【云端学术 同心战“疫”】信息科学与工程学院线上学术讲座专场

0

1

现实世界包括文本、知识和多媒体等多种信息源，近年来随着自然语言处理和计算机视觉的快速发展，多模态问答因其能提供“文本+视觉”之间关系的洞察而引起越来越多的关注。目前，问答系统仍然面临着很多挑战，其一是如何理解用户多种多样的问题表达，其二是如何从海量问答知识中匹配出精准的答案。多模态信息增加了问答的难度，系统必须同时理解语言以及学习相应视觉的表示。本报告分别介绍文本+知识的多策略问答和两种形式的多模态问答如视觉问答（Visual Question Answering）和视频问答（Video Question Answering）等的关键技术和常用数据集。

0

2

Non-Markovian models of stochastic biochemical kinetics often incorporate explicit time delays to effectively model large numbers of intermediate biochemical processes.Analysis and simulation of these models, as well as the inference of their parameters from data, are fraught with difficulties because the dynamics depends on the system’s history. Here we use an artificial neural network to approximate the time-dependent distributions of non-Markovian models by the solutions of much simpler time-inhomogeneous Markovian models; the approximation does not increase the dimensionality of the model and simultaneously leads to inference of the kinetic parameters. The training of the neural network uses a relatively small set of noisy measurements generated by experimental data or stochastic simulations of the non-Markovian model. We show using a variety of models, where the delays stem from transcriptional processes and feedback control, that the Markovian models learnt by the neural network accurately reflect the stochastic dynamics across parameter space.