Deep-State-Machine: Generative Model for Graph Generation Thesis Book

- Deep Auto-Regressive Machines for Graph Construction

Deep State Machine (DSM): Envisioned a novel appraoch for learning and generating graphs.

- Multiple states
- Complex states
- Customizable states
- Graph Deconstruction Tree Model (GDTM) for construction sequence generation
- GDTM: A non-deterministic approach for traversing graph
- Unparameterized construction sequences
- Learning complex embeddings in single state
- Learn and generate complex structures of graph in single state
- Graph generation in fewer steps
- DSM combined with GDTM to learn various alternative paths of graph generation

Graph Deconstruction Tree Model:

- Non-deterministic approach of graph traversal
- Navigates several alternative paths through which graphs can be generated
- Generates construction sequence through deconstruction & construction method
- Policy of randomly transitioning between simple to complex valid decision operations

DGMG, DeepGG, GraphRNN: Previous Research

- Basic two states, [add node, add edge]
- Simple and non-customizable states
- Traversal algorithm: bfs, dfs,...
- Parameterized construction sequencess
- Learning embeddings of either node or edge in a single state