Hierarchical representations in deep neural networks

Last updated: 18.03.2022

Short introduction: Explainability in AI and interpretability in Machine learning are very active areas. However, one of the key concepts in these domains, namely the hierarchical latent representations of deep neural networks and their characteristics, is far from being simple to understand. I propose here a brief definition, extracted from a scientific article. You can find the original article at this link and cited at the end, 

One key property of the Representation learning domain in deep neural networks is the ability to provide both high-level features and low-level features for the same learned data. Recall that a deep neural network will encode a latent representation at each hidden layer. Since the layer n units can be all or partially connected to the layer n + 1 units, each layer uses the previous layer as input. If the previous layer is a hidden layer, then the input is already a latent representation, i.e. an abstract feature representation that characterizes the data. Thus, each layer extracts an abstract feature representation of the previous layer. As a result, a deep neural network learns multiple levels of abstraction and implicitly encodes a hierarchy of latent and abstract representations that are built progressively, layer by layer. The layers that are close to the input layer will encode a low-level feature representation, whereas those deeper inside the architecture will encode a high-level feature representation. In other words, the closer the considered layer is to the output layer, the more the representation is abstract [Bengio et al., 2013; Zhong et al., 2016; Lesort et al., 2018], as represented in the figure below.

Illustrative and schematic representation of the position of a low level representation and a high level representation in a deep neural network. hx refers to the x th hidden layer in the network.

It has also been shown that, in deep learning algorithms, hidden representations tend to keep dominant information and propagate them across hidden layers, regardless of the width or depth increase of the deep neural networks [Nguyen et al.,2021]. This characteristic of RL is also a key one for XAI: by extracting and comparing the low-level and the high-level representations of a deep architecture, we consider that it is possible to explicit the inner mechanism of the architecture by observing the differences between the representations

To cite this article, please cite the original scientific paper: 

Ikram Chraibi Kaadoud, Lina Fahed, Philippe Lenca. Explainable AI: a narrative review at the crossroad of Knowledge Discovery, Knowledge Representation and Representation Learning. Twelfth International Workshop Modelling and Reasoning in Context (MRC) @IJCAI 2021, Aug 2021, Montréal (virtual), Canada. pp.28-40. ⟨hal-03343687⟩


  • [Bengio et al., 2013]  Y. Bengio, A. Courville, and P. Vincent. Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence, 35(8):1798–1828, 2013.
  • [Lesort et al., 2018] Timothee Lesort, Natalia DiazRodriguez, Jean-Franc¸ois Goudou, and David Filliat. State Representation Learning for Control: An Overview. Neural Networks, 108:379–392, December 2018.
  • [Nguyen et al.,2021] Thao Nguyen, Maithra Raghu, and Simon Kornblith. Do wide and deep networks learn the same things? Uncovering how neural network representations vary with width and depth. In 9th International Conference on Learning Representations, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021.
  • [Zhong et al., 2016] Guoqiang Zhong, Li-Na Wang, Xiao Ling, and Junyu Dong. An overview on data representation learning: From traditional feature learning to recent deep learning. The Journal of Finance and Data Science, 2(4):265–278, 2016.