Understanding the role of individual units in a deep neural network

By PNAS - 2020-12-13

Description

Deep neural networks excel at finding hierarchical representations that solve complex tasks over large datasets. How can we humans understand these learned representations? In this work, we present ne ...

Summary

Abstract Deep neural networks excel at finding hierarchical representations that solve complex tasks over large datasets.
D) Concepts that match units in the final convolutional layer are summarized, showing a broad diversity of detectors for objects, object parts, materials, and colors.
Recognizable people in dataset images have been anonymized by pixelating faces in visualizations.
To locate human-interpretable visual concepts within large-scale datasets of images, we use the Unified Perceptual Parsing image segmentation network (33) trained on the ADE20K scene dataset (53, 60) and an assignment of numerical color values to color names (61).

Topics

Machine_Learning (0.4)
Coding (0.13)
NLP (0.09)

Similar Articles

What is translation equivariance, and why do we use convolutions to get it

By Medium - 2020-10-05

Multi-layer Perceptrons (MLPs) are standard neural networks with fully connected layers, where each input unit is connected with each output unit. We rarely use these kind of layers when working with…

Multi-class object detection and bounding box regression with Keras, TensorFlow, and Deep Learning

By PyImageSearch - 2020-10-12

In this tutorial, you will learn how to train a custom multi-class object detector using bounding box regression with the Keras and TensorFlow deep learning libraries. Last week’s tutorial covered how ...

Dual Contradistinctive Generative Autoencoder

By DeepAI - 2020-11-19

11/19/20 - We present a new generative autoencoder model with dual contradistinctive losses to improve generative autoencoder that performs s...

ResNets — Residual Blocks & Deep Residual Learning

By Medium - 2020-12-02

Deep Learning harnesses the power of Big Data by building deep neural architectures that try to approximate a function f(x) that can map an input, x to its corresponding label, y. The Universal…