Analysis of Classification-Reconstruction Trade-off in Predictive Coding Networks Theory of Neural Systems


According to the predictive coding theory in neuroscience, the brain works like a hierarchical generative model, continuously predicting incoming stimuli. Learning and inference are achieved by minimizing the difference between predicted and actual stimuli [1].

Computational models of predictive coding, known as predictive coding networks (PCNs), encapsulate this concept within a mathematical structure. In the context of image classification, these models approach prediction as a task of reconstruction, aiming to infer an input image's class label by attempting its reconstruction from that label. Nevertheless, PCNs typically excel in either reconstruction or classification tasks, but not both [2].

Recent work in our group suggests that classification- and reconstruction-driven information must be traded off when integrated into shared representation in deep learning architectures [3]. This trade-off effect suggests that the observed specialization in PCNs might stem from the inherent trade-off between these two types of information, potentially elucidating their limitations in performing both tasks proficiently.

This project aims to explore the potential occurrence of the trade-off effect in PCNs. The candidate will implement a PCN in the line of [4] and analyze the training dynamics from the perspective of the classification-reconstruction trade-off. Additionally, the candidate will implement recently suggested alterations of PCNs and, again, evaluate them from the classification-reconstruction trade-off perspective [2, 5, 6, 7].

- Interest in machine learning and computational neuroscience
- Programming experience with Python and preferably a deep learning framework, e.g., PyTorch
Supervision: Jan Rathjens and Prof. Dr. Laurenz Wiskott









The Institut für Neuroinformatik (INI) is a central research unit of the Ruhr-Universität Bochum. We aim to understand the fundamental principles through which organisms generate behavior and cognition while linked to their environments through sensory systems and while acting in those environments through effector systems. Inspired by our insights into such natural cognitive systems, we seek new solutions to problems of information processing in artificial cognitive systems. We draw from a variety of disciplines that include experimental approaches from psychology and neurophysiology as well as theoretical approaches from physics, mathematics, electrical engineering and applied computer science, in particular machine learning, artificial intelligence, and computer vision.

Universitätsstr. 150, Building NB, Room 3/32
D-44801 Bochum, Germany

Tel: (+49) 234 32-28967
Fax: (+49) 234 32-14210