Leveraging Semantic Embeddings for Safety-Critical Applications
This work addresses safety-critical applications by providing a method for error detection without modifying existing networks, though it appears incremental as it builds on existing semantic embedding techniques.
The paper tackled the problem of improving safety in neural network classifiers by using semantic embeddings for introspection and error detection, resulting in a semantic distance score that achieves near state-of-the-art performance on a traffic sign classifier while being faster to compute.
Semantic Embeddings are a popular way to represent knowledge in the field of zero-shot learning. We observe their interpretability and discuss their potential utility in a safety-critical context. Concretely, we propose to use them to add introspection and error detection capabilities to neural network classifiers. First, we show how to create embeddings from symbolic domain knowledge. We discuss how to use them for interpreting mispredictions and propose a simple error detection scheme. We then introduce the concept of semantic distance: a real-valued score that measures confidence in the semantic space. We evaluate this score on a traffic sign classifier and find that it achieves near state-of-the-art performance, while being significantly faster to compute than other confidence scores. Our approach requires no changes to the original network and is thus applicable to any task for which domain knowledge is available.