ML Graphs Interp Robustness NLP Alignment CSS Machine Learning AI for Mathematical Reasoning Advancing AI capabilities in theorem proving, formalization, and mathematical discovery MathAI @ NeurIPS Machine Unlearning Investigating methods to selectively remove information from trained models while preserving overall performance TMLR ICML DMLR @ ICLR Random Representations for Continual Learning Demonstrating superior performance of fixed random transforms over learned representations in online continual learning NeurIPS Vocabulary Compression for LLM Pretraining A simple approach to compress the vocabulary layer of an LLM during pre-training to reduce memory requirements and increase throughput. MLC @ NeurIPS Graphs Graph Explainability Enhancing GNN explanations through integration of higher-order structures for improved accuracy and scalability AAAI Graph Unlearning Developing techniques for selective information removal in graph-based models CoLLAs Workshop Homophily Studying the impact of node similarity and connectivity patterns in graph neural networks Knowledge Graph Reasoning Enhancing foundation models for knowledge graph reasoning through semantic relationship exploration Infonets Exploring information flow and network architectures in graph-based learning systems Interpretability Representation Surgery Editing the neural model representations to alter model behavior through approaches such as steering functions. ICML Interpreting OCR Investigating textual semantic subspaces within image encoders for rendered text. This work shows strong evidence for a textual semantic subspace inside the image encoder for rendered text. UniReps @ NeurIPS Atomics Attempt to develop an interpretable LM architecture by incorporating orthogonality into Backpack Language Models. Interpretability in Instruction Tuning Understanding model behavior through counterfactual analysis Robustness LLM Consistency Evaluations Developing frameworks to evaluate and improve the logical consistency of language models. COLING Cross-Modal Adversarial Attacks Investigating adversarial vulnerabilities and attack transferability across different modalities in AI systems. Agentic Robustness Enhancing AI system robustness through policy generation and adaptive strategies Knowledge-infused Distracted Driver Detection Novel framework integrating scene graphs and pose information to detect distracted driving behaviors KIL @ KDD Natural Language Processing Code-Mixed Text Analysis & Generation Developing tools and methodologies for analyzing multilingual and code-switched text ACL Counter Turing Test for Hindi Evaluation framework for detecting AI-generated Hindi text across 26 LLMs and development of Hindi AI Detectability Index EMNLP-F Bias in Legal AI Examining bias and accuracy in Large Language Models for legal reasoning tasks in the Indian context JURIX South Asian Coreference Resolution Machine learning framework for multilingual coreference resolution across 31 South Asian languages COLING Information Disguise Towards effective paraphrasing techniques for protecting sensitive information Small Language Models Exploring Small Language Models (SLMs) as a transformative interface between natural language and real-world systems through on-device function calling. Legal Citation Detection System AI-powered system for identifying citation-worthy content in legal documents using deep learning JURIX AI Alignment Restoring the Safety of Fine-Tuned LLMs Exploring methods to restore safety alignment in fine-tuned language models Bias and Context Investigating the role of context in model bias and developing mitigation strategies Toxcity Detection in Memes Leveraging knowledge distillation and infusion to enhance multimodal toxicity detection, with strong performance on hateful meme benchmarks. Pluralistic Alignment Exploring diverse approaches to AI alignment considering multiple value systems and perspectives Impersonation Investigating LLMs' capabilities in combining style and factual information for realistic impersonation Political and Social Alignment Evaluating current LLMs on recent surveys beyond western populations to understand value alignment Computational Social Science Advertising Patterns in Print Media Analyzing patterns and influence of advertising in print newspapers Analyzing Media Framing in Indian Elections Computational analysis of conflict and game frames in election news coverage (2014-2019)