Article
Computer Science, Software Engineering
Yuankang Chen, Yifan Lu, Xiaohua Zhang, Nine Xie
Summary: In this paper, a lightweight cascaded network is proposed to denoise 1-spp Monte Carlo images through pixel and kernel prediction methods. Experimental results show that the approach achieves state-of-the-art denoising qualities for 1-spp images at an interactive frame speed.
Article
Multidisciplinary Sciences
Stephen Whitelam, Viktor Selin, Sang-Won Park, Isaac Tamblyn
Summary: Authors derive an analytic equivalence between neural network training under conditioned stochastic mutations and under gradient descent, showing that in the presence of small mutations, training a neural network by conditioned stochastic mutation or neuroevolution of its weights is equivalent to gradient descent on the loss function with Gaussian white noise. Neuroevolution is found to be equivalent to gradient descent on the loss function when averaged over independent realizations of the learning process, which is demonstrated through numerical simulations across finite mutations and various neural network architectures. This provides a connection between two families of neural-network training methods that are usually considered to be fundamentally different.
NATURE COMMUNICATIONS
(2021)
Article
Geochemistry & Geophysics
Xin Zhang, Andrew Curtis
Summary: This study introduces invertible neural networks (INNs) as an alternative to solving nonlinear and nonunique inverse problems in geophysics. By including data uncertainties as additional model parameters and training the network by maximizing the likelihood of the training data, INNs can provide comparable posterior probability density functions to Monte Carlo methods, including correlations between parameters.
JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH
(2021)
Article
Computer Science, Software Engineering
Cheng Zhang, Zihan Yu, Shuang Zhao
Summary: This paper introduces a physics-based differentiable rendering technique that utilizes differential path integrals for estimating arbitrary scene parameters. The method efficiently handles challenging geometric discontinuities and light transport phenomena, such as volumetric caustics.
ACM TRANSACTIONS ON GRAPHICS
(2021)
Article
Computer Science, Software Engineering
Shinyoung Yi, Donggun Kim, Kiseok Choi, Adrian Jarabo, Diego Gutierrez, Min H. Kim
Summary: Recent differentiable rendering techniques have become important tools for solving inverse problems in graphics and vision. However, existing models assume infinite speed of light, which may not be suitable for ultrafast imaging applications. This paper introduces a novel differentiable transient rendering framework that takes into account the finite speed of light, and successfully applies it in challenging scenarios such as optimizing indices of refraction and non-line-of-sight tracking.
ACM TRANSACTIONS ON GRAPHICS
(2021)
Article
Computer Science, Software Engineering
Hanggao Xin, Shaokun Zheng, Kun Xu, Ling-Qi Yan
Summary: This paper presents a novel method to generate single-bounce indirect illumination for dynamic scenes at interactive framerates. The method uses a lightweight neural network to predict screen-space indirect illumination, and achieves high quality and good temporal coherence through bilateral convolution layers and simplified input information.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2022)
Article
Computer Science, Information Systems
Yuliang Yuan, JinZhu Yang, Qi Sun, Yan Huang, Shuang Ma
Summary: Cinematic volume rendering is the next-generation volume rendering technology that uses photon mapping algorithm to overcome the limitations of slow convergence speed, high computational cost, and random noise. The algorithm supports multi-light illumination and enhances the perception of depth and shape in the region of interest.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Review
Computer Science, Software Engineering
Junqiu Zhu, Sizhe Zhao, Yanning Xu, Xiangxu Meng, Lu Wang, Ling-Qi Yan
Summary: In this article, we provide a comprehensive survey on recent glinty appearance rendering, starting with a definition based on microfacet theory. We summarize research works in terms of representation and practical rendering, and compare typical methods using a unified platform in terms of visual effects, rendering speed, and memory consumption. We also briefly discuss limitations and future research directions, aiming to provide insight for readers in choosing suitable methods for applications or conducting research.
COMPUTATIONAL VISUAL MEDIA
(2022)
Article
Radiology, Nuclear Medicine & Medical Imaging
Yazdan Salimi, Azadeh Akhavanallaf, Zahra Mansouri, Isaac Shiri, Habib Zaidi
Summary: We propose a deep learning-guided approach to generate voxel-based absorbed dose maps from whole-body CT acquisitions. Monte Carlo simulations (SP_MC) were used to calculate voxel-wise dose maps considering patient- and scanner-specific characteristics. A residual deep neural network (DNN) was trained to predict SP_MC using density map and SP_uniform dose maps. The performance of the DNN was evaluated for voxel-wise and organ-wise dose estimations using various error parameters.
EUROPEAN RADIOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Stephen Whitelam, Viktor Selin, Ian Benlolo, Corneel Casert, Isaac Tamblyn
Summary: The zero-temperature Metropolis Monte Carlo (MC) algorithm is examined as a tool for training neural networks. It can effectively train neural networks with comparable accuracy to gradient descent (GD), although not necessarily as quickly. The adaptive Monte Carlo algorithm (aMC) is introduced to overcome limitations when the network structure or neuron activations are strongly heterogeneous. The MC method allows training of deep neural networks and recurrent neural networks that cannot be trained by GD due to insignificant or excessive gradients. MC methods offer a complementary approach to gradient-based methods for training neural networks, providing access to different network architectures and principles.
MACHINE LEARNING-SCIENCE AND TECHNOLOGY
(2022)
Article
Multidisciplinary Sciences
Javier Robledo Moreno, Giuseppe Carleo, Antoine Georges, James Stokes
Summary: We introduce a new family of variational wave functions for simulating strongly correlated fermionic systems. By incorporating hidden additional degrees of freedom and optimizing the constraint and single-particle orbitals using a neural network parameterization, our construction overcomes limitations of hidden-particle representations and is proven to be universal. Applied to the ground-state properties of the Hubbard model, our approach achieves competitive levels of accuracy with state-of-the-art variational methods.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2022)
Article
Computer Science, Software Engineering
Megane Bati, Stephane Blanco, Christophe Coustet, Vincent Eymet, Vincent Forest, Richard Fournier, Jacques Gautrais, Nicolas Mellado, Mathias Paulin, Benjamin Piaud
Summary: This paper introduces a coupled path-space Monte Carlo algorithm for efficient multi-physics problems solving. By combining knowledge from physics and computer graphics, different simulation spaces are formulated and arranged into a single path space, enabling coupled heat transfer simulation using Monte Carlo. The proposed method allows for interactive computation of multiple simulations with different conditions in the same scene, validated through various thermal simulation scenarios. The theoretical framework presented in this work is expected to foster collaboration and multidisciplinary studies in resolving coupled PDEs at the interface of physics and computer graphics.
ACM TRANSACTIONS ON GRAPHICS
(2023)
Article
Computer Science, Interdisciplinary Applications
B. S. Kronheim, M. P. Kuchera, H. B. Prosper
Summary: TensorBNN is a new package that implements Bayesian inference for modern neural network models based on TensorFlow, sampling the posterior density of model parameters using Hamiltonian Monte Carlo. It leverages TensorFlow's architecture and GPU utilization in both training and prediction stages.
COMPUTER PHYSICS COMMUNICATIONS
(2022)
Article
Nuclear Science & Technology
Lesego E. Moloko, Pavel M. Bokov, Xu Wu, Kostadin N. Ivanov
Summary: This study uses Deep Neural Networks (DNNs) to predict assembly axial neutron flux profiles in the SAFARI-1 research reactor and quantifies the uncertainties in DNN predictions. Uncertainty Quantification is done using Monte Carlo Dropout (MCD) and Bayesian Neural Networks solved by Variational Inference (BNN VI). The results show that regular DNNs, DNNs with MCD, and BNN VI all have good prediction and generalization capabilities, and the uncertainty bands produced by MCD and BNN VI accurately envelope the measurement data points.
ANNALS OF NUCLEAR ENERGY
(2023)
Article
Computer Science, Interdisciplinary Applications
Jeffmin Lin, Gil Goldshlager, Lin Lin
Summary: The combination of neural networks and quantum Monte Carlo methods shows potential for highly accurate electronic structure calculations. This study introduces explicitly antisymmetrized universal neural network layers to address the challenge of measuring the expressiveness of the antisymmetric layer. The results demonstrate that applying these layers to the FermiNet model can accurately predict ground state energy for small atomic and molecular systems.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Computer Science, Software Engineering
Jonas Deyson Brito dos Santos, Pradeep Sen, Manuel M. Oliveira
Article
Computer Science, Software Engineering
Qiaodong Cui, Pradeep Sen, Theodore Kim
ACM TRANSACTIONS ON GRAPHICS
(2018)
Article
Computer Science, Software Engineering
Ben Mildenhall, Pratul P. Srinivasan, Rodrigo Ortiz-Cayon, Nima Khademi Kalantari, Ravi Ramamoorthi, Ren Ng, Abhishek Kar
ACM TRANSACTIONS ON GRAPHICS
(2019)
Article
Computer Science, Software Engineering
Nima Khademi Kalantari, Ravi Ramamoorthi
COMPUTER GRAPHICS FORUM
(2019)
Article
Computer Science, Software Engineering
Alexandr Kuznetsov, Milos Hasan, Zexiang Xu, Ling-Qi Yan, Bruce Walter, Nima Khademi Kalantari, Steve Marschner, Ravi Ramamoorthi
ACM TRANSACTIONS ON GRAPHICS
(2019)
Article
Computer Science, Information Systems
Chieh-Chi Kao, Yuxiang Wang, Jonathan Waltman, Pradeep Sen
IEEE TRANSACTIONS ON MULTIMEDIA
(2020)
Article
Computer Science, Software Engineering
Qiaodong Cui, Timothy Langlois, Pradeep Sen, Theodore Kim
COMPUTER GRAPHICS FORUM
(2020)
Article
Computer Science, Artificial Intelligence
Avinash Paliwal, Nima Khademi Kalantari
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2020)
Article
Computer Science, Software Engineering
Qinbo Li, Nima Khademi Kalantari
ACM TRANSACTIONS ON GRAPHICS
(2020)
Article
Computer Science, Software Engineering
Junqiu Zhu, Yaoyi Bai, Zilin Xu, Steve Bako, Edgar Velazquez-Armendariz, Lu Wang, Pradeep Sen, Milos Hasan, Ling-Qi Yan
Summary: Inspired by the success of deep networks, this research proposes using a machine learning framework to compress a complex luminaire's lightfield into an implicit neural representation. By training three networks to perform essential operations for evaluating, importance sampling, and blending complex luminaires, the approach achieves favorable results compared to state-of-the-art methods with reduced computation and storage costs.
ACM TRANSACTIONS ON GRAPHICS
(2021)
Article
Computer Science, Software Engineering
Xilong Zhou, Nima Khademi Kalantari
Summary: This paper proposes an optimization-based method to estimate the reflectance properties of a near planar surface from a single input image. By introducing a training mechanism and a learned reflectance loss, the overfitting problem of test-time optimization can be addressed. Experimental results demonstrate that the proposed method has good convergence and produces better results.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Software Engineering
N. Milef, S. Sueda, N. Khademi Kalantari
Summary: We propose a learning-based approach for reconstructing full-body pose from sparse upper body tracking data obtained from a virtual reality (VR) device. We use a conditional variational autoencoder with gated recurrent units to synthesize plausible and temporally coherent motions from 4-point tracking. To ensure the plausibility of poses, we introduce a novel sample selection and interpolation strategy with an anomaly detection algorithm. Our system is lightweight, real-time, and capable of producing coherent and realistic motions.
COMPUTER GRAPHICS FORUM
(2023)
Article
Computer Science, Software Engineering
Xilong Zhou, Nima Khademi Kalantari
Summary: This paper proposes a deep learning approach for estimating spatially-varying BRDFs from a single image, overcoming limitations of existing methods by using an adversarial framework and training on both synthetic and real examples. The method shows improved handling of various cases compared to state-of-the-art methods.
COMPUTER GRAPHICS FORUM
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Abhishek Badki, Alejandro Troccoli, Kihwan Kim, Jan Kautz, Pradeep Sen, Orazio Gallo
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
(2020)
Proceedings Paper
Computer Science, Artificial Intelligence
Ekta Prashnani, Hong Cai, Yasamin Mostofi, Pradeep Sen
2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
(2018)