Large Language Models (LLMs) have seen widespread adoption due to their remarkable natural language capabilities. However, when deploying them in real-world settings, it is important to align LLMs to generate texts according to acceptable human standards. Methods such as Proximal Policy Optimization (PPO) and Direct Preference Optimization (DPO) have enabled significant progress in refining LLMs using human preference data. However, the privacy concerns inherent in utilizing such preference data have yet to be adequately studied. In this paper, we investigate the vulnerability of LLMs aligned using two widely used methods - DPO and PPO - to membership inference attacks (MIAs). Our study has two main contributions: first, we theoretically motivate that DPO models are more vulnerable to MIA compared to PPO models; second, we introduce a novel reference-based attack framework specifically for analyzing preference data called PREMIA (Preference data MIA). Using PREMIA and existing baselines we empirically show that DPO models have a relatively heightened vulnerability towards MIA.

Siva Rajesh Kasa, Karan Gupta, Sumegh Roychowdhury, Ashutosh Kumar, Yaswanth Biruduraju, Santhosh Kumar Kasa, Pattisapu Nikhil Priyatam, Arindam Bhattacharya, Shailendra Agarwal, Vijay Huddar. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Pre-trained language models (PLMs) have seen tremendous success in text classification (TC) problems in the context of Natural Language Processing (NLP). In many real-world text classification tasks, the class definitions being learned do not remain constant but rather change with time - this is known as concept shift. Most techniques for handling concept shift rely on retraining the old classifiers with the newly labelled data. However, given the amount of training data required to fine-tune large DL models for the new concepts, the associated labelling costs can be prohibitively expensive and time consuming. In this work, we propose a reformulation, converting vanilla classification into an entailment-style problem that requires significantly less data to re-train the text classifier to adapt to new concepts. We demonstrate the effectiveness of our proposed method on both real world & synthetic datasets achieving absolute F1 gains upto ~6% and ~30% respectively in few-shot settings. Further, upon deployment, our solution also helped save 75% direct labeling costs and 40% downstream labeling costs overall in a span of 3 months.

MOTIVATION: The identification of sub-populations of patients with similar characteristics, called patient subtyping, is important for realizing the goals of precision medicine. Accurate subtyping is crucial for tailoring therapeutic strategies that can potentially lead to reduced mortality and morbidity. Model-based clustering, such as Gaussian mixture models, provides a principled and interpretable methodology that is widely used to identify subtypes. However, they impose identical marginal distributions on each variable; such assumptions restrict their modeling flexibility and deteriorates clustering performance. RESULTS: In this paper, we use the statistical framework of copulas to decouple the modeling of marginals from the dependencies between them. Current copula-based methods cannot scale to high dimensions due to challenges in parameter inference. We develop HD-GMCM, that addresses these challenges and, to our knowledge, is the first copula-based clustering method that can fit high-dimensional data. Our experiments on real high-dimensional gene-expression and clinical datasets show that HD-GMCM outperforms state-of-the-art model-based clustering methods, by virtue of modeling non-Gaussian data and being robust to outliers through the use of Gaussian mixture copulas. We present a case study on lung cancer data from TCGA. Clusters obtained from HD-GMCM can be interpreted based on the dependencies they model, that offers a new way of characterizing subtypes. Empirically, such modeling not only uncovers latent structure that leads to better clustering but also meaningful clinical subtypes in terms of survival rates of patients. AVAILABILITY AND IMPLEMENTATION: An implementation of HD-GMCM in R is available at: https://bitbucket.org/cdal/hdgmcm/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.