News & milestones. 新闻与里程碑。

EquiSSL — Gauge-Equivariant Attention for Rotation-Stable 360° Scene Understanding — is published in ACM Transactions on Graphics (Vol. 45, No. 6, Article 242) as part of the SIGGRAPH Asia 2026 Journal Track. The project page and the code are public.

EquiSSL（Gauge-Equivariant Attention for Rotation-Stable 360° Scene Understanding）正式发表于 ACM Transactions on Graphics（第 45 卷第 6 期，文章 242），入选 SIGGRAPH Asia 2026 期刊轨。项目主页与代码均已公开。

CORE — Less Is More: Minimal Components for Contrastive Learning Delivering Superior Accuracy and Efficiency — is published in IEEE Transactions on Neural Networks and Learning Systems (DOI ↗).

CORE（Less Is More: Minimal Components for Contrastive Learning Delivering Superior Accuracy and Efficiency）发表于 IEEE Transactions on Neural Networks and Learning Systems（DOI ↗）。

SDI — Semantic-Guided Data Augmentation for Self-Supervised Visual Pattern Learning: Leveraging Foundation Model Priors — is accepted by Pattern Recognition (Vol. 160) (DOI ↗).

SDI（Semantic-Guided Data Augmentation for Self-Supervised Visual Pattern Learning: Leveraging Foundation Model Priors）被 Pattern Recognition 接收（第 160 卷）（DOI ↗）。

DAGA — Dynamic Attention-Guided Adaptation for Self-Supervised Vision Transformers — is accepted to NeurIPS 2026. The code is available see the project page; code is on GitHub ↗.

DAGA（Dynamic Attention-Guided Adaptation for Self-Supervised Vision Transformers）被 NeurIPS 2026 接收，详见 项目主页，代码开源于 GitHub ↗。

HTNet — A Self-Supervised Heterogeneous Triple Network for Multi-Modal Data — is published in Neural Networks (Vol. 178) (DOI ↗), on unaligned multi-modal representation learning and cross-modal retrieval.

HTNet（A Self-Supervised Heterogeneous Triple Network for Multi-Modal Data）发表于 Neural Networks（第 178 卷）（DOI ↗），研究未对齐多模态数据的表征学习与跨模态检索。

CMAE — Contrastive Manifold Autoencoder for Efficient Latent Diffusion (Tianjian Zhou and Yishan Li, co-first authors) — is accepted for an Oral presentation at IEEE ICME 2026.

CMAE（Contrastive Manifold Autoencoder for Efficient Latent Diffusion，周天健与李易珊为共同第一作者）被 IEEE ICME 2026 接收，并获 Oral 报告。

P-SLCR — Unsupervised Point Cloud Semantic Segmentation via Prototypes Structure Learning and Consistent Reasoning — is published in AAAI 2026 (DOI ↗), with Lixin Zhan as first author.

P-SLCR（Unsupervised Point Cloud Semantic Segmentation via Prototypes Structure Learning and Consistent Reasoning）发表于 AAAI 2026（DOI ↗），Lixin Zhan 为第一作者。

Two national invention patents on self-supervised feature extraction and on multi-scale semantic enhancement for contrastive learning (CN119540575B and CN119580024B) are granted by the China National Intellectual Property Administration (CNIPA).

两项国家发明专利——基于自监督三重神经网络的图像特征提取方法（CN119540575B）与提升对比学习模型性能的多尺度语义信息增强方法和装置（CN119580024B）——获国家知识产权局（CNIPA）授权。