Graph Memory Learning: Imitating Lifelong Remembering and Forgetting of Brain Networks

Jiaxing Miao, Liang Hu, Qi Zhang, Longbin Cao

This paper introduces a new concept of graph memory learning. Its core idea is to enable a graph model to selectively remember new knowledge but forget old knowledge. Building on this idea, the paper presents a novel graph memory learning framework Brain-inspired Graph Memory Learning (BGML), inspired by brain network dynamics and function-structure coupling strategies.

Reliable Curriculum Unlearning via the Right Forgotten Pathway

Jiaxing Miao, Liang Hu, Qi Zhang, Zhongyuan Lai, Tangwei Ye and Usman Naseem

This paper proposes CUFG (Curriculum Unlearning via Forgetting Gradients), a novel framework that enhances the stability of approximate unlearning through innovations in both forgetting mechanisms and data scheduling strategies. Specifically, CUFG integrates a new gradient corrector guided by forgetting gradients for fine-tuning-based unlearning and a curriculum unlearning paradigm that progressively forgets from easy to hard.

Triplet teaching graph contrastive networks with self-evolving adaptive augmentation

Jiaxing Miao, Feilong Cao, Ming Li, Bing Yang, Hailiang Ye

Building on the teaching concept, this paper proposes a novel triplet-based teaching graph contrastive network with self-evolving adaptive augmentation (T-GCSA). It addresses the critical challenges of generating reasonable augmented views, utilizing them effectively, and constructing efficient, comprehensive contrastive objectives.

Revisiting graph neural networks from hybrid regularized graph signal reconstruction

Jiaxing Miao, Feilong Cao, Hailiang Ye, Ming Li, Bing Yang,

This paper presents a unified optimization framework from hybrid regularized graph signal reconstruction to establish the connection between the aggregation operations of different GNNs. We use this new framework to mathematically explain several classic GNN models and summarizes their commonalities and differences from a macro perspective. Based on this, we design the new model, GNN based on model-driven and data-driven (GNN-MD). We also theoretically analyze its convergence.

[A Durable Unlearning Enhancement Framework to Nullify Recall of Sensitive Data on Incremental Training]

Qingqing Cao, Liang Hu, Dora Dongmei Liu, Jiaxing Miao, Zhongyuan Lai, Cao Jian, Wei Cao

This work addresses the problem of post-unlearning knowledge recall, where previously forgotten sensitive information can be inadvertently restored during incremental model updates. The proposed Durable Unlearning Enhancement (DUE) framework detects sensitive samples in new data and suppresses their gradients to prevent the re-emergence of unlearned knowledge.

Causal-Inference Machine Learning for Unveiling Hydroxyl Radical Reactivity with Antibiotics in Water Purification

Shihua Zou, Zonglin Li, Yicen Dai, Jiaxing Miao, Zhiyu Zhao, Liang Hu, Hongying Zhao

This work develops an interpretable machine learning framework for AI-driven chemistry to uncover the intrinsic molecular factors governing HO· reactivity with antibiotic pollutants. By integrating DFT descriptors and attention-based features, the model identifies key molecular drivers of reactivity. The causal-interface model uncovers intrinsic cause–effect relationships and enables reliable mechanistic insight.

Accurate Trajectory Recovery in Underserved Areas via Location Inference from Web Crowdsourced Data

Tangwei Ye, Liang Hu, Zhongyuan Lai, Qi Zhang, Yiming Wu, Jiaxing Miao, Yijun Yang, Kun Yi

This paper introduces Region-aware Hierarchical Trajectory Recovery (RHTR), a region-aware framework that leverages web crowdsourced data and a coarse-to-fine multi-scale representation to recover trajectories in roadless and data-sparse regions without relying on explicit road networks or reliable GPS signals.

AMID: Model-Agnostic Dataset Distillation by Adversarial Mutual Information Minimization

Aoqi Wu, Junming Liu, Evelyn Zhang, Weiquan Huang, Yifan Yang, Jiaxing Miao, Qi Zhang, Lai Zhong Yuan, Liang Hu

This paper introduces an information-theoretic approach to model-agnostic dataset distillation. It proposes Adversarial Mutual Information Distillation (AMID), which minimizes architectural bias via adversarial learning, enabling strong cross-architecture generalization.