革新知能統合研究センター

理研AIPの研究者らが中心となって、深層学習分野のトップ国際会議であるThe Thirteenth International Conference on Learning Representations (ICLR2025)において、ワークショップを開催することとなりました。
皆様の投稿・参加をお待ちしております。

• Workshop Title: Deep Generative Model in Machine Learning: Theory, Principle and Efficacy
• Website: https://delta-workshop.github.io/
• Location: Singapore EXPO
• Date: April 27 or 28, 2025 (TBD)
• Paper Submission Deadline: February 5, 2025 (AOE)
• Topics of Interest:
  Theory Topics:
  • Expressivity of deep generative models: investigating expressivity and performance variations across datasets
  • Optimization and generalization in deep generative models
  • Solving stochastic processes for deep generative models
  • Sampling methods
  • Model stability and convergence analysis
  • Implicit bias and regularization in generative models
  • Robustness and generalization boundaries
  • Latent space geometry and manifold learning
• Application Areas:
  • Improved sampling schemes
  • Adversarial robustness and defense mechanisms
  • Scalability and efficiency in high-dimensional generative modeling
  • Multimodal generative modeling algorithms
  • Structured data modeling
  • Generative models for scientific discovery (AI4Science)