Научные семинары
Ближайший семинар:
Когда: 16.10.2023, 14:40 МСК
Лектор: Цгоев Чермен, аспирант ММФ НГУ
Название: Эффективное решение дифференциальных уравнений с помощью PINN-ADT
Аннотация: Данная работа представляет разработку универсального и гибкого программного комплекса, основанного на методе Physics-Informed Neural Networks (PINN), для решения разнообразных задач, связанных с дифференциальными уравнениями. Программный комплекс обеспечивает исследователей и инженеров средствами, позволяющими эффективно моделировать и анализировать различные типы систем дифференциальных уравнений, независимо от их количества и порядка. Дополнительно в работе представлены интегрированные инструменты, такие как адаптивные сетки и адаптивные регуляризационные члены, а также разнообразные нейросетевые архитектуры, улучшающие производительность и способность адаптации комплекса к различным задачам. Исследуется эффективность внедренных инструментов в различных конфигурациях.
Где: Покровский бульвар, 11, R604; zoom
Прошедшие семинары:
| Дата | Докладчик | Организация | Название | Аннотация |
| 20.03.2023 | Mr. Nand Kumar Yadav | - | Image to Image Translations Using Generative Adversarial Networks | Image-to-image translation is a widespread computer vision task capable of dealing with problems related to image generation, image colorization, semantic map to scene generation, etc. Visible image synthesis from complex input image modalities, such as sketch face to visual face transformation, thermal face to visual face transformation, and nearinfrared (NIR) image to visual optical image generation, is a crucial computer vision task using deep learning. Usually, complex input image modalities lacks with the rich visual information such as texture, color, fine details, etc., than the actual ground truth images in target domain. In such problems, there is a large domain gap between the input images and the corresponding target images. The traditional methods of image-to-image translation is unable to learn the accurate mapping between input and target images. However, in recent years, deep learning based Generative Adversarial Networks (GANs) have shown very promising results for image generation as well as image-to-image translation. The GANbased models are generally trained in an adversarial manner which is advantageous for the high quality image synthesis. GAN consists of generator and discriminator networks for training purposes. While for inference, only the trained generator network is required for synthesizing the images. For the better generalization ability in fewer training epoch, attention based methods have been proposed. Such methods focus over the important regions in the learning phase leading to better output. The existing image-to-image translation methods, such as Pix2pix, CycleGAN, DualGAN, CSGAN, and PCSGAN, miss the ingredients of attention mechanism. The existing GAN models require intense training. Moreover, these models suffer heavily with artifacts while synthesizing the complex scenes. The Attention based AGGAN and AttentionGAN result to in better unpaired image-to-image translation, but fails to effectively handle the complex image to real image translation. We propose different GAN methods by exploiting the attention mechanism for handling complex image modalities in the context of image-to-image translation. CSA-GAN is proposed with attention mechanism without using any extra network to handle the sketch face to real face translation as well as thermal face to visible face translation. By using the attention mechanism, it converges faster than the non-attention based methods and reports better results. We also propose a novel and efficient MobileAR-GAN model using attention-gates with MobileNet for complex scene translation, such as near-infrared to visible scene translation. The MobileAR-GAN is suitable for edge devices such as Jetson Board. We also propose Attention Guided Thermal to Visible Face Translation Network using GAN (TVAGAN). The proposed TVA-GAN is capable to generate the realistic samples with diverse face datasets, such as people from different races and poses. With the emergence of self-attention mechanisms image generation tasks reported promising results. Self-attention mechanism based parallel self-attention block is used with the inception block for improved performance over the thermal face to visible face transformation in the proposed Inception based Self-Attention GAN model (ISA-GAN). Further, a full-fledged Self-Attention driven Transformer network based Transcoder-GAN model is proposed for thermal face images into realistic face images. The generator network of the proposed Transcoder-GAN model utilizes the stack of multi-head self-attention blocks in an encoder-decoder fashion. |
| 27.02.2023 | Бузаев Федор | Младший научный сотрудник, Huawei RRI Theory Lab | Обзор на Physics Informed Neural Networks (PINNs), решающие уравнения частных производных, и для чего они нужны. | Нейронные сети, которые учитывают при обучении законы физики, помогают решать различные задачи, например такие как: прогноз погоды, описание электростатического поля и др.В этом обзоре будут разобраны нейронные сети, которые при обучении соблюдают любой заданный закон физики. Также в этом обзоре будут разобраны сферы применения таких нейронных сетей, как нейронные сети научились решать уравнения частных производных (уравнение Пуассона, уравнение Гельмгольца, уравнение Аллена-Кана), основные концепции обучения моделей. |
|
20.02.2023 |
Nikita Kazeev |
PhD, NUS Institute for Functional Intelligent Materials |
Sparse representation for machine learning the properties of defects is 2D materials |
The family of two-dimensional materials now includes dozens of crystals, which exhibit a wide range of electronic and optical properties, utilisable for the next generation devices. Furthermore, 2D crystals offer an opportunity of controlling their properties through a variety of knobs. Our particular interest is the possibility of obtaining desirable properties via controlled defect introduction. However, the search space for such structures is enormous, and proper ab-initio computations become prohibitively expensive. It makes essential to develop a way of precise and efficient prediction of the properties of a crystal with a specific defect configuration. We propose a machine learning approach for rapidly estimating 2D material properties given lattice structure and defect configuration. The method suggests a way to represent a 2D material configuration that allows a neural network to train quickly and accurately. We compare it with the state-of-the-art approaches and demonstrate at least $3.7$ times energy prediction error drop. Also, our approach is an order of magnitude more resource-efficient than its contenders both for the training and inference part. |
Нашли опечатку?
Выделите её, нажмите Ctrl+Enter и отправьте нам уведомление. Спасибо за участие!
Сервис предназначен только для отправки сообщений об орфографических и пунктуационных ошибках.