Evaluating the Perception, Understanding, and Forgetting of Progressive Neural Networks: A Quantitative and Qualitative Analysis

L. Güitta-López, J. Boal, A.J. López López

AI Vol. 7, nº. 4, pp. 120

Summary:

The use of virtual environments to collect the experience required by deep reinforcement learning models is accelerating the deployment of these algorithms in industrial environments. However, once the experience-gathering problem is solved, it is necessary to address how to efficiently transfer the knowledge from the virtual scenario to reality. This paper focuses on examining Progressive Neural Networks (PNNs) as a promising transfer learning technique. The analyses carried out range from studying the capabilities and limits of the layers responsible for learning the state representation from a pixel space, which could arguably be the convolutional blocks, to the forgetting agents suffer when learning a new task. Introducing controlled visual changes in the environment scene can lead to a performance degradation of 50.3% in the worst-case scenario. These visual discrepancies significantly impact the agent’s learning time and accuracy when using a PNN architecture. Regarding the PNN forgetting assessment, partial forgetting occurs in two of the three environments analyzed, those where the agent masters its new task. This could be due to a balance between the relevance of the new features learned and the ones inherited from the teacher agent.

Spanish layman's summary:

El uso de entornos virtuales facilita el entrenamiento de modelos de aprendizaje por refuerzo profundo, pero trasladar ese conocimiento a la realidad sigue siendo un reto. Este estudio analiza las Redes Neuronales Progresivas (PNNs) como técnica de transferencia, evaluando sus límites y el impacto de cambios visuales. Se observa que estas discrepancias pueden reducir el rendimiento hasta un 50,3% y provocar olvido parcial en algunos entornos.

English layman's summary:

Virtual environments accelerate training for deep reinforcement learning, but transferring knowledge to real-world scenarios remains challenging. This paper studies Progressive Neural Networks (PNNs) as a transfer learning approach, analyzing their limits and sensitivity to visual changes. Results show up to a 50.3% performance drop and partial forgetting in certain environments.

Keywords: deep reinforcement learning; progressive neural networks; sim-to-real; sample efficiency; representation learning

JCR-JIF Impact Factor and WoS quartile: 6,500 - Q1 (2025)

DOI reference: https://doi.org/10.3390/ai7040120

Published on paper: April 2026.

Published on-line: March 2026.

Citation:
L. Güitta-López, J. Boal, A.J. López López, "Evaluating the Perception, Understanding, and Forgetting of Progressive Neural Networks: A Quantitative and Qualitative Analysis", AI, Vol. 7, nº. 4, pp. 120, April 2026. [Online: March 2026] doi: 10.3390/ai7040120