GIN-Graph: A Generative Interpretation Network for Model-Level Explanation of Graph Neural Networks

One significant challenge of exploiting Graph neural networks (GNNs) in real-life scenarios is that they are always treated as black boxes, therefore leading to the requirement of interpretability. Model-level interpretations explain what patterns maximize probability of predicting to a certain class. However, existing model-level interpretation methods pose several limitations such as generating invalid explanation graphs and requiring extreme fine-tuning on hyperparameters manually. In this paper, we propose a new Generative Interpretation Network for Model-Level Explanation of Graph Neural Networks (GIN-Graph), to generate reliable model-level explanation graphs. The implicit and likelihood-free generative adversarial networks are exploited to construct explanation graphs similar to original graphs, meanwhile maximizing the prediction probability for a certain class by adopting a novel objective function. Experimental results indicate that GIN-Graph can be easily applied to GNN models trained on a variety of graph datasets to create meaningful explanation graphs without requiring extensive fine-tuning on hyperparameters.

North Atlantic Right Whale Contact Call Detection

The North Atlantic right whale (Eubalaena glacialis) is an endangered species. These whales continuously suffer from deadly vessel impacts alongside the eastern coast of North America. There have been countless efforts to save the remaining 350 - 400 of them. One of the most prominent works is done by Marinexplore and Cornell University. A system of hydrophones linked to satellite connected-buoys has been deployed in the whales habitat. These hydrophones record and transmit live sounds to a base station. These recording might contain the right whale contact call as well as many other noises. The noise rate increases rapidly in vessel-busy areas such as by the Boston harbor. This paper presents and studies the problem of detecting the North Atlantic right whale contact call with the presence of noise and other marine life sounds. A novel algorithm was developed to preprocess the sound waves before a tree based hierarchical classifier is used to classify the data and provide a score. The developed model was trained with 30,000 data points made available through the Cornell University Whale Detection Challenge program. Results showed that the developed algorithm had close to 85% success rate in detecting the presence of the North Atlantic right whale.