Yanbang Wang

I am a fourth-year Ph.D. candidate in Computer Science at Cornell University, extremely fortunate to be advised by Prof. Jon Kleinberg.

Prior to Cornell, I obtained my M.S. in Computer Science from Stanford University, where I worked with Prof. Jure Leskovec. I received my B.S. in Computer Science and Mathematics, graduating Summa Cum Laude (top 1%), from Hong Kong University of Science and Technology.

I visited the ALFA Group at MIT CSAIL in 2018, and interned at Meta AI in 2023 and at Microsoft Research in 2022. I am serving as an organizer for the Learning on Graphs (LoG) Conference.

My name is pronounced "Yen-bahng Wong".

 Google Scholar/  CV /  Email  /  LinkedIn / Twitter

I research machine learning and graph mining algorithms in large-scale social and information systems, as well as their broader societal implications.

• [10/2024] We are organizing the NYC Meetup for Learning on Graphs (LoG) conference on Nov 21-22.
• [05/2024] I start a research internship at Meta this summer.
• [05/2024] Two submissions accepted as oral at IC2S2'24.
• [03/2024] I have joined the organizing commmitte of the Learning on Graphs Conference (LoG) 2024/25 .
• [12/2023] Accepted at ICLR'24: Paper on hypergraph projection and reconstruction.
• [09/2023] Accepted at NeurIPS'23: Paper on how recommender systems influence opinions in large-scale social systems.
• [09/2023] Our proposal has been awarded the Microsoft Accelerating Foundation Models Research (AFMR) Grant .
• [01/2023] Our Graph Machine Learning chapter is now online as a part of our AI for Science blog series.

Accurate graph recall from text is essential for LLMs to perform reasoning tasks involving graph-structured information. While human graph recall has been studied extensively and shows structural biases aligned with social relationships, little is known about how LLMs handle similar tasks. This work presents the first systematic study of graph recall by LLMs, revealing that they often underperform and tend to favor microstructures like triangles and alternating 2-paths. Additionally, more advanced LLMs show improved recall accuracy when the graph is narrated in a language style consistent with its original domain.

We study the implications of the modeling choice to use a graph, instead of a hypergraph, to represent real-world interconnected systems whose constituent relationships are of higher order by nature. Such a modeling choice typically involves an underlying projection process that maps the original hypergraph onto a graph, and is common in graph-based analysis. While hypergraph projection can potentially lead to loss of higher-order relations, there exists very limited studies on the consequences of doing so, as well as its remediation.

Link recommendations in online social networks help users discover connections, enhancing engagement. However, the added links can also escalate conflicts and polarization. There's limited understanding of the relationship between these two outcomes. This work provides one of the first rigorous analysis of this question, using the recently popular Friedkin-Johnsen model of opinion dynamics.

We propose Causal Anonymous Walks (CAWs) for inductive representation learning in temporal networks. CAWs are extracted by temporal random walks and work as automatic retrieval of temporal network motifs to represent network dynamics, while avoiding the time-consuming selection and counting of those motifs. CAWs adopt a novel anonymization strategy that replaces node identities with the hitting counts of the nodes based on a set of sampled walks to keep the method inductive, and simultaneously establish the correlation between motifs. CAWs is the SOTA method on both transductive and inductive link prediction tasks in temporal networks.

We propose a novel framework, termed Temporal Network-diffusion Convolutional Networks (TEDIC), for generic representation learning on dynamic social interaction networks. TEDIC adopts diffusion of node attributes over a combination of the original network and its complement to capture long-hop interactive patterns embedded in the behaviors of people making or avoiding contact. It also leverages temporal convolution networks with hierarchical set-pooling operation to flexibly extract patterns from different-length interactions scattered over a long time span. TEDIC is evaluated over four social character prediction tasks: detecting deception detection, dominance, nervousness detection and community. It consistently outperforms previous SOTAs and provides interesting social insights.

We propose and mathematically analyze a general class of structure-related features, termed Distance Encoding (DE). DE assists GNNs in representing any set of nodes, while providing strictly more expressive power than the 1-Weisfeiler-Lehman test. We also prove that DE can distinguish node sets embedded in almost all regular graphs where traditional GNNs always fail.

The high false positive (FP) rate of authentication alerts remains to be a prominent challenge in cybersecurity nowadays. We identify two problems that cause this issue, which are unaddressed in existing learning-based anomaly detection methods. We address these problems by proposing a new framework based on self-supervised link prediction on dynamic authentication networks. We validate our framework on 4 months of authentication data of 125 randomly selected, real organizations that subscribe to Microsoft's defense services. Our model is intended to be integrated into Microsoft Security products.

We present a new organization of MOOC learner activity data at a resolution that is in between the fine granularity of the clickstream and coarse organizations that count activities, aggregate students or use long duration time units. A detailed access trajectory (DAT) consists of binary values and is two dimensional with one axis that is a time series, and the other that is a chronologically ordered list of a MOOC component type's instances, videos in instructional order, for example. Four empirical mini-studies suggest that DATs contain rich information about students' learning behaviors and facilitate MOOC learning analyses.

This guy makes a nice webpage.