chande, Het Nagda

Mortgage lending in the United States exhibits persistent racial and gender disparities. We investigate whether standard data preprocessing steps, specifically attribute binning, amplify these disparities in downstream pattern mining. Using 103,481 cleaned mortgage applications from the HMDA 2023 dataset (Chicago metropolitan area), we build a three-stage pipeline: (1) a PySpark data cleaning and binning pipeline that implements both standard equal-frequency binning and the epsilon-biased fair binning algorithm from Asudeh et al., (2) FP-Growth association rule mining that compares denial patterns under both binning regimes, and (3) K-Means clustering with a per-cluster disparate impact audit. Our standard binning shows 9.63% racial bias in income discretization, consistent with the 8-10% reported in prior work. Fair binning with seven race groups is infeasible at epsilon=0.03 and only succeeds at epsilon=0.08 with a Price of Fairness of 29.4%. FP-Growth reveals that high debt-to-income ratio is the dominant denial predictor (67.2% confidence, 2.81 lift), while racial bias does not appear as explicit high-support rules. However, K-Means clustering followed by a disparate impact audit flags 10 out of 45 cluster-group pairs, showing that Black applicants face significantly higher denial rates than White applicants even among financially similar groups.

  • Algorithmic Fairness
  • Mortgage Lending
  • HMDA
  • Fair Binning
  • FP-Growth
  • Association Rules
  • Disparate Impact
2026
arXiv ↗Code ↗
Technical Report · UIC Spring 2026

Responsible AI for Scientific Discovery: Evaluating Explainability Methods for Galaxy Morphology Classification across Multiple Architectures and Datasets

Gargi Sathe, Archit Rathod

Deep learning classifiers are now standard tools for automated galaxy morphology classification at the scale demanded by next-generation astronomical surveys. Their black-box nature, however, undermines scientific trust. We present a comparative study of four post-hoc explainability methods — Grad-CAM, LIME, Integrated Gradients, and GradientSHAP — applied to four convolutional architectures (ResNet-18, VGG-16, EfficientNet-B0, and a lightweight Custom CNN) on the Galaxy10 DECaLS dataset, with a secondary case study on the Galaxy Zoo Evo tiny subset. The secondary GZ Evo experiment surfaces a key lesson: under strict vote-fraction filtering and small-sample conditions, XAI evaluation can become unstable in ways that practitioners must report honestly. Together the experiments argue that XAI faithfulness rankings depend on both architecture and dataset conditions, and that no single explainer dominates universally.

  • Galaxy Morphology
  • Explainability
  • XAI
  • Grad-CAM
  • LIME
  • Integrated Gradients
  • GradientSHAP
2026
Technical Report · UIC Fall 2025

Measuring and Mitigating Toxicity in Large Language Models: A Comprehensive Replication Study

Mokshit Surana, Archit Rathod, Akshaj Kurra Satishkumar

Large Language Models (LLMs), when trained on web-scale corpora, inherently absorb toxic patterns from their training data. This leads to "toxic degeneration" where even innocuous prompts can trigger harmful outputs. This phenomenon poses significant risks for real-world deployments, necessitating effective mitigation strategies that maintain model utility while ensuring safety. In this comprehensive replication study, we evaluate the efficacy of DExperts (Decoding-time Experts), an inference-time mitigation technique that steers generation without requiring model retraining. We structured our research into three systematic phases: (1) establishing baseline toxicity measurements using RealToxicityPrompts on standard GPT-2 models; (2) implementing and evaluating DExperts to mitigate explicit toxicity; and (3) stress-testing the method against implicit hate speech using the adversarial ToxiGen dataset. Our empirical results confirm that while DExperts achieves near-perfect safety rates (100%) on explicit toxicity benchmarks, it exhibits brittleness against adversarial, implicit hate speech, with safety rates dropping to 98.5%. Furthermore, we quantify a critical trade-off: the method introduces a ~10x latency penalty (from 0.2s to 2.0s per generation), posing challenges for real-time deployment scenarios.

  • LLM Safety
  • Toxicity Mitigation
  • DExperts
  • GPT-2
  • ToxiGen
  • RealToxicityPrompts
  • AI Safety
2025
arXiv ↗
CS 418 · UIC Fall 2025

Coordinated Amplification and Misinformation Detection in Global YouTube Conflict Narratives

Archit Rathod, Srinath Ganesh, Vishaal Dayashanker, Harsh Shelke, Vignesh Pathak

YouTube serves as a major conduit for viral, multilingual political narratives, particularly during global conflicts. This project investigates coordinated amplification patterns and misinformation detection in YouTube content related to the Russia-Ukraine conflict. We analyzed approximately 5.9 million comments across 440,772 videos from 1,561 channels using a multi-method approach combining network science, anomaly detection, and natural language processing. Our findings validate three core hypotheses: (1) misinformation is amplified by highly interconnected channel and commenter clusters, (2) periods of intense real-world conflict correlate with statistically significant engagement anomalies, and (3) narratives evolve predictably over time in alignment with external war events. The project demonstrates the effectiveness of combining PageRank centrality analysis, Isolation Forest anomaly detection, and BERTopic modeling for detecting coordinated information campaigns at scale.

  • Misinformation Detection
  • YouTube
  • Network Science
  • PageRank
  • BERTopic
  • Anomaly Detection
  • NLP
2025
Technical Report · UIC 2024

Benchmarking Algorithms for Heterogeneous Treatment Effect Estimation in Networks

Archit Rathod, Mokshit Surana, Gargi Sathe

This research focuses on benchmarking heterogeneous treatment effect (HTE) estimation algorithms in networked environments to enhance our understanding of causal relationships. By evaluating models such as X-Learner, T-Learner, and Causal Forest across synthetic, semi-synthetic, and real-world datasets, this work addresses the challenges posed by confounding, mediation, and interference in social networks. Through rigorous dataset generation, model tuning, and performance evaluation using metrics like ATE error and PEHE, the study highlights the strengths and limitations of these algorithms. Key findings demonstrate the variability in model performance under different conditions and underscore the need for context-aware model selection. This comprehensive benchmarking framework aims to inform future developments in causal inference methodologies, advancing robust and scalable solutions for complex network environments.

  • Causal Inference
  • Heterogeneous Treatment Effects
  • X-Learner
  • T-Learner
  • Causal Forest
  • Social Networks
  • PEHE
2024
✦ Peer-reviewed/ICDSA 2024 · Springer Nature/pp. 63–77

Ascend.AI — Building Confidence Through Technology: A Technical Exploration of Facial Expression, Tone, and Pitch Analysis with Chatbot Guidance

Archit Rathod, Gargi Sathe, Siddh Shah, Kumkum Saxena

The proposed approach embarks on an intricate and comprehensive exploration of advanced and innovative technologies to enhance interview skills. Leveraging the power of OpenCV and Xception, the paper delves into the nuances of facial expression analysis, unraveling the intricacies of emotion recognition. The system analyzes tone and pitch with the aid of tools like LIBROSA to extract vocal features in order to understand the intensity of emotions, and has developed a 1D-CNN model for classification using RAVDESS, TESS, SAVEE, and CREMA-D datasets. The system includes a chatbot using vector database Qdrant and an open-source LLM Mixtral 8x7b, offering personalized interview guidance derived from scraping 30 diverse websites for interview-related questions. This technical exploration extends from conventional interview preparation to introducing an innovative framework that intertwines machine learning models with real-time analysis.

  • Interview Preparation
  • Facial Expression Analysis
  • Tone Analysis
  • Pitch Analysis
  • OpenCV
  • Xception
  • LIBROSA
2024
✦ Peer-reviewed/NeurIPS 2023 · MASec Workshop

Multiagent Simulators for Social Networks

Aditya Surve, Archit Rathod, Mokshit Surana, Gautam Malpani, Aneesh Shamraj, Sainath Reddy Sankepally, Raghav Jain, Swapneel S Mehta

Multiagent social network simulations are an avenue that can bridge the communication gap between the public and private platforms in order to develop solutions to a complex array of issues relating to online safety. While there are significant challenges relating to the scale of multiagent simulations, efficient learning from observational and interventional data to accurately model micro and macro-level emergent effects, there are equally promising opportunities — not least with the advent of large language models that provide an expressive approximation of user behavior. In this position paper, we review prior art relating to social network simulation, highlighting challenges and opportunities for future work exploring multiagent security using agent-based models of social networks.

  • Multiagent Systems
  • Social Network Simulation
  • LLMs
  • Online Safety
  • Agent-Based Models
  • Emergent Behavior
  • Network Security
2023
arXiv ↗
✦ Peer-reviewed/ICSISCET 2023 · Springer Nature/pp. 311–326

Leveraging CNNs and Ensemble Learning for Automated Disaster Image Classification

Archit Rathod, Veer Pariawala, Mokshit Surana, Kumkum Saxena

Natural disasters act as a serious threat globally, requiring effective and efficient disaster management and recovery. This paper focuses on classifying natural disaster images using Convolutional Neural Networks (CNNs). Multiple CNN architectures were built and trained on a dataset containing images of earthquakes, floods, wildfires, and volcanoes. A stacked CNN ensemble approach proved to be the most effective, achieving 95% accuracy and an F1 score going up to 0.96 for individual classes. Tuning hyperparameters of individual models for optimization was critical to maximize the models' performance. The stacking of CNNs with XGBoost acting as the meta-model utilizes the strengths of the CNN and ResNet models to improve the overall accuracy of the classification. Results obtained from the models illustrated the potency of CNN-based models for automated disaster image classification. This lays the foundation for expanding these techniques to build robust systems for disaster response, damage assessment, and recovery management.

  • Disaster Classification
  • CNN
  • Ensemble Learning
  • XGBoost
  • ResNet
  • Computer Vision
  • Image Classification
2023