A5061687942- Generative Adversarial Networks and Image Synthesis
- Advanced Vision and Imaging
- Neural Networks and Applications
- Model Reduction and Neural Networks
- Computer Graphics and Visualization Techniques
- Advanced Neural Network Applications
- RNA Research and Splicing
- Neural dynamics and brain function
- Optical measurement and interference techniques
- Acupuncture Treatment Research Studies
- 3D Shape Modeling and Analysis
- Domain Adaptation and Few-Shot Learning
- Image Retrieval and Classification Techniques
- Liver Disease Diagnosis and Treatment
- Telomeres, Telomerase, and Senescence
- Advanced Neuroimaging Techniques and Applications
- Cell Image Analysis Techniques
- Vibration Control and Rheological Fluids
- Topological and Geometric Data Analysis
- Software Testing and Debugging Techniques
- Image Processing and 3D Reconstruction
- Topic Modeling
- Myofascial pain diagnosis and treatment
- Machine Learning and ELM
- Robotics and Sensor-Based Localization
Capital Medical University
2017-2025
Northeast Forestry University
2023-2024
Ministry of Agriculture and Rural Affairs
2022-2024
South China Agricultural University
2022-2024
Air Force Medical University
2024
Shandong Jianzhu University
2023-2024
Huazhong University of Science and Technology
2023-2024
Nanjing University
2021-2023
Collaborative Innovation Center of Advanced Microstructures
2021-2023
Laboratoire de physique des Solides
2022-2023
We present Imagen Video, a text-conditional video generation system based on cascade of diffusion models. Given text prompt, Video generates high definition videos using base model and sequence interleaved spatial temporal super-resolution describe how we scale up the as text-to-video including design decisions such choice fully-convolutional models at certain resolutions, v-parameterization In addition, confirm transfer findings from previous work diffusion-based image to setting. Finally,...
Classifier-free guided diffusion models have recently been shown to be highly effective at high-resolution image generation, and they widely used in large-scale frameworks including DALL.E 2, Stable Diffusion Imagen. However, a downside of classifier-free is that are computationally expensive inference time since require evaluating two models, class-conditional model an unconditional model, tens hundreds times. To deal with this limitation, we propose approach distilling into fast sample...
This paper proposes a 3D shape descriptor network, which is deep convolutional energy-based model, for modeling volumetric patterns. The maximum likelihood training of the model follows an "analysis by synthesis" scheme and can be interpreted as mode seeking shifting process. synthesize patterns sampling from probability distribution via MCMC such Langevin dynamics. used to train generator network teaching. conditional version net object recovery super-resolution. Experiments demonstrate...
This paper studies the cooperative training of two generative models for image modeling and synthesis. Both are parametrized by convolutional neural networks (ConvNets). The first model is a deep energy-based model, whose energy function defined bottom-up ConvNet, which maps observed to energy. We call it descriptor network. second generator network, non-linear version factor analysis. It top-down latent factors image. maximum likelihood learning algorithms both involve MCMC sampling such as...
This paper studies a training method to jointly estimate an energy-based model and flow-based model, in which the two models are iteratively updated based on shared adversarial value function. joint has following traits. (1) The update of is noise contrastive estimation, with flow serving as strong distribution. (2) approximately minimizes Jensen-Shannon divergence between data (3) Unlike generative networks (GAN) estimates implicit probability distribution defined by generator our explicit...
This paper proposes a cooperative learning algorithm to train both the undirected energy-based model and directed latent variable jointly. The interweaves maximum likelihood algorithms for two models, each iteration consists of following steps: (1) Modified contrastive divergence model: is based on divergence, but finite-step MCMC sampling initialized from synthesized examples generated by instead being observed examples. (2) teaching how in changes initial model, where variables that...
This paper proposes a multi-grid method for learning energy-based generative ConvNet models of images. For each grid, we learn an probabilistic model where the energy function is defined by bottom-up convolutional neural network (ConvNet or CNN). Learning such requires generating synthesized examples from model. Within iteration our algorithm, observed training image, generate images at multiple grids initializing finite-step MCMC sampling minimal 1 Ã- version image. The image subsequent...
Neural networks are vulnerable to adversarial examples, i.e. inputs that imperceptibly perturbed from natural data and yet incorrectly classified by the network. Adversarial training, a heuristic form of robust optimization alternates between minimization maximization steps, has proven be among most successful methods train against pre-defined family perturbations. This paper provides partial answer success showing it converges network where surrogate loss with respect attack algorithm is...
Byzantine agreement, the underlying core of blockchain, aims to make every node in a decentralized network reach consensus. Classical agreements unavoidably face two major problems. One is 1/3 fault-tolerance bound, which means that system tolerate f malicious players requires at least 3 + 1 players. The other security loopholes from its classical cryptography methods. Here, we propose agreement framework with unconditional break this bound nearly 1/2 fault tolerance due multiparty...
Chrysanthemum (
Schizophrenia is a kind of serious mental illness. Due to the lack an objective physiological data supporting and unified analysis method, doctors can only rely on subjective experience distinguish normal people patients, which easily lead misdiagnosis. In recent years, functional Near-Infrared Spectroscopy (fNIRS) has been widely used in clinical diagnosis, it get hemoglobin concentration through variation optical intensity. Firstly, prefrontal brain networks were constructed based oxy-Hb...
We present a deformable generator model to disentangle the appearance and geometric information in purely unsupervised manner. The models related information, including color, illumination, identity or category, of an image, while performs warping, such as rotation stretching, through generating displacement coordinates each pixel obtain final image. Two generators act upon independent latent factors extract disentangled from proposed scheme is general can be easily integrated into different...
This paper addresses the limitations of traditional portfolio theory centered on mean-variance model and expected utility theory, proposes establishment a that takes into account subjective psychological factors investors, taking fact investors are susceptible to influence various biases, affective cognitive biases in actual decision-making process, with respect consistency assumptions investor’s risk attitude. The based fuzzy is proposed, combined development application linear programming...
This paper studies the dynamic generator model for spatialtemporal processes such as textures and action sequences in video data. In this model, each time frame of sequence is generated by a which non-linear transformation latent state vector, where parametrized top-down neural network. The vectors follows auto-regressive vector next current well an independent noise that provides randomness transition. transition can be feedforward We show learned alternating back-propagation through...
First-order methods such as stochastic gradient descent (SGD) are currently the standard algorithm for training deep neural networks. Second-order methods, despite their better convergence rate, rarely used in practice due to prohibitive computational cost calculating second-order information. In this paper, we propose a novel Gram-Gauss-Newton (GGN) train networks regression problems with square loss. Our method draws inspiration from connection between network optimization and kernel of...
Flower color is an important characteristic of ornamental plants and determined by various chemical components, including anthocyanin. In the present study, combined metabolomics transcriptomics analysis was used to explore variations in chrysanthemums three cultivars, which JIN yellow, FEN pink, ZSH red. A total 29 different metabolites, nine anthocyanins, were identified common cultivars. Compared with light-colored all anthocyanin contents found be up-regulated dark-colored ones. The...
Hydrogen-bonded organic frameworks (HOFs) demonstrate significant potential for application in photocatalysis. However, the low efficiency of electron-hole separation and limited stability inhibit their practical utilization photocatalytic hydrogen evolution from water splitting. Herein, novel dual-pyrene-base supramolecular HOF/COF 2D/2D S-scheme heterojunction between HOF-H
While energy-based models (EBMs) exhibit a number of desirable properties, training and sampling on high-dimensional datasets remains challenging. Inspired by recent progress diffusion probabilistic models, we present recovery likelihood method to tractably learn sample from sequence EBMs trained increasingly noisy versions dataset. Each EBM is with likelihood, which maximizes the conditional probability data at certain noise level given their higher level. Optimizing more tractable than...
To achieve the highest perceptual quality, state-of-the-art diffusion models are optimized with objectives that typically look very different from maximum likelihood and Evidence Lower Bound (ELBO) objectives. In this work, we reveal model actually closely related to ELBO. Specifically, show all commonly used equate a weighted integral of ELBOs over noise levels, where weighting depends on specific objective used. Under condition monotonic weighting, connection is even closer: then equals...