Reproduce “Zero-shot Recognition via Semantic Embeddings and Knowledge Graphs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiaolong Wang, Yufei Ye, Abhinav Gupta
The Robotics Institute at Carnegie Mellon University developed a zero-shot recognition method that integrates semantic embeddings with knowledge graphs using deep Graph Convolutional Networks (GCNs) to directly predict visual classifiers for unseen categories. This approach improves top-10 accuracy by 18.7% on an ImageNet subset and shows robustness to noisy graph data.
Key insights
- Integrating implicit semantic embeddings with explicit relational knowledge from knowledge graphs through deep GCNs dramatically improves zero-shot recognition capabilities.
- Deeper GCN architectures (e.g., 6 layers) are essential for effective message passing across the graph, enabling the model to learn complex transformations for classifier prediction.
- The approach demonstrates robustness to noise and scale in knowledge graphs, effectively leveraging structured information to enhance generalization beyond semantic embeddings alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.