Exemplary Natural Images Explain CNN Activations Better than State-of-the-Art Feature Visualization
May 3, 2021
Speakers
Judy Borowski
Speaker · 0 followers
Roland S. Zimmermann
Speaker · 0 followers
Judith Schepers
Speaker · 0 followers
Robert Geirhos
Speaker · 0 followers
Thomas Wallis
Speaker · 0 followers
Matthias Bethge
Speaker · 0 followers
Wieland Brendel
Speaker · 1 follower
About
Feature visualizations such as synthetic maximally activating images are a widely used explanation method to better understand the information processing of convolutional neural networks (CNNs). At the same time, there are concerns that these visualizations might not accurately represent CNNs' inner workings. Here, we measure how much extremely activating images help humans to predict CNN activations. Using a well-controlled psychophysical paradigm, we compare the informativeness of synthetic images by Olah et al. (2017) with a simple baseline visualization, namely exemplary natural images that also strongly activate a specific feature map. Given either synthetic or natural reference images, human participants choose which of two query images leads to strong positive activation. The experiment is designed to maximize participants' performance, and is the first to probe \emph{intermediate} instead of final layer representations. We find that synthetic images indeed provide helpful information about feature map activations ($82\pm4\%$ accuracy; chance would be $50\%$). However, natural images---originally intended to be a baseline---outperform these synthetic images by a wide margin ($92\pm2\%$ accuracy). Additionally, participants are faster and more confident for natural images, whereas subjective impressions about the interpretability of the feature visualizations by Olah et al. (2017) are mixed. The higher informativeness of natural images holds across most layers, for both expert and lay participants as well as for hand- and randomly-picked feature visualizations. Even if only a single reference image is given, synthetic images provide less information than natural images ($65\pm5\%$ vs. $73\pm4\%$). In summary, synthetic images from a popular feature visualization method are significantly less informative for assessing CNN activations than natural images. We argue that visualization methods should improve over this simple baseline.Feature visualizations such as synthetic maximally activating images are a widely used explanation method to better understand the information processing of convolutional neural networks (CNNs). At the same time, there are concerns that these visualizations might not accurately represent CNNs' inner workings. Here, we measure how much extremely activating images help humans to predict CNN activations. Using a well-controlled psychophysical paradigm, we compare the informativeness of synthetic im…
Categories
AI & Data Science
Category · 10.8k presentations
About ICLR 2021
The International Conference on Learning Representations (ICLR) is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence called representation learning, but generally referred to as deep learning. ICLR is globally renowned for presenting and publishing cutting-edge research on all aspects of deep learning used in the fields of artificial intelligence, statistics and data science, as well as important application areas such as machine vision, computational biology, speech recognition, text understanding, gaming, and robotics.
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Workshop on Distributed and Private Machine Learning
Watch later
Learning N:M Fine-grained Structured Sparse Neural Networks From Scratch
Watch later
Aojun Zhou, …
How Robots Can Learn End-to-End From Data
Watch later
Self-supervised Representation Learning with Relative Predictive Coding
Watch later
Machine Learning for Preventing and Combating Pandemics
Watch later
