14 Papers Accepted at AAAI-25

Held from 25 February to 4 March in Philadelphia, Pennsylvania, the Thirty-Ninth AAAI Conference on Artificial Intelligence (AAAI-25) aims to promote research in AI and foster scientific exchange between researchers, practitioners, scientists, students, and engineers.

A total of 14 papers from A*STAR Centre for Frontier AI Research (A*STAR CFAR) have been accepted at AAAI-25. Congratulations to the following scientists on this achievement:

Prof Ong Yew Soon, Chief Artificial Intelligence (AI) Scientist and Advisor
Dr Lim Joo Hwee, Principal Scientist
Dr Joey Zhou, Deputy Director and Principal Scientist
Dr Chen Zhenghua, Senior Scientist
Dr Du Jiawei, Senior Scientist
Dr Feng Shanshan, Senior Scientist
Dr Guo Qing, Senior Scientist
Dr Pan Yuangang, Senior Scientist
Dr Emadeldeen Eldele, Scientist
Dr Yan Ming, Senior Scientist
Dr Yin Haiyan, Senior Scientist
Dr Yu Xingrui, Scientist
Dr Zhang Jie, Scientist
Mr Neoh Tzeh Yuan, Engineer

List of accepted papers:

1.	KPL: Training-Free Medical Knowledge Mining of Vision-Language Models Jiaxiang Liu, Tianxiang Hu, Jiawei Du, Ruiyuan Zhang, Joey Tianyi Zhou, Zuozhu Liu This paper proposes Knowledge Proxy Learning (KPL) to address challenges in applying CLIP to zero-shot medical image classification, including inadequate class representation and visual-text modal gaps. KPL enriches semantic proxies with knowledge-relevant descriptions and generates stable multimodal proxies, significantly improving classification performance. Experiments show KPL outperforms baselines, highlighting its potential for medical imaging and beyond.
2.	Hierarchical Classification Auxiliary Network for Time Series Forecasting Yanru Sun, Zongxia Xie, Dongyue Chen, Emadeldeen Eldele, Qinghua Hu HCAN tackles the over-smoothing issue in time-series forecasting caused by the Mean Squared Error (MSE) loss. It enhances feature diversity by reframing the problem as a hierarchical classification task. It integrates hierarchy-aware attention, uncertainty-aware classifier, and a hierarchical consistency loss to capture both fine and coarse-grained time-series representations.
3.	POI Recommendation via Multi-Objective Adversarial Imitation Learning Zhenglin Wan, Anjun Gao, Xingrui Yu, Pingfu Chao, Jun Song, Maohao Ran Point-of-Interest (POI) recommendation aims to predict users’ future locations based on their historical check-ins. This paper introduces Multi-Objective Adversarial Imitation Recommender (MOAIR), a novel framework that integrates Generative Adversarial Imitation Learning with multi objective to address the data sparsity and incompleteness issues in POI recommendation.
4.	An LLM-empowered Adaptive Evolutionary Algorithm for Multi-Component Deep Learning Systems Haoxiang Tian, Xingshuo Han, Guoquan Wu, An Guo, Yuan Zhou, Jie Zhang, Shuo Li, Jun Wei, Tianwei Zhang The paper proposes µMOEA, an LLM-powered adaptive evolutionary algorithm that enhances efficiency and diversity in detecting safety violations in multi-component deep learning systems, outperforming state-of-the-art methods.
5.	SPASCA: Social Presence and Support with Conversational Agent for Persons Living with Dementia (demo paper) Ali Koksal, Jingjing Gu, Kotaro Hara, Jing Jiang, Joo Hwee Lim, Qianli Xu In this paper, we demonstrated an AI-driven interactive avatar that provides engaging conversational experience for persons living with dementia (PLWD) with two key innovative components, a dialogue model to generate speech content that is suitable to PLWD, and a video synthesis model that generates lip-synchronised talking head with proper head motions and facial expressions.
6.	RoDA: Robust Domain Alignment for Cross-domain Retrieval against Label Noise Ziniu Yin, Yanglin Feng, Ming YAN, Xiaomin Song, Dezhong Peng, *Xu Wang* This paper studies the complex challenge of cross-domain image retrieval under the condition of noisy labels (NCIR), a scenario that not only includes the inherent obstacles of traditional cross-domain image retrieval (CIR) but also requires alleviating the adverse effects of label noise.
7.	Balancing Privacy and Performance: A Many-in-One Approach for Image Anonymisation Xuemei Jia, Jiawei Du, Hui Wei, Ruinian Xue, Zheng Wang, Hongyuan Zhu, Jun Chen This paper proposes a novel Recombining for Obfuscation (FRO) approach to generate anonymised data for privacy-preserving. Unlike existing methods that generate one anonymised instance by perturbing the original data on a one- to-one basis, FRO generates an anonymised in-stance by reassembling mixed id-related features from multiple original data sources on a many-in-one basis.
8.	Multi-Edge Reinforced Collaborative Data Acquisition for Continuous Video Analytics by Prioritising Quality over Quantity Lei Zhang, Guanyu Gao, Haiyan Yin, Huaizheng Zhang The paper introduces a multi-edge collaborative active video acquisition (AVA) framework that uses RL to optimise the selection of informative video frames, minimising redundancy across edge devices while addressing data drift in video analytics. By integrating a Transformer-based encoder and an actor-critic RL model, the approach learns a policy that balances data efficiency and model accuracy in real-time. Experimental results show a significant reduction in training data (60%-70%) while maintaining comparable performance, enabling scalable and cost-efficient continuous learning in dynamic environments.
9.	WiFi CSI Based Temporal Activity Detection Via Dual Pyramid Network Zhendong Liu, Le Zhang, Bing Li, Yingjie Zhou, Zhenghua Chen, Ce Zhu The proposed Dual Pyramid Network advances sustainable AI through efficient feature extraction mechanisms like Signed Mask-Attention and ContraNorm. Its scalable design and reusable dataset minimise redundancy, promoting energy-efficient research and resource optimisation.
10.	AsyncDSB: Schedule-Asynchronous Diffusion Schrödinger Bridge for Image Inpainting Zihao Han, Baoquan Zhang, Lisai Zhang, Shanshan Feng,* Kenghong Lin, Guotao Liang, Yunming Ye, Guangming Ye* A novel Asynchronous Schrödinger Diffusion Bridge framework is developed to effectively mitigate schedule-restoration mismatches in image inpainting.
11.	Understanding EFX Allocations: Counting and Variants *Tzeh Yuan Neoh*, Nicholas Teh This paper studied the problem of counting EFX allocations and resolved open problems regarding WEFX allocations.
12.	Max-Mahalanobis Anchors Guidance for Multi-View Clustering *Pei Zhang, Yuangang Pan*, Siwei Wang, Shengju Yu, Huiying Xu, En Zhu, Xinwang Liu, and *Ivor Tsang* Anchor selection or learning is a critical component in large-scale multi-view clustering. We provide formal definitions for the desirable properties (Diversity, Balance, and Compactness) of anchors, revealing the deficiencies of current anchor-based methods. Further, we propose a rational-design anchor strategy, termed Max-Mahalanobis Anchors (MMA), which consistently achieves superior performance in extensive experiments.
13.	Concept Matching with Agent for Out-of-Distribution Detection Yuxiao Lee, Xiaofeng Cao, Jingcai Guo, Wei Ye, Qing Guo, Yi Chang The remarkable achievements of Large Language Models (LLMs) have captivated the attention of both academia and industry, transcending their initial role in dialogue generation. To expand the usage scenarios of LLM, some works enhance the effectiveness and capabilities of the model by introducing more external information, which is called the agent paradigm. Based on this idea, we propose a new method that integrates the agent paradigm into out-of-distribution (OOD) detection task, aiming to improve its robustness and adaptability.
14.	Active Large Language Model-based Knowledge Distillation for Session-based Recommendation Yingpeng Du, Zhu Sun, Ziyan Wang, Haoyan Chua, Jie Zhang, *Yew-Soon Ong* Leveraging the LLM's vast knowledge to enhance the recommendation system while maintaining computational efficiency.

Find out more about AAAI-25.

News

14 Papers Accepted at AAAI-25

A*STAR celebrates International Women's Day