James R. Foulds

Philip Feldman, James R. Foulds, Shimei Pan

Recent advances in large language models (LLMs), such as ChatGPT, have led to highly sophisticated conversation agents. However, these models suffer from "hallucinations," where the model generates false or fabricated information. Addressing this challenge is crucial, particularly with AI-driven platforms being adopted across various sectors. In this paper, we propose a novel method to recognize and flag instances when LLMs perform outside their domain knowledge, and ensuring users receive accurate information. We find that the use of context combined with embedded tags can successfully combat hallucinations within generative language models. To do this, we baseline hallucination frequency in no-context prompt-response pairs using generated URLs as easily-tested indicators of fabricated data. We observed a significant reduction in overall hallucination when context was supplied along with question prompts for tested generative engines. Lastly, we evaluated how placing tags within contexts impacted model responses and were able to eliminate hallucinations in responses with 98.88% effectiveness.

Philip Feldman, Aaron Dant, James R. Foulds et al.

Text analysis of social media for sentiment, topic analysis, and other analysis depends initially on the selection of keywords and phrases that will be used to create the research corpora. However, keywords that researchers choose may occur infrequently, leading to errors that arise from using small samples. In this paper, we use the capacity for memorization, interpolation, and extrapolation of Transformer Language Models such as the GPT series to learn the linguistic behaviors of a subgroup within larger corpora of Yelp reviews. We then use prompt-based queries to generate synthetic text that can be analyzed to produce insights into specific opinions held by the populations that the models were trained on. Once learned, more specific sentiment queries can be made of the model with high levels of accuracy when compared to traditional keyword searches. We show that even in cases where a specific keyphrase is limited or not present at all in the training corpora, the GPT is able to accurately generate large volumes of text that have the correct sentiment.

George J. Cancro, Shimei Pan, James Foulds

When a human receives a prediction or recommended course of action from an intelligent agent, what additional information, beyond the prediction or recommendation itself, does the human require from the agent to decide whether to trust or reject the prediction or recommendation? In this paper we survey literature in the area of trust between a single human supervisor and a single agent subordinate to determine the nature and extent of this additional information and to characterize it into a taxonomy that can be leveraged by future researchers and intelligent agent practitioners. By examining this question from a human-centered, information-focused point of view, we can begin to compare and contrast different implementations and also provide insight and directions for future work.

Pronob Kumar Barman, James R. Foulds, Tera L. Reynolds

Peer support is critical to managing chronic health conditions. Online health communities (OHCs) enable patients and caregivers to connect with similar others, yet their large scale makes it challenging to find the most relevant peers and content. This study assessed perceived value, preferred features, and acceptance conditions for algorithmically personalized support group formation within OHCs. A two-phase, mixed-methods survey (N=165) examined OHC participation patterns, personalization priorities, and acceptance of a simulated personalized support group. Perceived value of the simulated support group was high (mean 4.55/5; 62.8% rated 5/5) and 91.5% would join this group. The importance participants placed on peer matching strongly correlated with perceived value (\r{ho}=0.764, p<0.001). Qualitative findings revealed conditional acceptance: participants demand security, transparency, human oversight, and user control over data. Personalized support groups may be desired, but they will not be adopted unless trust, privacy, and algorithmic governance concerns are addressed.

Fatema Hasan, Yulong Li, James Foulds et al.

Speech data has rich acoustic and paralinguistic information with important cues for understanding a speaker's tone, emotion, and intent, yet traditional large language models such as BERT do not incorporate this information. There has been an increased interest in multi-modal language models leveraging audio and/or visual information and text. However, current multi-modal language models require both text and audio/visual data streams during inference/test time. In this work, we propose a methodology for training language models leveraging spoken language audio data but without requiring the audio stream during prediction time. This leads to an improved language model for analyzing spoken transcripts while avoiding an audio processing overhead at test time. We achieve this via an audio-language knowledge distillation framework, where we transfer acoustic and paralinguistic information from a pre-trained speech embedding (OpenAI Whisper) teacher model to help train a student language model on an audio-text dataset. In our experiments, the student model achieves consistent improvement over traditional language models on tasks analyzing spoken transcripts.

Rashidul Islam, Shimei Pan, James R. Foulds

It is now well understood that machine learning models, trained on data without due care, often exhibit unfair and discriminatory behavior against certain populations. Traditional algorithmic fairness research has mainly focused on supervised learning tasks, particularly classification. While fairness in unsupervised learning has received some attention, the literature has primarily addressed fair representation learning of continuous embeddings. In this paper, we conversely focus on unsupervised learning using probabilistic graphical models with discrete latent variables. We develop a fair stochastic variational inference technique for the discrete latent variables, which is accomplished by including a fairness penalty on the variational distribution that aims to respect the principles of intersectionality, a critical lens on fairness from the legal, social science, and humanities literature, and then optimizing the variational parameters under this penalty. We first show the utility of our method in improving equity and fairness for clustering using naïve Bayes and Gaussian mixture models on benchmark datasets. To demonstrate the generality of our approach and its potential for real-world impact, we then develop a special-purpose graphical model for criminal justice risk assessments, and use our fairness approach to prevent the inferences from encoding unfair societal biases.

Pronob Kumar Barman, Tera L. Reynolds, James Foulds

Online health communities (OHCs) are vital for fostering peer support and improving health outcomes. Support groups within these platforms can provide more personalized and cohesive peer support, yet traditional support group formation methods face challenges related to scalability, static categorization, and insufficient personalization. To overcome these limitations, we propose two novel machine learning models for automated support group formation: the Group specific Dirichlet Multinomial Regression (gDMR) and the Group specific Structured Topic Model (gSTM). These models integrate user generated textual content, demographic profiles, and interaction data represented through node embeddings derived from user networks to systematically automate personalized, semantically coherent support group formation. We evaluate the models on a large scale dataset from MedHelp.org, comprising over 2 million user posts. Both models substantially outperform baseline methods including LDA, DMR, and STM in predictive accuracy (held out log likelihood), semantic coherence (UMass metric), and internal group consistency. The gDMR model yields group covariates that facilitate practical implementation by leveraging relational patterns from network structures and demographic data. In contrast, gSTM emphasizes sparsity constraints to generate more distinct and thematically specific groups. Qualitative analysis further validates the alignment between model generated groups and manually coded themes, showing the practical relevance of the models in informing groups that address diverse health concerns such as chronic illness management, diagnostic uncertainty, and mental health. By reducing reliance on manual curation, these frameworks provide scalable solutions that enhance peer interactions within OHCs, with implications for patient engagement, community resilience, and health outcomes.

Philip Feldman, Aaron Dant, James R. Foulds

The accelerating advancements in Artificial Intelligence (AI) and Machine Learning (ML), highlighted by the development of cutting-edge Generative Pre-trained Transformer (GPT) models by organizations such as OpenAI, Meta, and Anthropic, present new challenges and opportunities in warfare and security. Much of the current focus is on AI's integration within weapons systems and its role in rapid decision-making in kinetic conflict. However, an equally important but often overlooked aspect is the potential of AI-based psychological manipulation at internet scales within the information domain. These capabilities could pose significant threats to individuals, organizations, and societies globally. This paper explores the concept of AI weapons, their deployment, detection, and potential countermeasures.

Alden Dima, James Foulds, Shimei Pan et al.

Large Language Models (LLMs) are often provided as a service via an API, making it challenging for developers to detect changes in their behavior. We present an approach to monitor LLMs for changes by comparing the distributions of linguistic and psycholinguistic features of generated text. Our method uses a statistical test to determine whether the distributions of features from two samples of text are equivalent, allowing developers to identify when an LLM has changed. We demonstrate the effectiveness of our approach using five OpenAI completion models and Meta's Llama 3 70B chat model. Our results show that simple text features coupled with a statistical test can distinguish between language models. We also explore the use of our approach to detect prompt injection attacks. Our work enables frequent LLM change monitoring and avoids computationally expensive benchmark evaluations.

Philip Feldman, James R. Foulds, Shimei Pan

The recent explosion of "foundation" generative AI models has been built upon the extensive extraction of value from online sources, often without corresponding reciprocation. This pattern mirrors and intensifies the extractive practices of surveillance capitalism, while the potential for enormous profit has challenged technology organizations' commitments to responsible AI practices, raising significant ethical and societal concerns. However, a promising alternative is emerging: the development of models that rely on content willingly and collaboratively provided by users. This article explores this "egalitarian" approach to generative AI, taking inspiration from the successful model of Wikipedia. We explore the potential implications of this approach for the design, development, and constraints of future foundation models. We argue that such an approach is not only ethically sound but may also lead to models that are more responsive to user needs, more diverse in their training data, and ultimately more aligned with societal values. Furthermore, we explore potential challenges and limitations of this approach, including issues of scalability, quality control, and potential biases inherent in volunteer-contributed content.

Tao Zhang, Ziqian Zeng, Yuxiang Xiao et al.

Large Language Models (LLMs) are prone to generating content that exhibits gender biases, raising significant ethical concerns. Alignment, the process of fine-tuning LLMs to better align with desired behaviors, is recognized as an effective approach to mitigate gender biases. Although proprietary LLMs have made significant strides in mitigating gender bias, their alignment datasets are not publicly available. The commonly used and publicly available alignment dataset, HH-RLHF, still exhibits gender bias to some extent. There is a lack of publicly available alignment datasets specifically designed to address gender bias. Hence, we developed a new dataset named GenderAlign, aiming at mitigating a comprehensive set of gender biases in LLMs. This dataset comprises 8k single-turn dialogues, each paired with a "chosen" and a "rejected" response. Compared to the "rejected" responses, the "chosen" responses demonstrate lower levels of gender bias and higher quality. Furthermore, we categorized the gender biases in the "rejected" responses of GenderAlign into 4 principal categories. The experimental results show the effectiveness of GenderAlign in reducing gender bias in LLMs.

Clarice Wang, Kathryn Wang, Andrew Bian et al.

Currently, there is a surge of interest in fair Artificial Intelligence (AI) and Machine Learning (ML) research which aims to mitigate discriminatory bias in AI algorithms, e.g. along lines of gender, age, and race. While most research in this domain focuses on developing fair AI algorithms, in this work, we show that a fair AI algorithm on its own may be insufficient to achieve its intended results in the real world. Using career recommendation as a case study, we build a fair AI career recommender by employing gender debiasing machine learning techniques. Our offline evaluation showed that the debiased recommender makes fairer career recommendations without sacrificing its accuracy. Nevertheless, an online user study of more than 200 college students revealed that participants on average prefer the original biased system over the debiased system. Specifically, we found that perceived gender disparity is a determining factor for the acceptance of a recommendation. In other words, our results demonstrate we cannot fully address the gender bias issue in AI recommendations without addressing the gender bias in humans.

Fatema Hasan, Kevin S. Xu, James R. Foulds et al.

User-generated data on social media contain rich information about who we are, what we like and how we make decisions. In this paper, we survey representative work on learning a concise latent user representation (a.k.a. user embedding) that can capture the main characteristics of a social media user. The learned user embeddings can later be used to support different downstream user analysis tasks such as personality modeling, suicidal risk assessment and purchase decision prediction. The temporal nature of user-generated data on social media has largely been overlooked in much of the existing user embedding literature. In this survey, we focus on research that bridges the gap by incorporating temporal/sequential information in user representation learning. We categorize relevant papers along several key dimensions, identify limitations in the current work and suggest future research directions.

Ziqian Zeng, Rashidul Islam, Kamrun Naher Keya et al.

Recently, much attention has been paid to the societal impact of AI, especially concerns regarding its fairness. A growing body of research has identified unfair AI systems and proposed methods to debias them, yet many challenges remain. Representation learning for Heterogeneous Information Networks (HINs), a fundamental building block used in complex network mining, has socially consequential applications such as automated career counseling, but there have been few attempts to ensure that it will not encode or amplify harmful biases, e.g. sexism in the job market. To address this gap, in this paper we propose a comprehensive set of de-biasing methods for fair HINs representation learning, including sampling-based, projection-based, and graph neural networks (GNNs)-based techniques. We systematically study the behavior of these algorithms, especially their capability in balancing the trade-off between fairness and prediction accuracy. We evaluate the performance of the proposed methods in an automated career counseling application where we mitigate gender bias in career recommendation. Based on the evaluation results on two datasets, we identify the most effective fair HINs representation learning techniques under different conditions.

Kamrun Naher Keya, Rashidul Islam, Shimei Pan et al.

Healthcare programs such as Medicaid provide crucial services to vulnerable populations, but due to limited resources, many of the individuals who need these services the most languish on waiting lists. Survival models, e.g. the Cox proportional hazards model, can potentially improve this situation by predicting individuals' levels of need, which can then be used to prioritize the waiting lists. Providing care to those in need can prevent institutionalization for those individuals, which both improves quality of life and reduces overall costs. While the benefits of such an approach are clear, care must be taken to ensure that the prioritization process is fair or independent of demographic information-based harmful stereotypes. In this work, we develop multiple fairness definitions for survival models and corresponding fair Cox proportional hazards models to ensure equitable allocation of healthcare resources. We demonstrate the utility of our methods in terms of fairness and predictive accuracy on two publicly available survival datasets.

Guohou Shan, James Foulds, Shimei Pan

Text features that are correlated with class labels, but do not directly cause them, are sometimesuseful for prediction, but they may not be insightful. As an alternative to traditional correlation-basedfeature selection, causal inference could reveal more principled, meaningful relationships betweentext features and labels. To help researchers gain insight into text data, e.g. for social scienceapplications, in this paper we investigate a class of matching-based causal inference methods fortext feature selection. Features used in document classification are often high dimensional, howeverexisting causal feature selection methods use Propensity Score Matching (PSM) which is known to beless effective in high-dimensional spaces. We propose a new causal feature selection framework thatcombines dimension reduction with causal inference to improve text feature selection. Experiments onboth synthetic and real-world data demonstrate the promise of our methods in improving classificationand enhancing interpretability.

Rashidul Islam, Kamrun Naher Keya, Ziqian Zeng et al.

A growing proportion of human interactions are digitized on social media platforms and subjected to algorithmic decision-making, and it has become increasingly important to ensure fair treatment from these algorithms. In this work, we investigate gender bias in collaborative-filtering recommender systems trained on social media data. We develop neural fair collaborative filtering (NFCF), a practical framework for mitigating gender bias in recommending sensitive items (e.g. jobs, academic concentrations, or courses of study) using a pre-training and fine-tuning approach to neural collaborative filtering, augmented with bias correction techniques. We show the utility of our methods for gender de-biased career and college major recommendations on the MovieLens dataset and a Facebook dataset, respectively, and achieve better performance and fairer behavior than several state-of-the-art models.

Kamrun Naher Keya, Yannis Papanikolaou, James R. Foulds

Word embedding models such as the skip-gram learn vector representations of words' semantic relationships, and document embedding models learn similar representations for documents. On the other hand, topic models provide latent representations of the documents' topical themes. To get the benefits of these representations simultaneously, we propose a unifying algorithm, called neural embedding allocation (NEA), which deconstructs topic models into interpretable vector-space embeddings of words, topics, documents, authors, and so on, by learning neural embeddings to mimic the topic models. We showcase NEA's effectiveness and generality on LDA, author-topic models and the recently proposed mixed membership skip gram topic model and achieve better performance with the embeddings compared to several state-of-the-art models. Furthermore, we demonstrate that using NEA to smooth out the topics improves coherence scores over the original topic models when the number of topics is large.

Nilavra Pathak, James Foulds, Nirmalya Roy et al.

Modeling buildings' heat dynamics is a complex process which depends on various factors including weather, building thermal capacity, insulation preservation, and residents' behavior. Gray-box models offer a causal inference of those dynamics expressed in few parameters specific to built environments. These parameters can provide compelling insights into the characteristics of building artifacts and have various applications such as forecasting HVAC usage, indoor temperature control monitoring of built environments, etc. In this paper, we present a systematic study of modeling buildings' thermal characteristics and thus derive the parameters of built conditions with a Bayesian approach. We build a Bayesian state-space model that can adapt and incorporate buildings' thermal equations and propose a generalized solution that can easily adapt prior knowledge regarding the parameters. We show that a faster approximate approach using variational inference for parameter estimation can provide similar parameters as that of a more time-consuming Markov Chain Monte Carlo (MCMC) approach. We perform extensive evaluations on two datasets to understand the generative process and show that the Bayesian approach is more interpretable. We further study the effects of prior selection for the model parameters and transfer learning, where we learn parameters from one season and use them to fit the model in the other. We perform extensive evaluations on controlled and real data traces to enumerate buildings' parameter within a 95% credible interval.

James Foulds, Rashidul Islam, Kamrun Keya et al.

Intersectionality is a framework that analyzes how interlocking systems of power and oppression affect individuals along overlapping dimensions including race, gender, sexual orientation, class, and disability. Intersectionality theory therefore implies it is important that fairness in artificial intelligence systems be protected with regard to multi-dimensional protected attributes. However, the measurement of fairness becomes statistically challenging in the multi-dimensional setting due to data sparsity, which increases rapidly in the number of dimensions, and in the values per dimension. We present a Bayesian probabilistic modeling approach for the reliable, data-efficient estimation of fairness with multi-dimensional protected attributes, which we apply to two existing intersectional fairness metrics. Experimental results on census data and the COMPAS criminal justice recidivism dataset demonstrate the utility of our methodology, and show that Bayesian methods are valuable for the modeling and measurement of fairness in an intersectional context.

James Foulds, Rashidul Islam, Kamrun Naher Keya et al.

We propose definitions of fairness in machine learning and artificial intelligence systems that are informed by the framework of intersectionality, a critical lens arising from the Humanities literature which analyzes how interlocking systems of power and oppression affect individuals along overlapping dimensions including gender, race, sexual orientation, class, and disability. We show that our criteria behave sensibly for any subset of the set of protected attributes, and we prove economic, privacy, and generalization guarantees. We provide a learning algorithm which respects our intersectional fairness criteria. Case studies on census data and the COMPAS criminal recidivism dataset demonstrate the utility of our methods.

James Foulds

Word embeddings improve the performance of NLP systems by revealing the hidden structural relationships between words. Despite their success in many applications, word embeddings have seen very little use in computational social science NLP tasks, presumably due to their reliance on big data, and to a lack of interpretability. I propose a probabilistic model-based word embedding method which can recover interpretable embeddings, without big data. The key insight is to leverage mixed membership modeling, in which global representations are shared, but individual entities (i.e. dictionary words) are free to use these representations to uniquely differing degrees. I show how to train the model using a combination of state-of-the-art training techniques for word embeddings and topic models. The experimental results show an improvement in predictive language modeling of up to 63% in MRR over the skip-gram, and demonstrate that the representations are beneficial for supervised learning. I illustrate the interpretability of the models with computational social science case studies on State of the Union addresses and NIPS articles.

Mijung Park, James Foulds, Kamalika Chaudhuri et al.

Many applications of Bayesian data analysis involve sensitive information, motivating methods which ensure that privacy is protected. We introduce a general privacy-preserving framework for Variational Bayes (VB), a widely used optimization-based Bayesian inference method. Our framework respects differential privacy, the gold-standard privacy criterion, and encompasses a large class of probabilistic models, called the Conjugate Exponential (CE) family. We observe that we can straightforwardly privatise VB's approximate posterior distributions for models in the CE family, by perturbing the expected sufficient statistics of the complete-data likelihood. For a broadly-used class of non-CE models, those with binomial likelihoods, we show how to bring such models into the CE family, such that inferences in the modified model resemble the private variational Bayes algorithm as closely as possible, using the Polya-Gamma data augmentation scheme. The iterative nature of variational Bayes presents a further challenge since iterations increase the amount of noise needed. We overcome this by combining: (1) an improved composition method for differential privacy, called the moments accountant, which provides a tight bound on the privacy cost of multiple VB iterations and thus significantly decreases the amount of additive noise; and (2) the privacy amplification effect of subsampling mini-batches from large-scale data in stochastic learning. We empirically demonstrate the effectiveness of our method in CE and non-CE models including latent Dirichlet allocation, Bayesian logistic regression, and sigmoid belief networks, evaluated on real-world datasets.

Mijung Park, James Foulds, Kamalika Chaudhuri et al.

We develop a privatised stochastic variational inference method for Latent Dirichlet Allocation (LDA). The iterative nature of stochastic variational inference presents challenges: multiple iterations are required to obtain accurate posterior distributions, yet each iteration increases the amount of noise that must be added to achieve a reasonable degree of privacy. We propose a practical algorithm that overcomes this challenge by combining: (1) an improved composition method for differential privacy, called the moments accountant, which provides a tight bound on the privacy cost of multiple variational inference iterations and thus significantly decreases the amount of additive noise; and (2) privacy amplification resulting from subsampling of large-scale data. Focusing on conjugate exponential family models, in our private variational inference, all the posterior distributions will be privatised by simply perturbing expected sufficient statistics. Using Wikipedia data, we illustrate the effectiveness of our algorithm for large-scale data.

James Foulds, Joseph Geumlek, Max Welling et al.

Bayesian inference has great promise for the privacy-preserving analysis of sensitive data, as posterior sampling automatically preserves differential privacy, an algorithmic notion of data privacy, under certain conditions (Dimitrakakis et al., 2014; Wang et al., 2015). While this one posterior sample (OPS) approach elegantly provides privacy "for free," it is data inefficient in the sense of asymptotic relative efficiency (ARE). We show that a simple alternative based on the Laplace mechanism, the workhorse of differential privacy, is as asymptotically efficient as non-private posterior inference, under general assumptions. This technique also has practical advantages including efficient use of the privacy budget for MCMC. We demonstrate the practicality of our approach on a time-series analysis of sensitive military records from the Afghanistan and Iraq wars disclosed by the Wikileaks organization.

Yannis Papanikolaou, James R. Foulds, Timothy N. Rubin et al.

We introduce a novel approach for estimating Latent Dirichlet Allocation (LDA) parameters from collapsed Gibbs samples (CGS), by leveraging the full conditional distributions over the latent variable assignments to efficiently average over multiple samples, for little more computational cost than drawing a single additional collapsed Gibbs sample. Our approach can be understood as adapting the soft clustering methodology of Collapsed Variational Bayes (CVB0) to CGS parameter estimation, in order to get the best of both techniques. Our estimators can straightforwardly be applied to the output of any existing implementation of CGS, including modern accelerated variants. We perform extensive empirical comparisons of our estimators with those of standard collapsed inference algorithms on real-world data for both unsupervised LDA and Prior-LDA, a supervised variant of LDA for multi-label classification. Our results show a consistent advantage of our approach over traditional CGS under all experimental conditions, and over CVB0 inference in the majority of conditions. More broadly, our results highlight the importance of averaging over multiple samples in LDA parameter estimation, and the use of efficient computational techniques to do so.

James Foulds, Levi Boyles, Christopher Dubois et al.

In the internet era there has been an explosion in the amount of digital text information available, leading to difficulties of scale for traditional inference algorithms for topic models. Recent advances in stochastic variational inference algorithms for latent Dirichlet allocation (LDA) have made it feasible to learn topic models on large-scale corpora, but these methods do not currently take full advantage of the collapsed representation of the model. We propose a stochastic algorithm for collapsed variational Bayesian inference for LDA, which is simpler and more efficient than the state of the art method. We show connections between collapsed variational Bayesian inference and MAP estimation for LDA, and leverage these connections to prove convergence properties of the proposed algorithm. In experiments on large-scale text corpora, the algorithm was found to converge faster and often to a better solution than the previous method. Human-subject experiments also demonstrated that the method can learn coherent topics in seconds on small corpora, facilitating the use of topic models in interactive document analysis software.