27 research outputs found
MOTIF-Driven Contrastive Learning of Graph Representations
No full textWe propose a MOTIF-driven contrastive framework to pretrain a graph neural network in a self-supervised manner so that it can automatically mine motifs from large graph datasets. Our framework achieves state-of-the-art results on various graph-level downstream tasks with few labels, like molecular property prediction
Recommended from our members
On the Contextual Unfairness of Modern Machine Learning: Graph Neural Networks to Large Language Models
No full textGraph neural networks (GNNs) and large language models (LLMs) have emerged as popular machine learning (ML) models for powering applications such as social recommendation on social media platforms and chat-based assistants, respectively. GNNs and LLMs are both contextual models: GNNs operate on social context in social networks, while LLMs process syntactic and semantic context in language. In conjunction with the proliferation of GNNs and LLMs, there is decreasing trust in the fairness of ML. Unfair ML models cause real-world harm, such as the reinforcement of stereotypes and discrimination in hiring. The unfairness of GNNs is exacerbated by social context (e.g., graph structure, message passing). However, this aspect is not explored in research on the fairness of traditional ML models and requires a deeper principled understanding. Moreover, the open-ended nature of LLM generations can make automatic evaluations of syntactic and semantic context-dependent unfairness difficult.This dissertation tackles technical challenges in addressing the unfairness of GNNs and LLMs. In the first part, we theoretically and empirically investigate different forms of GNN unfairness (i.e., imputation bias, preferential attachment bias, degree bias), and how they are affected by graph structure and the choice of graph filter. We further propose principled metrics and methods to alleviate GNN unfairness. In the second part of this dissertation, we assess the measurement validity of evaluations of LLM misgendering. In the final part, we return to the relatively simple setting of feedforward neural networks, and even in this setting, we identify and tackle major challenges in obtaining a precise analytical theory of how model design choices and data properties contribute to unfairness. Such a theory for GNNs and LLMs could aid in interpreting model outputs and designing stronger evaluation and mitigation methods for unfairness. Overall, this dissertation develops a principled understanding of and addresses the unfairness of modern ML models, towards preventing the further entrenchment of social inequalities and promoting justice
Flight control simulator using optic-flow Hardware-in-the-loop
No full textThe most simplistic acts in nature can be incredibly complex to replicate. Among which, insects in general, are profound micro-machines with evolution as the backbone for their most optimized features. This study focuses on developing an efficient sensing and processing strategy for autonomous flight maneuvers in Atalanta, a bio-inspired robotic-fly that uses unconventional flapping wing propulsion for flights. Visual guidance has proven a very significant role in the fields of autonomous robot navigation. Optical-flow-based solutions provide map-less navigation strategies, especially for miniaturized robots that possess stringent size, weight, and processing (SWaP) constraints. The primary objective focuses on the evaluation of a compact optical mouse sensor (ADNS 9800) for use as an optical-flow sensor in autonomous robots for indoor applications. The sensor is programmed for optic-flow detection and enhanced with suitable optics to improve the existing focal range from 3 mm to a few meters. A customizable MATLAB GUI is developed to create a digital sensing environment for the sensor. After necessary modifications, the sensor has proven to be sensitive under 70% brightness and contrast in the digital optical-flow environment. Despite the limited resolution and field-of-view, the sensor is identified and proven to perform well as an optical-flow sensor. Combinations of these sensors are used to develop a Hardware-in-the-loop (HITL) flight simulator to understand pure optic-flow based controls-strategies. Results indicate that the flight simulator works as intended for indoor situations within an error tolerance in Optic-flow of 4.83% from the actual path.Atalanta projectMechanical Engineering | Precision and Microsystems Engineerin
Theoretical and Empirical Insights into the Origins of Degree Bias in Graph Neural Networks
No full textGraph Neural Networks (GNNs) often perform better for high-degree nodes than low-degree nodes on node classification tasks. This degree bias can reinforce social marginalization by, e.g., privileging celebrities and other high-degree actors in social networks during social and content recommendation. While researchers have proposed numerous hypotheses for why GNN degree bias occurs, we find via a survey of 38 degree bias papers that these hypotheses are often not rigorously validated, and can even be contradictory. Thus, we provide an analysis of the origins of degree bias in message-passing GNNs with different graph filters. We prove that high-degree test nodes tend to have a lower probability of misclassification regardless of how GNNs are trained. Moreover, we show that degree bias arises from a variety of factors that are associated with a node\u27s degree (e.g., homophily of neighbors, diversity of neighbors). Furthermore, we show that during training, some GNNs may adjust their loss on low-degree nodes more slowly than on high-degree nodes; however, with sufficiently many epochs of training, message-passing GNNs can achieve their maximum possible training accuracy, which is not significantly limited by their expressive power. Throughout our analysis, we connect our findings to previously-proposed hypotheses for the origins of degree bias, supporting and unifying some while drawing doubt to others. We validate our theoretical findings on 8 common real-world networks, and based on our theoretical and empirical insights, describe a roadmap to alleviate degree bias.Accepted to NeurIPS 202
Networked Inequality: Preferential Attachment Bias in Graph Neural Network Link Prediction
No full textGraph neural network (GNN) link prediction is increasingly deployed in
citation, collaboration, and online social networks to recommend academic
literature, collaborators, and friends. While prior research has investigated
the dyadic fairness of GNN link prediction, the within-group (e.g., queer
women) fairness and "rich get richer" dynamics of link prediction remain
underexplored. However, these aspects have significant consequences for degree
and power imbalances in networks. In this paper, we shed light on how degree
bias in networks affects Graph Convolutional Network (GCN) link prediction. In
particular, we theoretically uncover that GCNs with a symmetric normalized
graph filter have a within-group preferential attachment bias. We validate our
theoretical analysis on real-world citation, collaboration, and online social
networks. We further bridge GCN's preferential attachment bias with unfairness
in link prediction and propose a new within-group fairness metric. This metric
quantifies disparities in link prediction scores within social groups, towards
combating the amplification of degree and power disparities. Finally, we
propose a simple training-time strategy to alleviate within-group unfairness,
and we show that it is effective on citation, social, and credit networks.Comment: Accepted to ICML 202
LINGUISTIC FUZZY MODELING IN LASER MACHINING QUALITY EVALUATION
No full textComputational method combining the knowledge of skilled operator with appropriate rules will enable the modeling of a powerful Fuzzy~Model which helps far more than what human can do in this modern era of precision manufacturing. The resurgence of interest in computational modeling system over the past few decades has opened many new avenues in its applications. Fuzzy computational modeling leads to greater generality and better rapport with reality. It is driven by the need for methods of analysis and design, which can come to grips with the pervasive imprecision of the real world and exploit the tolerance for imprecision to achieve tractability, robustness and low cost solution. Fuzzy modeling and approximation are the most interesting fields where fuzzy theory can be effectively applied. As far as modeling and approximation is concerned, one can say that the main interest is towards its applications. When one intends to apply fuzzy modeling and approximation to an industrial process, one of the key problems to be solved is to find fuzzy rules. In this research, the inputs are the key variables of the design parameters which generates the singleton output to evaluate the cut edge quality in laser machining. The aim of this scientific research is to design knowledge based linguistic rules, algorithm, architecture & learning ability and further develop fuzzy model for laser machining kerf edge quality prediction. Besides that, the author also investigated the effect of using same rules or different work thicknesses of Mn-Mo pressure vessel plates. The finding shows that, the developed lingual fuzzy model has produced a sound output as it matches closely and agrees well to the experimental result
Weisfeiler and Leman Go Measurement Modeling: Probing the Validity of the WL Test
No full textThe expressive power of graph neural networks is usually measured by
comparing how many pairs of graphs or nodes an architecture can possibly
distinguish as non-isomorphic to those distinguishable by the -dimensional
Weisfeiler-Leman (-WL) test. In this paper, we uncover misalignments between
graph machine learning practitioners' conceptualizations of expressive power
and -WL through a systematic analysis of the reliability and validity of
-WL. We conduct a survey () of practitioners to surface their
conceptualizations of expressive power and their assumptions about -WL. In
contrast to practitioners' beliefs, our analysis (which draws from graph theory
and benchmark auditing) reveals that -WL does not guarantee isometry, can be
irrelevant to real-world graph tasks, and may not promote generalization or
trustworthiness. We argue for extensional definitions and measurement of
expressive power based on benchmarks. We further contribute guiding questions
for constructing such benchmarks, which is critical for graph machine learning
practitioners to develop and transparently communicate our understandings of
expressive power
Strong Model Collapse
No full textWithin the scaling laws paradigm, which underpins the training of large neural networks like ChatGPT and Llama, we consider a supervised regression setting and establish the existance of a strong form of the model collapse phenomenon, a critical performance degradation due to synthetic data in the training corpus. Our results show that even the smallest fraction of synthetic data (e.g., as little as 1\% of the total training dataset) can still lead to model collapse: larger and larger training sets do not enhance performance. We further investigate whether increasing model size, an approach aligned with current trends in training large language models, exacerbates or mitigates model collapse. In a simplified regime where neural networks are approximated via random projections of tunable size, we both theoretically and empirically show that larger models can amplify model collapse. Interestingly, our theory also indicates that, beyond the interpolation threshold (which can be extremely high for very large datasets), larger models may mitigate the collapse, although they do not entirely prevent it. Our theoretical findings are empirically verified through experiments on language models and feed-forward neural networks for images
Stop! In the Name of Flaws: Disentangling Personal Names and Sociodemographic Attributes in NLP
No full textPersonal names simultaneously differentiate individuals and categorize them in ways that are important in a given society. While the natural language processing community has thus associated personal names with sociodemographic characteristics in a variety of tasks, researchers have engaged to varying degrees with the established methodological problems in doing so. To guide future work that uses names and sociodemographic characteristics, we provide an overview of relevant research: first, we present an interdisciplinary background on names and naming. We then survey the issues inherent to associating names with sociodemographic attributes, covering problems of validity (e.g., systematic error, construct validity), as well as ethical concerns (e.g., harms, differential impact, cultural insensitivity). Finally, we provide guiding questions along with normative recommendations to avoid validity and ethical pitfalls when dealing with names and sociodemographic characteristics in natural language processing.Gender Bias in Natural Language Processing Workshop at ACL 202
Factoring the Matrix of Domination: A Critical Review and Reimagination of Intersectionality in AI Fairness
No full textIntersectionality is a critical framework that, through inquiry and praxis,
allows us to examine how social inequalities persist through domains of
structure and discipline. Given AI fairness' raison d'etre of "fairness", we
argue that adopting intersectionality as an analytical framework is pivotal to
effectively operationalizing fairness. Through a critical review of how
intersectionality is discussed in 30 papers from the AI fairness literature, we
deductively and inductively: 1) map how intersectionality tenets operate within
the AI fairness paradigm and 2) uncover gaps between the conceptualization and
operationalization of intersectionality. We find that researchers
overwhelmingly reduce intersectionality to optimizing for fairness metrics over
demographic subgroups. They also fail to discuss their social context and when
mentioning power, they mostly situate it only within the AI pipeline. We: 3)
outline and assess the implications of these gaps for critical inquiry and
praxis, and 4) provide actionable recommendations for AI fairness researchers
to engage with intersectionality in their work by grounding it in AI
epistemology.Comment: To appear at AIES 202
