Papers by Siddharth Varia
Balancing Classification and Calibration Performance in Decision-Making LLMs via Calibration Aware Reinforcement Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are increasingly deployed in decision-making tasks where accuracy and reliable confidence estimates are essential. |
| Approach: | They propose a calibration-aware reinforcement learning formulation that directly adjusts decision-token probabilities. |
| Outcome: | The proposed model preserves RLVR’s accuracy level while mitigating overconfidence, reducing ECE scores up to 9 points. |
DeSePtion: Dual Sequence Prediction and Adversarial Examples for Improved Fact-Checking (2020.acl-main)
Copied to clipboard
Christopher Hidey, Tuhin Chakrabarty, Tariq Alhindi, Siddharth Varia, Kriste Krstovski, Mona Diab, Smaranda Muresan
| Challenge: | Fact Extraction and Verification datasets provide a resource for end-to-end fact-checking, requiring retrieval of evidence from Wikipedia to validate a veracity prediction. |
| Approach: | They propose a system that is resilient to attacks by multiple propositions, temporal reasoning, ambiguity and lexical variation and a sequence of evidence sentences and veracity relation predictions. |
| Outcome: | The proposed system is resilient to three realistic “attacks” and obtains state-of-the-art results due to improved evidence retrieval. |
Detecting Gang-Involved Escalation on Social Media Using Context (D18-1)
Copied to clipboard
Serina Chang, Ruiqi Zhong, Ethan Adams, Fei-Tzin Lee, Siddharth Varia, Desmond Patton, William Frey, Chris Kedzie, Kathy McKeown
| Challenge: | In cities such as Chicago, gang-involved youth have increasingly turned to social media to post about their experiences and intents online. |
| Approach: | They propose a system that uses domain-specific resources and contextual representations of the emotional and semantic content of the user’s recent tweets and their interactions with other users to detect Aggression and Loss in social media posts. |
| Outcome: | The proposed system improves on a large unlabeled dataset and incorporates contextual representations of the emotional and semantic content of the user’s recent tweets as well as their interactions with other users. |
A Weak Supervision Approach for Few-Shot Aspect Based Sentiment Analysis (2024.eacl-long)
Copied to clipboard
Robert Vacareanu, Siddharth Varia, Kishaloy Halder, Shuai Wang, Giovanni Paolini, Neha Anna John, Miguel Ballesteros, Smaranda Muresan
| Challenge: | Existing methods to improve few-shot performance in aspect-based sentiment analysis (ABSA) require complex interactions between the target and the polarity of the sentiment. |
| Approach: | They propose a pipeline approach to construct a noisy ABSA dataset and adapt it to the ABSA tasks. |
| Outcome: | The proposed model outperforms the state-of-the-art on the aspect extraction sentiment classification task and is capable of performing the harder aspect sentiment triplet extraction task. |
A Multi-Modal Multilingual Benchmark for Document Image Classification (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing document image classification datasets have several limitations and we present two new datasets that overcome these limitations. |
| Approach: | They propose to use two newly curated multilingual datasets that overcome these limitations and propose to develop multilingual Document AI models. |
| Outcome: | The proposed datasets overcome limitations in document image classification and open the door for future research into improving Document AI models. |
Agent vs. Agent: Automated Data Generation and Red-Teaming for Custom Agentic Workflows (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Existing red-teaming frameworks like AgentHarm use static prompts and hardcoded toolsets . |
| Approach: | They propose a red-teaming framework that generates adversarial tasks and evaluation functions tailored to arbitrary toolsets and uses iterative prompt refinement with self-reflection to develop more effective attacks. |
| Outcome: | The proposed approach achieves 162% increase in attack success rate on o4-mini and 86% success on gemini 2.5 Pro. |