New Research at ICML 2026: Graph Models and Diffusion

🔬 New Research at ICML 2026: Graph Models and Diffusion

At the ICML 2026 conference, cutting-edge research in graph and diffusion models was presented. Among the highlights are GraphPFN from Yandex Research for training graph foundation models and ReCache for optimizing diffusion transformers.

🌍 The works confirm a transition toward universal graph models and open pathways to efficient diffusion scaling.

👤 GraphPFN simplifies graph tasks through in-context learning, while ReCache will accelerate media generation on user devices.

Source 1: https://arxiv.org/abs/2509.21489 Source 2: https://arxiv.org/abs/2606.06060