Tyler Derr
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Tyler Derr has a total of 213 co-authors. Here are the top co-authors by number of shared publications:
- Yu Wang
University of Oregon
20 shared publications - Yuying Zhao
Vanderbilt University, Nashville, TN, USA
11 shared publications - Jiliang Tang
Michigan State University, East Lansing, MI, USA
10 shared publications - Ryan Anthony Rossi
Adobe Research, San Jose, USA
8 shared publications - Yuying Zhao
Vanderbilt University
8 shared publications - Yao Ma
Rensselaer Polytechnic Institute
7 shared publications
Latest Publications
- Open Access
- 6 November 2026
- Published byĀ Association for Computing Machinery (ACM)
- p. 909-919
RAD: Rule-Augmented Relational Anomaly Detection
- 6 November 2026
- Published byĀ Association for Computing Machinery (ACM)
- p. 909-919
Open AccessRAD: Rule-Augmented Relational Anomaly Detection
- Open Access
- 26 October 2026
- Published byĀ ElsevierĀ inĀ Neurocomputing
- Vol.Ā 711
A survey of graph unlearning
- 26 October 2026
- Published byĀ ElsevierĀ inĀ Neurocomputing
- Vol.Ā 711
Open AccessA survey of graph unlearning
- Open Access
- 8 September 2026
- Published byĀ ElsevierĀ inĀ SSRN Electronic Journal
Model Routing for Large Language Models: A Unified Taxonomy and Survey
The landscape of LLMs is increasingly defined not by any single model, but by heterogeneous models that differ in scale, specialization, cost, latency, and reliability. This heterogeneity gives rise to a fundamental question at inference time: which model should be used for a given input? Model routing addresses this by selecting, composing, or orchestrating models at inference time, with the goal of achieving strong task performance while respecting operational constraints. Despite the growing importance and rapid progress, existing works remain fragmented across disparate objectives, input signals, levels of granularity, and learning paradigms, making it difficult to compare approaches or identify which design choices are most appropriate for a given setting. In this work, we study existing LLM routing approaches and propose a few intuitive taxonomies that serve as a fundamental basis for understanding and developing better techniques for model routing. In particular, we unify work on routing by introducing a multi-dimensional taxonomy that decomposes routing along four orthogonal dimensions: routing objectives (why to route), routing inputs (what information informs the decision), routing granularity (at what level routing occurs), and learning regime (how the routing policy is learned). We then use the proposed taxonomies to systematically categorize and compare representative methods. Additionally, we identify key challenges that remain to be addressed and potential opportunities to solve them. Finally, our proposed taxonomies unify existing approaches and distill key facets of the routing problem, providing a foundation for both researchers and practitioners.
- 8 September 2026
- Published byĀ ElsevierĀ inĀ SSRN Electronic Journal
Open AccessModel Routing for Large Language Models: A Unified Taxonomy and Survey
The landscape of LLMs is increasingly defined not by any single model, but by heterogeneous models that differ in scale, specialization, cost, latency, and reliability. This heterogeneity gives rise to a fundamental question at inference time: which model should be used for a given input? Model routing addresses this by selecting, composing, or orchestrating models at inference time, with the goal of achieving strong task performance while respecting operational constraints. Despite the growing importance and rapid progress, existing works remain fragmented across disparate objectives, input signals, levels of granularity, and learning paradigms, making it difficult to compare approaches or identify which design choices are most appropriate for a given setting. In this work, we study existing LLM routing approaches and propose a few intuitive taxonomies that serve as a fundamental basis for understanding and developing better techniques for model routing. In particular, we unify work on routing by introducing a multi-dimensional taxonomy that decomposes routing along four orthogonal dimensions: routing objectives (why to route), routing inputs (what information informs the decision), routing granularity (at what level routing occurs), and learning regime (how the routing policy is learned). We then use the proposed taxonomies to systematically categorize and compare representative methods. Additionally, we identify key challenges that remain to be addressed and potential opportunities to solve them. Finally, our proposed taxonomies unify existing approaches and distill key facets of the routing problem, providing a foundation for both researchers and practitioners.
- Top Cited PapersOpen Access
- 4 September 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Computing Surveys
- Vol.Ā 58 Ā (15) , 1-36
Towards Trustworthy Retrieval Augmented Generation for Large Language Models: A Survey
Retrieval-Augmented Generation (RAG) enhances AI-generated content by integrating external knowledge, improving relevance, and reducing hallucinations. However, RAG also introduces risks related to reliability, safety, privacy, fairness, explainability, and accountability, which impact trustworthiness. While various methods aim to address these concerns, a unified framework is lacking. This survey bridges that gap by presenting a comprehensive roadmap for trustworthy RAG systems. We provide a structured analysis of key challenges, existing solutions, and future directions across these aspects. We further organize the field around how the retrieval and generation stages and the six trustworthiness dimensions interact, demonstrating their synergies and trade-offs through a controlled cross-dimensional analysis. Additionally, we highlight downstream applications where trustworthy RAG can make a significant impact, encouraging further research and adoption in real-world AI systems.
- 4 September 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Computing Surveys
- Vol.Ā 58 Ā (15) , 1-36
Top Cited PapersOpen AccessTowards Trustworthy Retrieval Augmented Generation for Large Language Models: A Survey
Retrieval-Augmented Generation (RAG) enhances AI-generated content by integrating external knowledge, improving relevance, and reducing hallucinations. However, RAG also introduces risks related to reliability, safety, privacy, fairness, explainability, and accountability, which impact trustworthiness. While various methods aim to address these concerns, a unified framework is lacking. This survey bridges that gap by presenting a comprehensive roadmap for trustworthy RAG systems. We provide a structured analysis of key challenges, existing solutions, and future directions across these aspects. We further organize the field around how the retrieval and generation stages and the six trustworthiness dimensions interact, demonstrating their synergies and trade-offs through a controlled cross-dimensional analysis. Additionally, we highlight downstream applications where trustworthy RAG can make a significant impact, encouraging further research and adoption in real-world AI systems.
- Open Access
- 31 August 2026
- Published byĀ Association for Computing Machinery (ACM)
- p. 1-5
Personalized Auto-Research: Towards a True AI Co-Scientist
- 31 August 2026
- Published byĀ Association for Computing Machinery (ACM)
- p. 1-5
Open AccessPersonalized Auto-Research: Towards a True AI Co-Scientist
- Open Access
- 3 August 2026
- Published byĀ ElsevierĀ inĀ SSRN Electronic Journal
Hop-Affinity-Based Graph Decomposition and Adaptive Model Routing for HeterogeneousGraph Learning
Graph Neural Networks (GNNs) are highly effective at capturing structural patterns in graphstructured data; however, most existing architectures employ a single message-passing mechanism across an entire graph, assuming uniform structural behavior. Many real-world networks exhibit a mixture of homophilic and heterophilic regions, where node relationships vary across local neighborhoods. Such structural heterogeneity limits the effectiveness of graph-wide learning strategies and motivates region-aware graph modeling. This paper proposes Hop Affinity based Context Analysis (HACA), a framework that identifies structural contexts and enables region-specific model selection within a graph. HACA performs multi-hop affinity analysis to characterize a node based on neighbourhood hops, distinguishing dominant homophilic, heterophilic and mixed-affinity regions. Based on the identified context, the framework associates specialized GNN models with different regions. A routing mechanism assigns node predictions according to inferred regional characteristics, enabling localized adaptation without imposing a single representation function over the entire graph. The framework is evaluated on benchmark datasets spanning varying graph sizes and homophilyāheterophily distributions. Results indicate that region-aware prediction provides measurable improvements over graph-wide homophilyoriented baselines and competitive performance on heterophilic graphs. Analysis on large-scale heterophilic datasets confirms the effectiveness of region-wise decomposition and specialized model assignment, suggesting a pathway toward adaptive graph learning.
- 3 August 2026
- Published byĀ ElsevierĀ inĀ SSRN Electronic Journal
Open AccessHop-Affinity-Based Graph Decomposition and Adaptive Model Routing for HeterogeneousGraph Learning
Graph Neural Networks (GNNs) are highly effective at capturing structural patterns in graphstructured data; however, most existing architectures employ a single message-passing mechanism across an entire graph, assuming uniform structural behavior. Many real-world networks exhibit a mixture of homophilic and heterophilic regions, where node relationships vary across local neighborhoods. Such structural heterogeneity limits the effectiveness of graph-wide learning strategies and motivates region-aware graph modeling. This paper proposes Hop Affinity based Context Analysis (HACA), a framework that identifies structural contexts and enables region-specific model selection within a graph. HACA performs multi-hop affinity analysis to characterize a node based on neighbourhood hops, distinguishing dominant homophilic, heterophilic and mixed-affinity regions. Based on the identified context, the framework associates specialized GNN models with different regions. A routing mechanism assigns node predictions according to inferred regional characteristics, enabling localized adaptation without imposing a single representation function over the entire graph. The framework is evaluated on benchmark datasets spanning varying graph sizes and homophilyāheterophily distributions. Results indicate that region-aware prediction provides measurable improvements over graph-wide homophilyoriented baselines and competitive performance on heterophilic graphs. Analysis on large-scale heterophilic datasets confirms the effectiveness of region-wise decomposition and specialized model assignment, suggesting a pathway toward adaptive graph learning.
- 18 July 2026
- Published byĀ Springer Nature
- p. 271-298
Reachability-Based Formal Verification of Graph Neural Networks with Node and Edge Features
- 18 July 2026
- Published byĀ Springer Nature
- p. 271-298
Reachability-Based Formal Verification of Graph Neural Networks with Node and Edge Features
- Top Cited PapersOpen Access
- 12 June 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Intelligent Systems and Technology
- Vol.Ā 17 Ā (5) , 1-43
A Survey of Mamba
Deep learning (DL), as a vital technique, has sparked a notable revolution in artificial intelligence (AI), resulting in a great change in human lifestyles. As one of the most representative DL techniques, the Transformer architecture has empowered numerous advanced models, especially the large language models (LLMs) that comprise billions of parameters, becoming a cornerstone in deep learning. Despite the impressive achievements, Transformers still face inherent limitations, particularly the time-consuming inference resulting from the quadratic computation complexity of attention calculation. Recently, a novel architecture named Mamba , drawing inspiration from classical state space models (SSMs), has emerged as a promising alternative for building foundation models, delivering comparable modeling abilities to Transformers while preserving near-linear scalability concerning sequence length. This has sparked an increasing number of studies actively exploring Mamba's potential to achieve impressive performance across diverse domains. Given such rapid evolution, there is a critical need for a systematic review that consolidates existing Mamba-empowered models, offering a comprehensive understanding of this emerging model architecture. In this survey, we therefore conduct an in-depth investigation of recent Mamba-associated studies, covering three main aspects: the advancements of Mamba-based models , the techniques of adapting Mamba to diverse data , and the applications where Mamba can excel . Specifically, we first review the foundational knowledge of various representative deep learning models and the details of Mamba-1&2 as preliminaries. Then, to showcase the significance of Mamba for AI, we comprehensively review the related studies focusing on Mamba modelsā architecture design, data adaptability, and applications. Finally, we present a discussion of current limitations and explore various promising research directions to provide deeper insights for future investigations.
- 12 June 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Intelligent Systems and Technology
- Vol.Ā 17 Ā (5) , 1-43
Top Cited PapersOpen AccessA Survey of Mamba
Deep learning (DL), as a vital technique, has sparked a notable revolution in artificial intelligence (AI), resulting in a great change in human lifestyles. As one of the most representative DL techniques, the Transformer architecture has empowered numerous advanced models, especially the large language models (LLMs) that comprise billions of parameters, becoming a cornerstone in deep learning. Despite the impressive achievements, Transformers still face inherent limitations, particularly the time-consuming inference resulting from the quadratic computation complexity of attention calculation. Recently, a novel architecture named Mamba , drawing inspiration from classical state space models (SSMs), has emerged as a promising alternative for building foundation models, delivering comparable modeling abilities to Transformers while preserving near-linear scalability concerning sequence length. This has sparked an increasing number of studies actively exploring Mamba's potential to achieve impressive performance across diverse domains. Given such rapid evolution, there is a critical need for a systematic review that consolidates existing Mamba-empowered models, offering a comprehensive understanding of this emerging model architecture. In this survey, we therefore conduct an in-depth investigation of recent Mamba-associated studies, covering three main aspects: the advancements of Mamba-based models , the techniques of adapting Mamba to diverse data , and the applications where Mamba can excel . Specifically, we first review the foundational knowledge of various representative deep learning models and the details of Mamba-1&2 as preliminaries. Then, to showcase the significance of Mamba for AI, we comprehensively review the related studies focusing on Mamba modelsā architecture design, data adaptability, and applications. Finally, we present a discussion of current limitations and explore various promising research directions to provide deeper insights for future investigations.
- Open Access
- 9 June 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM AI Letters
- Vol.Ā 1 Ā (3) , 1-6
Referral Required: When AI Blocks Access to Human Doctors
As AI systems evolve from support tools to decision-making elements in clinical workflows, a pivotal shift is occurring in how care is accessed. This paper considers a speculative but increasingly plausible scenario: what if seeing a human physician in the future requires a referral from an AI system? Just as insurers gate access to specialists through primary care referrals, we envision AI care access systemsāpowered by large language models (LLMs)āserving as arbiters of whether patients may proceed to any human clinician. We examine risks along this pathāparticularly around opacity, inequity, and denial at scaleāand propose a roadmap for building systems that support explainability, appealability, and ethical oversight. We introduce the concept of medicolegal LLMs: agents optimized not only for accuracy, but for accountable refusal. Finally, we outline a system-level blueprint for trustworthy AI care access decision systems, integrating evidence grounding, uncertainty quantification, transparent policies, audit logging, and responsive escalation to ensure that automation supports rather than obstructs access to care.
- 9 June 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM AI Letters
- Vol.Ā 1 Ā (3) , 1-6
Open AccessReferral Required: When AI Blocks Access to Human Doctors
As AI systems evolve from support tools to decision-making elements in clinical workflows, a pivotal shift is occurring in how care is accessed. This paper considers a speculative but increasingly plausible scenario: what if seeing a human physician in the future requires a referral from an AI system? Just as insurers gate access to specialists through primary care referrals, we envision AI care access systemsāpowered by large language models (LLMs)āserving as arbiters of whether patients may proceed to any human clinician. We examine risks along this pathāparticularly around opacity, inequity, and denial at scaleāand propose a roadmap for building systems that support explainability, appealability, and ethical oversight. We introduce the concept of medicolegal LLMs: agents optimized not only for accuracy, but for accountable refusal. Finally, we outline a system-level blueprint for trustworthy AI care access decision systems, integrating evidence grounding, uncertainty quantification, transparent policies, audit logging, and responsive escalation to ensure that automation supports rather than obstructs access to care.
- Open Access
- 20 May 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Knowledge Discovery From Data
- Vol.Ā 20 Ā (5) , 1-54
A Survey on Signed Network Reconstruction Modeling and Its Applications
The proliferation in the use of Online Social Networks has revolutionized information sharing and consumption, leading to the development of advanced techniques such as link prediction, recommendation systems, community detection, node classification, and network representation learning. However, the availability and quality of real-world datasets for testing these algorithms pose challenges. Synthetic signed datasets generated through signed network reconstruction models offer alternatives for algorithm testing and experimentation. This survey presents an overview of state-of-the-art signed network reconstruction modeling techniques, evaluates their performance through rigorous experimental analysis, explores real-world applications, discusses challenges and open research problems, and guides future research efforts in the field. By consolidating knowledge and providing insights into existing models, this survey contributes to advancing the understanding and improvement of signed network reconstruction modeling. Various research papers discussed in this survey along with publicly available links to their codes are available at: https://github.com/Aikta-Arya/Signed-Network-Reconstruction-Modeling .
- 20 May 2026
- Published byĀ Association for Computing Machinery (ACM)Ā inĀ ACM Transactions on Knowledge Discovery From Data
- Vol.Ā 20 Ā (5) , 1-54
Open AccessA Survey on Signed Network Reconstruction Modeling and Its Applications
The proliferation in the use of Online Social Networks has revolutionized information sharing and consumption, leading to the development of advanced techniques such as link prediction, recommendation systems, community detection, node classification, and network representation learning. However, the availability and quality of real-world datasets for testing these algorithms pose challenges. Synthetic signed datasets generated through signed network reconstruction models offer alternatives for algorithm testing and experimentation. This survey presents an overview of state-of-the-art signed network reconstruction modeling techniques, evaluates their performance through rigorous experimental analysis, explores real-world applications, discusses challenges and open research problems, and guides future research efforts in the field. By consolidating knowledge and providing insights into existing models, this survey contributes to advancing the understanding and improvement of signed network reconstruction modeling. Various research papers discussed in this survey along with publicly available links to their codes are available at: https://github.com/Aikta-Arya/Signed-Network-Reconstruction-Modeling .