Timezone: »

Classification-by-Components: Probabilistic Modeling of Reasoning over a Set of Components
Sascha Saralajew · Lars Holdijk · Maike Rees · Ebubekir Asan · Thomas Villmann

Wed Dec 11 05:00 PM -- 07:00 PM (PST) @ East Exhibition Hall B + C #13

Neural networks are state-of-the-art classification approaches but are generally difficult to interpret. This issue can be partly alleviated by constructing a precise decision process within the neural network. In this work, a network architecture, denoted as Classification-By-Components network (CBC), is proposed. It is restricted to follow an intuitive reasoning based decision process inspired by Biederman's recognition-by-components theory from cognitive psychology. The network is trained to learn and detect generic components that characterize objects. In parallel, a class-wise reasoning strategy based on these components is learned to solve the classification problem. In contrast to other work on reasoning, we propose three different types of reasoning: positive, negative, and indefinite. These three types together form a probability space to provide a probabilistic classifier. The decomposition of objects into generic components combined with the probabilistic reasoning provides by design a clear interpretation of the classification decision process. The evaluation of the approach on MNIST shows that CBCs are viable classifiers. Additionally, we demonstrate that the inherent interpretability offers a profound understanding of the classification behavior such that we can explain the success of an adversarial attack. The method's scalability is successfully tested using the ImageNet dataset.

Author Information

Sascha Saralajew (Dr. Ing. h.c. F. Porsche AG)
Lars Holdijk (Radboud University Nijmegen)
Maike Rees (Dr. Ing. h.c. F. Porsche AG)
Ebubekir Asan (Porsche AG)
Thomas Villmann (University of Applied Sciences Mittweida)

More from the Same Authors