Neural Ordinary Differential Equations - part 2 (results & discussion) | AISC
Toronto Deep Learning Series, 14-Jan-2019
https://tdls.a-i.science/events/2019-01-14
Paper: https://arxiv.org/abs/1806.07366
Discussion Panel: Jodie Zhu, Helen Ngo, Lindsay Brin
Host: SAS Institute Canada
NEURAL ORDINARY DIFFERENTIAL EQUATIONS
We introduce a new family of deep neural network models. Instead of specifying a discrete sequence of hidden layers, we parameterize the derivative of the hidden state using a neural network. The output of the network is computed using a black-box differential equation solver. These continuous-depth models have constant memory cost, adapt their evaluation strategy to each input, and can explicitly trade numerical precision for speed. We demonstrate these properties in continuous-depth residual networks and continuous-time latent variable models. We also construct continuous normalizing flows, a generative model that can train by maximum likelihood, without partitioning or ordering the data dimensions. For training, we show how to scalably backpropagate through any ODE solver, without access to its internal operations. This allows end-to-end training of ODEs within larger models.
Видео Neural Ordinary Differential Equations - part 2 (results & discussion) | AISC канала ML Explained - Aggregate Intellect - AI.SCIENCE
https://tdls.a-i.science/events/2019-01-14
Paper: https://arxiv.org/abs/1806.07366
Discussion Panel: Jodie Zhu, Helen Ngo, Lindsay Brin
Host: SAS Institute Canada
NEURAL ORDINARY DIFFERENTIAL EQUATIONS
We introduce a new family of deep neural network models. Instead of specifying a discrete sequence of hidden layers, we parameterize the derivative of the hidden state using a neural network. The output of the network is computed using a black-box differential equation solver. These continuous-depth models have constant memory cost, adapt their evaluation strategy to each input, and can explicitly trade numerical precision for speed. We demonstrate these properties in continuous-depth residual networks and continuous-time latent variable models. We also construct continuous normalizing flows, a generative model that can train by maximum likelihood, without partitioning or ordering the data dimensions. For training, we show how to scalably backpropagate through any ODE solver, without access to its internal operations. This allows end-to-end training of ODEs within larger models.
Видео Neural Ordinary Differential Equations - part 2 (results & discussion) | AISC канала ML Explained - Aggregate Intellect - AI.SCIENCE
Показать
Комментарии отсутствуют
Информация о видео
4 февраля 2019 г. 19:55:35
00:42:05
Другие видео канала
Diving Into Document Question and Answering Systems with LLMsBuilding with LLMs Using LangChainRunning LLMs in Your EnvironmentChatGPT-like application for construction of mathematical financial modelsModern Innovations in Fine-Tuning Large Language ModelsBuilding ResearchLLM: automated statistical research and interpretationExploring the agency limits of today's LLMsLearning-free Controllable Text Generation for DebiasingLeveraging Language Models for Training Data Generation and Tool LearningGenerative AI: Ethics, Accessibility, Legal Risk MitigationIntegrating LLMs into Your Product: Considerations and Best PracticesIncorporating Large Language Models into Enterprise AnalyticsCommercializing LLMs: Lessons and Ideas for Agile InnovationThe Emergence of KnowledgeOpsNeural Search for Augmented Decision Making - Zeta Alpha - DRT S2E17Distributed Data Engineering for Science - OpSci - Holonym - DRT S2E16Data Products - Accumulation of Imperfect Actions Towards a Focused Goal - DRT S2E15Unfolding the Maze of Funding Deep Tech; Metafold - DRT S2E14 - Ft. Moien Giashi, Alissa rossData Structure for Knowledge = Language Models + Structured Data - DRT S2E13EVE - Explainable Vector Embeddings - DRT S2E12LabDAO - Decentralized Marketplace for Research in Life Sciences - DRT S2E11