Inceptionism: Going Deeper into Neural Networks

By Alexander Mordvintsev, Chris Olah, Christopher Olah, Mike Tyka, Michael D. Tyka. 2015

838 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

On the Anatomy of MCMC-based Maximum Likelihood Learning of Energy-Based Models (2019, background): “Previous AM research focuses on maximizing the response of pre-trained classifier networks [7, 24, 26, 29].”
Conditional Image Generation with PixelCNN Decoders (2016, background): “imilarly image processing tasks such as denoising, deblurring, inpainting, super-resolution and colorization rely on generating improved images conditioned on noisy or incomplete data. Neural artwork [18, 5] and content generation represent potential future uses for conditional generation. This pape”
CLIPDraw: Exploring Text-to-Drawing Synthesis through Language-Image Encoders (2021, background): “This framework is often referred to as activation maximization (Erhan et al., 2009; Nguyen et al., 2016; Mordvintsev et al., 2015), where a random image is optimized through backpropogation to increase certain neuron activations of a pretrained network.”
Semantic Image Inpainting with Deep Generative Models (2016, methodology): “Besides the proposed method, one may also consider using D to update y by maximizing D(y), similar to backpropagation in DeepDream [28] or neural style transfer [8].”
Synthesizing the preferred inputs for neurons in neural networks via deep generator networks (2016, background): “Many handdesigned natural image priors have been experimentally shown to improve image quality such as: Gaussian blur [7], α-norm [5, 7, 8], total variation [6, 9], jitter [10, 6, 9], data-driven patch priors [8], center-bias regularization [9], and initializing from mean images [9].”
Interpreting Deep Classifier by Visual Distillation of Dark Knowledge (2018, background): “…for example, by producing images that most activate individual units, ranging from low-level features like edges and textures to middle-level features like patterns and even high-level concepts like objects (Erhan et al., 2009; Mahendran & Vedaldi, 2015; Olah et al., 2015; Nguyen et al., 2015).”
Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV) (2017, methodology): “This technique is also applied to AI-aided art (Mordvintsev et al., 2015).”
Methods for interpreting and understanding deep neural networks (2017, methodology): “[50] used a similar trick as part of the inceptionism technique for visualizing DNNs.”
Causal Interpretability for Machine Learning - Problems, Methods and Evaluation (2020, methodology): “culated by solving the following optimization problem, x∗ = argmax x hij(θ,xˆ ), subject to ||x||2 = 1 (2) Despite this method being used as a tool in providing explanations for higher-layer features [55; 79; 75], it has been reported that due to the complexity of the input distribution, some return”
Deep Learning Stock Volatility with Google Domestic Trends (2015, neutral): “In addition, new visualization techniques have been demonstrated [7] which give users more insight into how artificial neural networks operate.”
Opportunities and obstacles for deep learning in biology and medicine (2017, background): “[348] and applied in subsequent work [335,346,347,349].”