Don't trust your eyes: on the (un)reliability of feature visualizations

• URL: http://arxiv.org/abs/2306.04719v6
• Date: Thu, 6 Jun 2024 18:05:40 GMT
• Title: Don't trust your eyes: on the (un)reliability of feature visualizations
• Authors: Robert Geirhos, Roland S. Zimmermann, Blair Bilodeau, Wieland Brendel, Been Kim,
• Abstract summary: We show how to trick feature visualizations into showing arbitrary patterns that are completely disconnected from normal network behavior on natural input. We then provide evidence for a similar phenomenon occurring in standard, unmanipulated networks. This can be used as a sanity check for feature visualizations.
• Score: 25.018840023636546
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: How do neural networks extract patterns from pixels? Feature visualizations attempt to answer this important question by visualizing highly activating patterns through optimization. Today, visualization methods form the foundation of our knowledge about the internal workings of neural networks, as a type of mechanistic interpretability. Here we ask: How reliable are feature visualizations? We start our investigation by developing network circuits that trick feature visualizations into showing arbitrary patterns that are completely disconnected from normal network behavior on natural input. We then provide evidence for a similar phenomenon occurring in standard, unmanipulated networks: feature visualizations are processed very differently from standard input, casting doubt on their ability to "explain" how neural networks process natural images. This can be used as a sanity check for feature visualizations. We underpin our empirical findings by theory proving that the set of functions that can be reliably understood by feature visualization is extremely small and does not include general black-box neural networks. Therefore, a promising way forward could be the development of networks that enforce certain structures in order to ensure more reliable feature visualizations.

Related papers

• Coding schemes in neural networks learning classification tasks [52.22978725954347]
  We investigate fully-connected, wide neural networks learning classification tasks. We show that the networks acquire strong, data-dependent features. Surprisingly, the nature of the internal representations depends crucially on the neuronal nonlinearity.
  arXiv  Detail & Related papers  (2024-06-24T14:50:05Z)
• Visualizing Neural Network Imagination [2.1749194587826026]
  In certain situations, neural networks will represent environment states in their hidden activations. Our goal is to visualize what environment states the networks are representing. We define a quantitative interpretability metric and use it to demonstrate that hidden states can be highly interpretable.
  arXiv  Detail & Related papers  (2024-05-10T11:43:35Z)
• Seeing in Words: Learning to Classify through Language Bottlenecks [59.97827889540685]
  Humans can explain their predictions using succinct and intuitive descriptions. We show that a vision model whose feature representations are text can effectively classify ImageNet images.
  arXiv  Detail & Related papers  (2023-06-29T00:24:42Z)
• Topological Understanding of Neural Networks, a survey [0.0]
  We look at the internal structure of neural networks which is usually treated as a black box. We review the significance of different activation functions, types of network architectures associated to them, and some empirical data.
  arXiv  Detail & Related papers  (2023-01-23T22:11:37Z)
• Data-driven emergence of convolutional structure in neural networks [83.4920717252233]
  We show how fully-connected neural networks solving a discrimination task can learn a convolutional structure directly from their inputs. By carefully designing data models, we show that the emergence of this pattern is triggered by the non-Gaussian, higher-order local structure of the inputs.
  arXiv  Detail & Related papers  (2022-02-01T17:11:13Z)
• Dynamic Inference with Neural Interpreters [72.90231306252007]
  We present Neural Interpreters, an architecture that factorizes inference in a self-attention network as a system of modules. inputs to the model are routed through a sequence of functions in a way that is end-to-end learned. We show that Neural Interpreters perform on par with the vision transformer using fewer parameters, while being transferrable to a new task in a sample efficient manner.
  arXiv  Detail & Related papers  (2021-10-12T23:22:45Z)
• Expressive Explanations of DNNs by Combining Concept Analysis with ILP [0.3867363075280543]
  We use inherent features learned by the network to build a global, expressive, verbal explanation of the rationale of a feed-forward convolutional deep neural network (DNN) We show that our explanation is faithful to the original black-box model.
  arXiv  Detail & Related papers  (2021-05-16T07:00:27Z)
• Leveraging Sparse Linear Layers for Debuggable Deep Networks [86.94586860037049]
  We show how fitting sparse linear models over learned deep feature representations can lead to more debuggable neural networks. The resulting sparse explanations can help to identify spurious correlations, explain misclassifications, and diagnose model biases in vision and language tasks.
  arXiv  Detail & Related papers  (2021-05-11T08:15:25Z)
• A neural anisotropic view of underspecification in deep learning [60.119023683371736]
  We show that the way neural networks handle the underspecification of problems is highly dependent on the data representation. Our results highlight that understanding the architectural inductive bias in deep learning is fundamental to address the fairness, robustness, and generalization of these systems.
  arXiv  Detail & Related papers  (2021-04-29T14:31:09Z)
• Visual Pattern Recognition with on On-chip Learning: towards a Fully Neuromorphic Approach [10.181725314550823]
  We present a spiking neural network (SNN) for visual pattern recognition with on-chip learning on neuromorphic hardware. We show how this network can learn simple visual patterns composed of horizontal and vertical bars sensed by a Dynamic Vision Sensor. During recognition, the network classifies the pattern's identity while at the same time estimating its location and scale.
  arXiv  Detail & Related papers  (2020-08-08T08:07:36Z)
• Teaching CNNs to mimic Human Visual Cognitive Process & regularise Texture-Shape bias [18.003188982585737]
  Recent experiments in computer vision demonstrate texture bias as the primary reason for supreme results in models employing Convolutional Neural Networks (CNNs) It is believed that the cost function forces the CNN to take a greedy approach and develop a proclivity for local information like texture to increase accuracy, thus failing to explore any global statistics. We propose CognitiveCNN, a new intuitive architecture, inspired from feature integration theory in psychology to utilise human interpretable feature like shape, texture, edges etc. to reconstruct, and classify the image.
  arXiv  Detail & Related papers  (2020-06-25T22:32:54Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.