Dual-Pathway Neural Networks: Harnessing Scene and Object Pathways for Enhanced Visual Understanding
Keywords: scene and object learning, disentangled representations, generalization
TL;DR: A dual-pathway neural network architecture inspired by human cognition which separates the processing of scene and object information and fuses them in a complemantary fashion for improved generalization and robustness.
Abstract: Standard artificial neural networks (ANNs) often struggle with generalization due to their reliance on surface-level cues, which can lead to suboptimal performance. Drawing inspiration from the distinct processing pathways for scenes and objects in the human brain, we explore the interactions between scene and object and introduce a dual-modality architecture aimed at emulating this cognitive processing mechanism within ANNs. Our approach features separate encodings for scene and object modalities, which are fused to facilitate enhanced visual understanding. By optimizing object recognition and scene reconstruction objectives, our architecture efficiently encodes scene and object information crucial for holistic representation learning. Empirical validation demonstrates significant improvements in generalization, lifelong learning, and adversarial robustness compared to conventional architectures. These findings underscore the potential of integrating biological insights into AI systems to bridge the gap between artificial and biological intelligence.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8324
Loading