Do Invariances in Deep Neural Networks Align with Human Perception?


  • Vedant Nanda University of Maryland, College Park Max Planck Institute for Software Systems (MPI-SWS)
  • Ayan Majumdar Max Planck Institute for Software Systems (MPI-SWS)
  • Camila Kolling Max Planck Institue for Software Systems (MPI-SWS)
  • John P. Dickerson University of Maryland, College Park
  • Krishna P. Gummadi Max Planck Institue for Software Systems (MPI-SWS)
  • Bradley C. Love University College London The Alan Turing Institute
  • Adrian Weller University of Cambridge The Alan Turing Institute



ML: Evaluation and Analysis (Machine Learning), ML: Representation Learning, ML: Transparent, Interpretable, Explainable ML, PEAI: Safety, Robustness & Trustworthiness


An evaluation criterion for safe and trustworthy deep learning is how well the invariances captured by representations of deep neural networks (DNNs) are shared with humans. We identify challenges in measuring these invariances. Prior works used gradient-based methods to generate identically represented inputs (IRIs), ie, inputs which have identical representations (on a given layer) of a neural network, and thus capture invariances of a given network. One necessary criterion for a network's invariances to align with human perception is for its IRIs look 'similar' to humans. Prior works, however, have mixed takeaways; some argue that later layers of DNNs do not learn human-like invariances yet others seem to indicate otherwise. We argue that the loss function used to generate IRIs can heavily affect takeaways about invariances of the network and is the primary reason for these conflicting findings. We propose an adversarial regularizer on the IRI generation loss that finds IRIs that make any model appear to have very little shared invariance with humans. Based on this evidence, we argue that there is scope for improving models to have human-like invariances, and further, to have meaningful comparisons between models one should use IRIs generated using the regularizer-free loss. We then conduct an in-depth investigation of how different components (eg architectures, training losses, data augmentations) of the deep learning pipeline contribute to learning models that have good alignment with humans. We find that architectures with residual connections trained using a (self-supervised) contrastive loss with l_p ball adversarial data augmentation tend to learn invariances that are most aligned with humans. Code:




How to Cite

Nanda, V., Majumdar, A., Kolling, C., Dickerson, J. P., Gummadi, K. P., Love, B. C., & Weller, A. (2023). Do Invariances in Deep Neural Networks Align with Human Perception?. Proceedings of the AAAI Conference on Artificial Intelligence, 37(8), 9277-9285.



AAAI Technical Track on Machine Learning III