arXiv subjects

Jiajun Tan

Publications and source records attributed to Jiajun Tan.

8 recordsLinked to original sources

arXiv2025

A Survey on Unlearning in Large Language Models

Large Language Models (LLMs) demonstrate remarkable capabilities, but their training on massive corpora poses significant risks from memorized sensitive information. To mitigate these issues and align with legal standards, unlearning has emerged as a critical technique to selectively erase specific knowledge from LLMs without compromising their overall performance. This survey provides a systematic review of over 180 papers on LLM unlearning published since 2021. First, it introduces a novel taxonomy that categorizes unlearning methods based on the phase in the LLM pipeline of the intervention. This framework further distinguishes between parameter modification and parameter selection strategies, thus enabling deeper insights and more informed comparative analysis. Second, it offers a multidimensional analysis of evaluation paradigms. For datasets, we compare 18 existing benchmarks from the perspectives of task format, content, and experimental paradigms to offer actionable guidance. For metrics, we move beyond mere enumeration by dividing knowledge memorization metrics into 10 categories to analyze their advantages and applicability, while also reviewing metrics for model utility, robustness, and efficiency. By discussing current challenges and future directions, this survey aims to advance the field of LLM unlearning and the development of secure AI systems.

cs.CL↗

arXiv2025

The Mirage of Model Editing: Revisiting Evaluation in the Wild

Despite near-perfect results reported in the literature, the effectiveness of model editing in real-world applications remains unclear. To bridge this gap, we introduce QAEdit, a new benchmark aligned with widely used question answering (QA) datasets, and WILD, a task-agnostic evaluation framework designed to better reflect real-world usage of model editing. Our single editing experiments show that current editing methods perform substantially worse than previously reported (38.5% vs. 96.8%). We demonstrate that it stems from issues in the synthetic evaluation practices of prior work. Among them, the most severe is the use of teacher forcing during testing, which leaks both content and length of the ground truth, leading to overestimated performance. Furthermore, we simulate practical deployment by sequential editing, revealing that current approaches fail drastically with only 1000 edits. This work calls for a shift in model editing research toward rigorous evaluation and the development of robust, scalable methods that can reliably update knowledge in LLMs for real-world use.

cs.CL↗

arXiv2024

Understanding the Collapse of LLMs in Model Editing

Despite significant progress in model editing methods, their application in real-world scenarios remains challenging as they often cause large language models (LLMs) to collapse. Among them, ROME is particularly concerning, as it could disrupt LLMs with only a single edit. In this paper, we study the root causes of such collapse. Through extensive analysis, we identify two primary factors that contribute to the collapse: i) inconsistent handling of prefixed and unprefixed keys in the parameter update equation may result in very small denominators, causing excessively large parameter updates; ii) the subject of collapse cases is usually the first token, whose unprefixed key distribution significantly differs from the prefixed key distribution in autoregressive transformers, causing the aforementioned issue to materialize. To validate our findings, we propose a simple yet effective approach: uniformly using prefixed keys during editing phase and adding prefixes during testing phase to ensure the consistency between training and testing. The experimental results show that the proposed solution can prevent model collapse while maintaining the effectiveness of the edits.

cs.CL↗

arXiv2024

Unlink to Unlearn: Simplifying Edge Unlearning in GNNs

As concerns over data privacy intensify, unlearning in Graph Neural Networks (GNNs) has emerged as a prominent research frontier in academia. This concept is pivotal in enforcing the \textit{right to be forgotten}, which entails the selective removal of specific data from trained GNNs upon user request. Our research focuses on edge unlearning, a process of particular relevance to real-world applications. Current state-of-the-art approaches like GNNDelete can eliminate the influence of specific edges yet suffer from \textit{over-forgetting}, which means the unlearning process inadvertently removes excessive information beyond needed, leading to a significant performance decline for remaining edges. Our analysis identifies the loss functions of GNNDelete as the primary source of over-forgetting and also suggests that loss functions may be redundant for effective edge unlearning. Building on these insights, we simplify GNNDelete to develop \textbf{Unlink to Unlearn} (UtU), a novel method that facilitates unlearning exclusively through unlinking the forget edges from graph structure. Our extensive experiments demonstrate that UtU delivers privacy protection on par with that of a retrained model while preserving high accuracy in downstream tasks, by upholding over 97.3\% of the retrained model's privacy protection capabilities and 99.8\% of its link prediction accuracy. Meanwhile, UtU requires only constant computational demands, underscoring its advantage as a highly lightweight and practical edge unlearning solution.

cs.LG↗

arXiv2022

Curvature Graph Generative Adversarial Networks

Generative adversarial network (GAN) is widely used for generalized and robust learning on graph data. However, for non-Euclidean graph data, the existing GAN-based graph representation methods generate negative samples by random walk or traverse in discrete space, leading to the information loss of topological properties (e.g. hierarchy and circularity). Moreover, due to the topological heterogeneity (i.e., different densities across the graph structure) of graph data, they suffer from serious topological distortion problems. In this paper, we proposed a novel Curvature Graph Generative Adversarial Networks method, named \textbf{\modelname}, which is the first GAN-based graph representation method in the Riemannian geometric manifold. To better preserve the topological properties, we approximate the discrete structure as a continuous Riemannian geometric manifold and generate negative samples efficiently from the wrapped normal distribution. To deal with the topological heterogeneity, we leverage the Ricci curvature for local structures with different topological properties, obtaining to low-distortion representations. Extensive experiments show that CurvGAN consistently and significantly outperforms the state-of-the-art methods across multiple tasks and shows superior robustness and generalization.

cs.LG↗

arXiv2021

ACE-HGNN: Adaptive Curvature Exploration Hyperbolic Graph Neural Network

Graph Neural Networks (GNNs) have been widely studied in various graph data mining tasks. Most existingGNNs embed graph data into Euclidean space and thus are less effective to capture the ubiquitous hierarchical structures in real-world networks. Hyperbolic Graph Neural Networks(HGNNs) extend GNNs to hyperbolic space and thus are more effective to capture the hierarchical structures of graphs in node representation learning. In hyperbolic geometry, the graph hierarchical structure can be reflected by the curvatures of the hyperbolic space, and different curvatures can model different hierarchical structures of a graph. However, most existing HGNNs manually set the curvature to a fixed value for simplicity, which achieves a suboptimal performance of graph learning due to the complex and diverse hierarchical structures of the graphs. To resolve this problem, we propose an Adaptive Curvature Exploration Hyperbolic Graph NeuralNetwork named ACE-HGNN to adaptively learn the optimal curvature according to the input graph and downstream tasks. Specifically, ACE-HGNN exploits a multi-agent reinforcement learning framework and contains two agents, ACE-Agent andHGNN-Agent for learning the curvature and node representations, respectively. The two agents are updated by a NashQ-leaning algorithm collaboratively, seeking the optimal hyperbolic space indexed by the curvature. Extensive experiments on multiple real-world graph datasets demonstrate a significant and consistent performance improvement in model quality with competitive performance and good generalization ability.

cs.LG↗

arXiv2018

Interstellar Scintillation observations for PSR B0355+54

In this paper, we report our investigation of pulsar scintillation phenomena by monitoring PSR B0355$+$54 at 2.25 GHz for three successive months using \emph{Kunming 40-m radio telescope}. We have measured the dynamic spectrum, the two-dimensional correlation function, and the secondary spectrum. In those observations with high signal-to-noise ratio ($S/N\ge100$), we have detected the scintillation arcs, which are rarely observable using such a small telescope. The sub-microsecond scale width of the scintillation arc indicates that the transverse scale of structures on scattering screen is as compact as AU size. Our monitoring has also shown that both the scintillation bandwidth, timescale, and arc curvature of PSR B0355$+$54 were varying temporally. The plausible explanation would need to invoke multiple-scattering-screen or multiple-scattering-structure scenario that different screens or ray paths dominate the scintillation process at different epochs.

astro-ph.GA↗

arXiv2013

Maxwell's Fishpond

Most of us will have at some time thrown a pebble into water, and watched the ripples spread outwards and fade away. But now there is a way to reverse the process, and make those ripples turn around and reconverge again, ... and again, and again. To do this we have designed the Maxwell's Fishpond, a water wave or "Transformation Aquatics" version of the Maxwell's Fisheye lens [Tyc et al. 2011, Luneberg 1964] that is now well-known from transformation optics. These are transformation devices where wave propagation on the surface of a sphere is not modelled on an actual sphere, but in a flat device with carefully designed spatially varying properties. And just as for rays from a point source on a sphere, a wave disturbance in a Maxwell's Fisheye or Fishpond spreads out at first, but then reforms itself at its opposing point in the device. Here we show how such a device can be made for water waves, partly in friendly competition with comparable electromagnetic devices [Ma et al. 2011] and partly as an accessible and fun demonstration of the power of transformation mechanics. To the eye, our Maxwell's Fishpond was capable of reforming a disturbance up to five times, although such a feat required taking considerable care, close observation, and a little luck. What can you see in the video at http://www.qols.ph.ic.ac.uk/~kinsle/files/MFishpond/ ?

physics.class-ph↗