用于图生成的自回归扩散模型 笔记

1 Title

        Autoregressive Diffusion Model for Graph Generation（Lingkai Kong、Jiaming Cui、Haotian Sun、Yuchen Zhuang、B. Aditya Prakash、Chao Zhang）【PMLR 2022】

2 Conclusion        

        This study propose an autoregressive diffusion model for graph generation. Unlike existing methods, we define a node-absorbing diffusion process that operates directly in the discrete graph space. For forward diffusion, it designs a diffusion ordering network, which learns a datadependent node absorbing ordering from graph topology. For reverse generation, it designs a denoising network that uses the reverse node ordering to efficiently reconstruct the graph by predicting the node type of the new node and its edges with previously denoised nodes at a time.Based on the permutation invariance of graph, it shows that the two networks can be jointly trained by optimizing a simple lower bound of data likelihood.

3 Good Sentences

        1、Diffusion-based graph generative models have recently obtained promising results for graph generation. However, existing diffusion-based graph generative models are mostly one-shot generative models that apply Gaussian diffusion in the dequantized adjacency matrix space. Such a strategy can suffer from difficulty in model training, slow sampling speed, and incapability of incorporating constraints.（The difficult problems of previous works has not solve.）
        2、Diffusion models decompose the full complex transformation between noise and real data into man small steps of simple diffusion. Compared with prior deep generative models, diffusion models enjoy both flexibility in modeling architecture and tractability of the model’s probability distributions.（The advantages of Diffusion Model which used in graph generates）
        3、While ARDM offers an efficient and general diffusion framework for discrete data, two key questions remain to be addressed for applying ARDM for graphs: (1) How do we define absorbing states for inter-dependent nodes and edges in graphs without losing the efficiency of ARDM? (2) While ARDM imposes a uniform ordering for arriving at an orderagnostic variational lower bound (VLB) of likelihood, a random ordering fails to capture graph topology（There are still two problems which needed to solve in ARDM）
        4、Compared with the random diffusion ordering, our design has two benefits: (1) We can automatically learn a datadependent node generation ordering which leverages the graph