Journal Paper | November 2022

Invariance of object detection in untrained deep neural networks

Jeonghwan Cheon*, Seungdae Baek*, Se-Bum Paik (*Equally contributed)

Abstract

The ability to perceive visual objects with various types of transformations, such as rotation, translation, and scaling, is crucial for consistent object recognition. In machine learning, invariant object detection for a network is often implemented by augmentation with a massive number of training images, but the mechanism of invariant object detection in biological brains—how invariance arises initially and whether it requires visual experience—remains elusive. Here, using a model neural network of the hierarchical visual pathway of the brain, we show that invariance of object detection can emerge spontaneously in the complete absence of learning. First, we found that units selective to a particular object class arise in randomly initialized networks even before visual training. Intriguingly, these units show robust tuning to images of each object class under a wide range of image transformation types, such as viewpoint rotation. We confirmed that this “innate” invariance of object selectivity enables untrained networks to perform an object-detection task robustly, even with images that have been significantly modulated. Our computational model predicts that invariant object tuning originates from combinations of non-invariant units via random feedforward projections, and we confirmed that the predicted profile of feedforward projections is observed in untrained networks. Our results suggest that invariance of object detection is an innate characteristic that can emerge spontaneously in random feedforward networks.

Highlights

• Object-selective units spontaneously emerge in untrained deep neural networks
• Object selectivity maintains robustly in a wide range of image transformations
• Feedforward model can explain spontaneous emergence of the invariance
• Innate invariance enables invariant object detection without learning to variations

Publication Information

• Invariance of object detection in untrained deep neural networks (2022)
• Jeonghwan Cheon*, Seungdae Baek*, Se-Bum Paik.
• Frontiers in Computational Neuroscience 16
Paper Code Dataset Software Bibtex

Preprint and Proceedings

• • Invariance of object detection in untrained deep neural networks (2022)
  • Jeonghwan Cheon*, Seungdae Baek*, Se-Bum Paik
  • BioRxiv 2022.09.08.507096
  Preprint
• • A model study on the developmental mechanism of spontaneous emergence of viewpoint invariant object detection in untrained bio-inspired neural networks (2022)
  • Jeonghwan Cheon*, Seungdae Baek*, Se-Bum Paik
  • KAIST Undergraduate Research Participant (URP) Program

Workshop Presentations

• • Open Lab Bio and Brain Engineering, KAIST (Department Event), Apr 3, 2023, Daejeon, KR. (Poster)
  • Jeonghwan Cheon*, Seungdae Baek*, Se-Bum Paik.
  • ”Invariance of object detection in untrained deep neural networks”
• • 2022 Winter/Spring KAIST Undergraduate Research Participant (URP) Program Workshop, Aug 24, 2022, Daejeon, KR. (Online, 3rd Prize, KRW 700,000)
  • Jeonghwan Cheon*, Seungdae Baek*, Se-Bum Paik.
  • ”A model study on the developmental mechanism of spontaneous emergence of viewpoint invariant object detection in untrained bio-inspired neural networks”