Deep Interactive Evolution

Bontrager, Philip; Lin, Wending; Togelius, Julian; Risi, Sebastian

doi:10.1007/978-3-319-77583-8_18

Cited by 53 publications

(67 citation statements)

References 21 publications

(23 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For instance, the possible compounds eligible for drug design is between 10 23 and 10 60 . Recent advances in DNN and specifically in GANs have enabled innovations in creating a new image or composing a symphony . This discovery paradigm can be applied to various materials and provided thoughtful guidance to the synthesis of new materials .…”

Section: Molecular Discoveries Using Gansmentioning

confidence: 99%

Machine learning and artificial neural network accelerated computational discoveries in materials science

Yang

Hou

Jiang

et al. 2019

WIREs Comput Mol Sci

View full text Add to dashboard Cite

Artificial intelligence (AI) has been referred to as the "fourth paradigm of science," and as part of a coherent toolbox of data-driven approaches, machine learning (ML) dramatically accelerates the computational discoveries. As the machinery for ML algorithms matures, significant advances have been made not only by the mainstream AI researchers, but also those work in computational materials science. The number of ML and artificial neural network (ANN) applications in the computational materials science is growing at an astounding rate. This perspective briefly reviews the state-of-the-art progress in some supervised and unsupervised methods with their respective applications. The characteristics of primary ML and ANN algorithms are first described. Then, the most critical applications of AI in computational materials science such as empirical interatomic potential development, ML-based potential, property predictions, and molecular discoveries using generative adversarial networks (GAN) are comprehensively reviewed. The central ideas underlying these ML applications are discussed, and future directions for integrating ML with computational materials science are given. Finally, a discussion on the applicability and limitations of current ML techniques and the remaining challenges are summarized.

show abstract

Section: Molecular Discoveries Using Gansmentioning

confidence: 99%

Machine learning and artificial neural network accelerated computational discoveries in materials science

Yang

Hou

Jiang

et al. 2019

WIREs Comput Mol Sci

View full text Add to dashboard Cite

show abstract

“…In the absence of such an ideal, we must develop alternative generative models to test alternative hypotheses of the brain's encoding function for categorisation. Modern systems like generative adversarial networks (Karras et al, 2020) and derivatives of the classical VAE like vectorquantised VAEs (Oord et al, 2018;Razavi et al, 2019) and Nouveau VAEs (Vahdat & Kautz, 2020), which can be trained on large, naturalistic face databases, can synthesise tantalisingly realistic faces, complete with hair, opening up an interesting avenue for future research and applications (Suchow et al, 2018;Bontrager et al, 2018;Todorov et al, 2020). However, understanding and disentangling their latent spaces remains challenging (Mathieu et al, 2019), and it also remains challenging to engineer generative models that afford the multiple categorisations of realistic faces, bodies, objects and scenes.…”

Section: Hypothesis-driven Research Using Generative Modelsmentioning

confidence: 99%

Grounding deep neural network predictions of human categorization behavior in understandable functional features: The case of face identity

Daube¹,

Xu²,

Zhan³

et al. 2020

Preprint

View full text Add to dashboard Cite

Deep neural networks (DNNs) can resolve real-world tasks with apparent human-like performance and could therefore offer accessible information-processing models of the brain. However, interpreting these models remains a challenge. Here, we address this challenge by leveraging the experimental control of a generative model of the stimuli, in the form of faces that vary in 3D shape and texture. We asked human participants to rate how similar randomly generated faces were to four familiar colleagues. We then predicted these human responses from the shape and texture parameters of the generative model and from the activations of four DNNs trained on different optimisation objectives. With information- theoretic redundancy and reverse correlation, we then grounded behavioural prediction performance into interpretable functional shape and texture features. Our study addresses a critical prerequisite to evaluating the similarity between the brain and its DNN models: both must process the same functional features in the task.

show abstract

“…Another approach was done by Bontrager et al that combines generative adversarial networks (GANs) with interactive evolutionary computation [13]. They show a GAN trained on a specific target domain can act as an efficient genotype-to-phenotype mapping.…”

Section: Deep Learning and Evolutionary Artmentioning

confidence: 99%

Deep Learning Concepts for Evolutionary Art

Tanjil

Ross

2019

Computational Intelligence in Music, Sound, Art and Design

View full text Add to dashboard Cite

A deep convolutional neural network (CNN) trained on millions of images forms a very high-level abstract overview of any given target image. Our primary goal is to use this high-level content information of a given target image to guide the automatic evolution of images. We use genetic programming (GP) to evolve procedural textures.We incorporate a pre-trained deep CNN model into the fitness. We are not performing any training, but rather, we pass a target image through the pre-trained deep CNN and use its the high-level representation as the fitness guide for evolved images. We develop a preprocessing strategy called Mean Minimum Matrix Strategy (MMMS) which reduces the dimensions and identifies the most relevant high-level activation maps. The technique using reduced activation matrices for a fitness shows promising results. GP is able to guide the evolution of textures such that they have shared characteristics with the target image. We also experiment with the fully connected "classifier" layers of the deep CNN. The evolved images are able to achieve high confidence scores from the deep CNN module for some tested target images. Finally, we implement our own shallow convolutional neural network with a fixed set of filters.Experiments show that the basic CNN had limited effectiveness, likely due to the lack of training. In conclusion, the research shows the potential for using deep learning concepts in evolutionary art. As deep CNN models become better understood, they will be able to be used more effectively for evolutionary art.

show abstract

Deep Interactive Evolution

Cited by 53 publications

References 21 publications

Machine learning and artificial neural network accelerated computational discoveries in materials science

Machine learning and artificial neural network accelerated computational discoveries in materials science

Grounding deep neural network predictions of human categorization behavior in understandable functional features: The case of face identity

Deep Learning Concepts for Evolutionary Art

Contact Info

Product

Resources

About