Large language models (LLMs) are increasingly deployed in multi-agent environments. However, the processes by which beliefs form and propagate among interacting LLMs remain poorly understood. We introduce CoevolveSim, a framework for studying belief diffusion within networked LLM populations. CoevolveSim allows us to isolate and study three factors: domain specialization, social-role assignment, and social network structure. Within this framework, generalist and specialist LLM agents exchange and revise beliefs. In each round, an LLM agent observes a summary of its neighbors' beliefs before updating its own. We run 1,280 controlled simulations spanning four scenarios, two network structures, and 20 medical-indication statements. We find that persona-style role assignment and network structure reshape individual belief revision but have minimal effect on population-level consensus. In contrast, introducing (finetuned) specialist LLMs more than doubles the shift in consensus and gives rise to consistent asymmetries in exerted influence. We further show that simple persistence-based opinion-dynamics models reproduce collective outcomes in all-generalist LLM populations, whereas heterogeneous LLM populations require population-level belief composition to reproduce consensus and agent identity to predict individual belief transitions. Our results indicate that realistic simulation of belief diffusion in multi-agent LLM systems requires a diverse set of underlying LLMs, not persona prompting alone.

Assigning agents to teams under strict task and effort constraints is crucial in business, science, and engineering, where disruptions can cause significant losses. Current methods do not explore hypergraph-based solutions that explicitly optimize algebraic connectivity under constraints, leaving unresolved how to systematically form robust, recoverable teams. We present a hypergraph-based team assignment algorithm where nodes represent agents and hyperedges represent tasks. The search is guided by input constraints and aims to optimize resilience and diffusion by maximizing the algebraic connectivity of an edge-dependent, vertex-weighted hypergraph. We employ constrained simulated annealing to find a satisfactory hypergraph by enforcing both the minimum effort required for task completion and the maximum effort agents can exert. We evaluate robustness by assessing solution recovery after node removal attacks. Our results demonstrate that the hypergraph formulation yields more robust solutions than the bipartite formulation and the greedy approach. Assigning entities to teams for task completion under constraints is a complex combinatorial optimization challenge. Here, the authors introduce a novel algorithm using hypergraph-based search to optimize resilience and diffusion, demonstrating superior robustness to node removal attacks compared to traditional methods, with implications for diverse fields requiring resilient team configurations

Faculty hiring shapes the flow of ideas, resources, and opportunities in academia, influencing not only individual career trajectories but also broader patterns of institutional prestige and scientific progress. While prior studies have found strong correlations between faculty hiring and attributes such as doctoral department prestige and publication record, they rarely assess whether these associations generalize to individual hiring outcomes, particularly for future candidates outside the original sample. Here, we consider faculty placement as an individual-level prediction task. Our data consist of temporal coauthorship networks with conventional attributes such as doctoral department prestige and bibliometric features. We observe that using the coauthorship network significantly improves predictive accuracy by up to 25

Recommender systems have been shown to exhibit popularity bias by over-recommending popular items and under-recommending relevant niche items. We seek to understand interactions with niche items in benchmark recommendation datasets as a step toward mitigating popularity bias. We find that, compared to mainstream users, niche-preferring users exhibit a longer-tailed activity-level distribution, indicating the existence of users who both prefer niche items and exhibit high activity levels. We partition users along two axes: (1) activity level ("power" vs. "light") and (2) item-popularity preference ("mainstream" vs. "niche"), and show that in several benchmark datasets, the number of power-niche users (high activity and niche preference) is statistically significantly larger than expected under a null configuration model. Motivated by this observation, we propose a framework for reweighting the Bayesian Personalized Ranking (BPR) loss that simultaneously reweights based on user activity level and item popularity. Our method introduces two interpretable parameters: one controlling the significance of user activity level, and the other of item popularity. Experiments on benchmark datasets show that upweighting power-niche users reduces popularity bias and can increase overall performance. In contrast to previous work that only considers user activity level or item popularity in isolation, our results suggest that considering their interaction leads to Pareto-dominant performance.

Neural ordinary differential equations (neural ODEs) can effectively learn dynamical systems from time series data, but their behavior on graph-structured data remains poorly understood, especially when applied to graphs with different size or structure than encountered during training. We study neural ODEs (𝚗𝙾𝙳𝙴s) with vector fields following the Barabási-Barzel form, trained on synthetic data from five common dynamical systems on graphs. Using the 𝕊^1-model to generate graphs with realistic and tunable structure, we find that degree heterogeneity and the type of dynamical system are the primary factors in determining 𝚗𝙾𝙳𝙴s' ability to generalize across graph sizes and properties. This extends to 𝚗𝙾𝙳𝙴s' ability to capture fixed points and maintain performance amid missing data. Average clustering plays a secondary role in determining 𝚗𝙾𝙳𝙴 performance. Our findings highlight 𝚗𝙾𝙳𝙴s as a powerful approach to understanding complex systems but underscore challenges emerging from degree heterogeneity and clustering in realistic graphs.

We study the fairness of dimensionality reduction methods for recommendations. We focus on the established method of principal component analysis (PCA), which identifies latent components and produces a low-rank approximation via the leading components while discarding the trailing components. Prior works have defined notions of "fair PCA"; however, these definitions do not answer the following question: what makes PCA unfair? We identify two underlying mechanisms of PCA that induce unfairness at the item level. The first negatively impacts less popular items, due to the fact that less popular items rely on trailing latent components to recover their values. The second negatively impacts the highly popular items, since the leading PCA components specialize in individual popular items instead of capturing similarities between items. To address these issues, we develop a polynomial-time algorithm, Item-Weighted PCA, a modification of PCA that uses item-specific weights in the objective. On a stylized class of matrices, we prove that Item-Weighted PCA using a specific set of weights minimizes a popularity-normalized error metric. Our evaluations on real-world datasets show that Item-Weighted PCA not only improves overall recommendation quality by up to 0.1 item-level AUC-ROC but also improves on both popular and less popular items.

Machine learning models for graphs in real-world applications are prone to two primary types of uncertainty: (1) those that arise from incomplete and noisy data and (2) those that arise from uncertainty of the model in its output. These sources of uncertainty are not mutually exclusive. Additionally, models are susceptible to targeted adversarial attacks, which exacerbate both of these uncertainties. In this work, we introduce Radius Enhanced Graph Embeddings (REGE), an approach that measures and incorporates uncertainty in data to produce graph embeddings with radius values that represent the uncertainty of the model's output. REGE employs curriculum learning to incorporate data uncertainty and conformal learning to address the uncertainty in the model's output. In our experiments, we show that REGE's graph embeddings perform better under adversarial attacks by an average of 1.5 against state-of-the-art methods.

A wide range of graph embedding objectives decompose into two components: one that enforces similarity, attracting the embeddings of nodes that are perceived as similar, and another that enforces dissimilarity, repelling the embeddings of nodes that are perceived as dissimilar. Without repulsion, the embeddings would collapse into trivial solutions. Skip-Gram Negative Sampling (SGNS) is a popular and efficient repulsion approach that prevents collapse by repelling each node from a sample of dissimilar nodes. In this work, we show that when repulsion is most needed and the embeddings approach collapse, SGNS node-wise repulsion is, in the aggregate, an approximate re-centering of the node embedding dimensions. Such dimension operations are more scalable than node operations and produce a simpler geometric interpretation of the repulsion. Our theoretical result establishes dimension regularization as an effective and more efficient, compared to skip-gram node contrast, approach to enforcing dissimilarity among embeddings of nodes. We use this result to propose a flexible algorithm augmentation framework that improves the scalability of any existing algorithm using SGNS. The framework prioritizes node attraction and replaces SGNS with dimension regularization. We instantiate this generic framework for LINE and node2vec and show that the augmented algorithms preserve downstream link-prediction performance while reducing GPU memory usage by up to 33.3% and training time by 23.4%. Moreover, we show that completely removing repulsion (a special case of our augmentation framework) in LINE reduces training time by 70.9% on average, while increasing link prediction performance, especially for graphs that are globally sparse but locally dense. Global sparsity slows down dimensional collapse, while local density ensures that node attraction brings the nodes near their neighbors. In general, however, repulsion is needed, and dimension regularization provides an efficient alternative to SGNS.

Large language models (LLMs) are widely used as information sources, yet small changes in semantic assumptions can destabilize their beliefs. We introduce P-StaT (Perturbation Stability of Truth), a framework for evaluating belief stability under matched semantic perturbations in both representational and behavioral settings. Across 21 LLMs and three domains, we compare perturbations involving familiar Fictional statements against synthetically generated and unfamiliar Synthetic statements. Unfamiliar Synthetic perturbations generally induce greater epistemic instability than familiar Fictional perturbations, with behavioral belief retraction rates frequently exceeding 0.5 (50

Machine learning (ML) approaches are increasingly being used to accelerate combinatorial optimization (CO) problems. We investigate the Set Cover Problem (SCP) and propose Graph-SCP, a graph neural network method that augments existing optimization solvers by learning to identify a smaller sub-problem that contains the solution space. Graph-SCP uses both supervised learning from prior solved instances and unsupervised learning to minimize the SCP objective. We evaluate the performance of Graph-SCP on synthetically weighted and unweighted SCP instances with diverse problem characteristics and complexities, and on instances from the OR Library, a canonical benchmark for SCP. We show that Graph-SCP reduces the problem size by 60–80

MOTIVATION:Unraveling the human interactome to uncover disease-specific patterns and discover drug targets hinges on accurate protein-protein interaction (PPI) predictions. However, challenges persist in machine learning (ML) models due to a scarcity of quality hard negative samples, shortcut learning, and limited generalizability to novel proteins. RESULTS:In this study, we introduce a novel approach for strategic sampling of protein-protein noninteractions (PPNIs) by leveraging higher-order network characteristics that capture the inherent complementarity-driven mechanisms of PPIs. Next, we introduce Unsupervised Pre-training of Node Attributes tuned for PPI (UPNA-PPI), a high throughput sequence-to-function ML pipeline, integrating unsupervised pre-training in protein representation learning with Topological PPNI (TPPNI) samples, capable of efficiently screening billions of interactions. By using our TPPNI in training the UPNA-PPI model, we improve PPI prediction generalizability and interpretability, particularly in identifying potential binding sites locations on amino acid sequences, strengthening the prioritization of screening assays and facilitating the transferability of ML predictions across protein families and homodimers. UPNA-PPI establishes the foundation for a fundamental negative sampling methodology in graph machine learning by integrating insights from network topology. AVAILABILITY AND IMPLEMENTATION:Code and UPNA-PPI predictions are freely available at https://github.com/alxndgb/UPNA-PPI.

Human-AI coevolution, defined as a process in which humans and AI algorithms continuously influence each other, increasingly characterises our society, but is understudied in artificial intelligence and complexity science literature. Recommender systems and assistants play a prominent role in human-AI coevolution, as they permeate many facets of daily life and influence human choices through online platforms. The interaction between users and AI results in a potentially endless feedback loop, wherein users' choices generate data to train AI models, which, in turn, shape subsequent user preferences. This human-AI feedback loop has peculiar characteristics compared to traditional human-machine interaction and gives rise to complex and often “unintended” systemic outcomes. This paper introduces human-AI coevolution as the cornerstone for a new field of study at the intersection between AI and complexity science focused on the theoretical, empirical, and mathematical investigation of the human-AI feedback loop. In doing so, we: (i) outline the pros and cons of existing methodologies and highlight shortcomings and potential ways for capturing feedback loop mechanisms; (ii) propose a reflection at the intersection between complexity science, AI and society; (iii) provide real-world examples for different human-AI ecosystems; and (iv) illustrate challenges to the creation of such a field of study, conceptualising them at increasing levels of abstraction, i.e., scientific, legal and socio-political.

Analysis of single-cell RNA sequencing data is often conducted through network projections such as coexpression networks, primarily due to the abundant availability of network analysis tools for downstream tasks. However, this approach has several limitations: loss of higher-order information, inefficient data representation caused by converting a sparse dataset to a fully connected network, and overestimation of coexpression due to zero-inflation. To address these limitations, we propose conceptualizing scRNA-seq expression data as hypergraphs, which are generalized graphs in which the hyperedges can connect more than two vertices. In the context of scRNA-seq data, the hypergraph nodes represent cells and the edges represent genes. Each hyperedge connects all cells where its corresponding gene is actively expressed and records the expression of the gene across different cells. This hypergraph conceptualization enables us to explore multi-way relationships beyond the pairwise interactions in coexpression networks without loss of information. We propose two novel clustering methods: (1) the Dual-Importance Preference Hypergraph Walk (DIPHW) and (2) the Coexpression and Memory-Integrated Dual-Importance Preference Hypergraph Walk (CoMem-DIPHW). They outperform established methods on both simulated and real scRNA-seq datasets. The improvement brought by our proposed methods is especially significant when data modularity is weak. Furthermore, CoMem-DIPHW incorporates the gene coexpression network, cell coexpression network, and the cell-gene expression hypergraph from the single-cell abundance counts data altogether for embedding computation. This approach accounts for both the local level information from single-cell level gene expression and the global level information from the pairwise similarity in the two coexpression networks.

Hypergraphs, which belong to the family of higher-order networks, are a natural and powerful choice for modeling group interactions in the real world. For example, when modeling collaboration networks, which may involve not just two but three or more people, the use of hypergraphs allows us to explore beyond pairwise (dyadic) patterns and capture groupwise (polyadic) patterns. The mathematical complexity of hypergraphs offers both opportunities and challenges for hypergraph mining. The goal of hypergraph mining is to find structural properties recurring in real-world hypergraphs across different domains, which we call patterns. To find patterns, we need tools. We divide hypergraph mining tools into three categories: (1) null models (which help test the significance of observed patterns), (2) structural elements (i.e., substructures in a hypergraph such as open and closed triangles), and (3) structural quantities (i.e., numerical tools for computing hypergraph patterns such as transitivity). There are also hypergraph generators, whose objective is to produce synthetic hypergraphs that are a faithful representation of real-world hypergraphs. In this survey, we provide a comprehensive overview of the current landscape of hypergraph mining, covering patterns, tools, and generators. We provide comprehensive taxonomies for each and offer in-depth discussions for future research on hypergraph mining.

Human-AI coevolution, defined as a process in which humans and AI algorithms continuously influence each other, increasingly characterises our society, but is understudied in artificial intelligence and complexity science literature. Recommender systems and assistants play a prominent role in human-AI coevolution, as they permeate many facets of daily life and influence human choices through online platforms. The interaction between users and AI results in a potentially endless feedback loop, wherein users' choices generate data to train AI models, which, in turn, shape subsequent user preferences. This human-AI feedback loop has peculiar characteristics compared to traditional human-machine interaction and gives rise to complex and often "unintended" systemic outcomes. This paper introduces human-AI coevolution as the cornerstone for a new field of study at the intersection between AI and complexity science focused on the theoretical, empirical, and mathematical investigation of the human-AI feedback loop. In doing so, we: (i) outline the pros and cons of existing methodologies and highlight shortcomings and potential ways for capturing feedback loop mechanisms; (ii) propose a reflection at the intersection between complexity science, AI and society; (iii) provide real-world examples for different human-AI ecosystems; and (iv) illustrate challenges to the creation of such a field of study, conceptualising them at increasing levels of abstraction, i.e., scientific, legal and socio-political.

We often attribute human characteristics to large language models (LLMs) and claim that they "know" certain things. LLMs have an internal probabilistic knowledge that represents information retained during training. How can we assess the veracity of this knowledge? We examine two common methods for probing the veracity of LLMs and discover several assumptions that are flawed. To address these flawed assumptions, we introduce sAwMIL (short for Sparse Aware Multiple-Instance Learning), a probing method that utilizes the internal activations of LLMs to separate statements into true, false, and neither. sAwMIL is based on multiple-instance learning and conformal prediction. We evaluate sAwMIL on 5 validity criteria across 16 open-source LLMs, including both default and chat-based variants, as well as on 3 new datasets. Among the insights we provide are: (1) the veracity signal is often concentrated in the third quarter of an LLM's depth; (2) truth and falsehood signals are not always symmetric; (3) linear probes perform better on chat models than on default models; (4) nonlinear probes may be required to capture veracity signals for some LLMs with reinforcement learning from human feedback or knowledge distillation; and (5) LLMs capture a third type of signal that is distinct from true and false and is neither true nor false. These findings provide a reliable method for verifying what LLMs "know" and how certain they are of their probabilistic internal knowledge.

The spread of information through socio-technical systems determines which individuals are the first to gain access to opportunities and insights. Yet, the pathways through which information flows can be skewed, leading to systematic differences in access across social groups. These inequalities remain poorly characterized in settings involving nonlinear social contagion and higher-order interactions that exhibit homophily. We introduce a generative model for Hypergraphs with Hyperedge Homophily, a hyperedge size-dependent property, and tunable degree distribution, called the H3 model, along with a model for nonlinear social contagion that incorporates asymmetric transmission between in-group and out-group nodes. Using stochastic simulations of a social contagion process on hypergraphs from the H3 model and diverse real-world datasets, we show that the interaction between social contagion dynamics and hyperedge homophily-an effect unique to higher-order networks due to its dependence on hyperedge size-can critically shape group-level differences in information access. By emphasizing how hyperedge homophily shapes interaction patterns, our findings underscore the need to rethink socio-technical system design through a higher-order perspective and suggest that dynamics-informed, targeted interventions at specific hyperedge sizes, embedded in a platform architecture, offer a powerful lever for reducing inequality.

The cyber-threat landscape has evolved tremendously in recent years, with new threat variants emerging daily, and large-scale coordinated campaigns becoming more prevalent. In this study, we propose CELEST (CollaborativE LEarning for Scalable Threat detection, a federated machine learning framework for global threat detection over HTTP, which is one of the most commonly used protocols for malware dissemination and communication. CELEST leverages federated learning in order to collaboratively train a global model across multiple clients who keep their data locally, thus providing increased privacy and confidentiality assurances. Through a novel active learning component integrated with the federated learning technique, our system continuously discovers and learns the behavior of new, evolving, and globally-coordinated cyber threats. We show that CELEST is able to expose attacks that are largely invisible to individual organizations. For instance, in one challenging attack scenario with data exfiltration malware, the global model achieves a three-fold increase in Precision-Recall AUC compared to the local model. We also design a poisoning detection and mitigation method, DTrust, specifically designed for federated learning in the collaborative threat detection domain. DTrust successfully detects poisoning clients using the feedback from participating clients to investigate and remove them from the training process. We deploy CELEST on two university networks and show that it is able to detect the malicious HTTP communication with high precision and low false positive rates. Furthermore, during its deployment, CELEST detected a set of previously unknown 42 malicious URLs and 20 malicious domains in one day, which were confirmed to be malicious by VirusTotal.

Vertex classification using graph convolutional networks is susceptible to targeted poisoning attacks, in which both graph structure and node attributes can be changed in an attempt to misclassify a target node. This vulnerability decreases users' confidence in the learning method and can prevent adoption in high-stakes contexts. Defenses have been proposed, focused on filtering edges before creating the model or aggregating information from neighbors more robustly. This paper considers an alternative: we investigate the ability to exploit network phenomena in the training data selection process to improve classifier robustness. We propose two alternative methods of selecting training data: (1) to select the highest-degree nodes and (2) to select nodes with many connections to the test data. In four real datasets, we show that changing the training set often results in far more perturbations required for a successful attack on the graph structure; often a factor of 2 over the random training baseline. We also run a simulation study in which we demonstrate conditions under which the proposed methods outperform random selection, finding that they improve performance most when homophily is higher, clustering coefficient is higher, node degrees are more homogeneous, and attributes are less informative. In addition, we show that the methods are effective when applied to adaptive attacks, alleviating concerns about generalizability.

Scalable addressing of high-dimensional constrained combinatorial optimization problems is a challenge that arises in several science and engineering disciplines. Recent work introduced novel applications of graph neural networks for solving quadratic-cost combinatorial optimization problems. However, effective utilization of models such as graph neural networks to address general problems with higher-order constraints is an unresolved challenge. This paper presents a framework, HypOp, that advances the state of the art for solving combinatorial optimization problems in several aspects: (1) it generalizes the prior results to higher-order constrained problems with arbitrary cost functions by leveraging hypergraph neural networks; (2) it enables scalability to larger problems by introducing a new distributed and parallel training architecture; (3) it demonstrates generalizability across different problem formulations by transferring knowledge within the same hypergraph; (4) it substantially boosts the solution accuracy compared with the prior art by suggesting a fine-tuning step using simulated annealing; and (5) it shows remarkable progress on numerous benchmark examples, including hypergraph MaxCut, satisfiability and resource allocation problems, with notable run-time improvements using a combination of fine-tuning and distributed training techniques. We showcase the application of HypOp in scientific discovery by solving a hypergraph MaxCut problem on a National Drug Code drug-substance hypergraph. Through extensive experimentation on various optimization problems, HypOp demonstrates superiority over existing unsupervised-learning-based solvers and generic optimization methods. Bolstering the broad and deep applicability of graph neural networks, Heydaribeni et al. introduce HypOp, a framework that uses hypergraph neural networks to solve general constrained combinatorial optimization problems. The presented method scales and generalizes well, improves accuracy and outperforms existing solvers on various benchmarking examples.