时间：2022年6月16日（星期四）20:00-22:15
会议号：腾讯会议874 816 168
主办：www.bob.com
承办：信息学院
报告人1：谢雨波 20:00-20:45
题目：开放领域的共情对话生成
Open-Domain Empathetic Dialog Generation
摘要：共情对话生成是开放域聊天机器人的一个热门问题。之前的工作将情感信息整合到神经对话模型的输入中，并使用人为定义的规则来确定所需的情绪反馈。然而，这些工作并没有显式地学习人类对话中所蕴含的情感交流。因此，我们开发了一个多轮情感对话模型 (Multi-Turn Emotionally Engaging Dialog Model, MEED)，使用 LIWC 词典提取情感信息，然后使用情感 编码器进行编码，最后将得到的结果输入到解码过程的每一步。作为 MEED 的第二次迭代，我们开发了 MEED2 模型，它先在 OpenSubtitles 对话数据集上进行了预训练，然后在较小的对话 框数据集上进行了微调，在情绪的反馈机制上更加可控，可以生成更高细粒度的情绪和共情意图。
Empathetic response generation is a hot issue in the domain of conversational agents. Previous work incorporated affect information into the input of the neural dialog models and used hand-crafted rules to determine the desired response emotion. However, they did not explicitly learn the subtle emotion exchanges captured in human dialogs. To this end, we developed a multi-turn emotionally engaging dialog model (MEED), by adopting an emotion encoder that extracts emotion information using the LIWC program, which is then incorporated into each step of the decoding process. As a next iteration of MEED, we developed MEED2, which was first pre-trained on OpenSubtitles dialogs and then fine-tuned on smaller dialog datasets. It is more controllable and can generate more fine-grained emotions and intents.
报告人2：黄绍广 20:45-21:30
题目：基于稀疏表示的高光谱遥感影像分类
Sparse coding in the classification of hyperspectral remote sensing images
摘要：高光谱影像超越人类视觉感知的光谱诊断能力，在精细地物识别上具有独特的优势，在国防军事、精准农业、环境监测、城市规划等领域有诸多应用。然而高光谱影像数据维度高、观测样本量大、标签样本匮乏、噪声和光谱变异等问题极大限制了高光谱影像信息解译的精度和效率。稀疏表示是处理高维信号的经典模型，本报告将介绍稀疏表示在高光谱图像分类的应用，以及汇报人在该领域的最新研究进展，最后展望了未来的研究方向。
Due to rich spectral information, hyperspectral images (HSI) discriminate much better between fine ground objects, which leads to numerous applications of HSI in national defense and military, precision agriculture, environmental monitoring, urban planning, etc. However, the high dimensionality of HSI, big data, lack of labelled samples, noise and spectral variation make the accurate and efficient interpretation of HSI difficult. Sparse representation is a classical technique in the processing of high-dimensional signals. This presentation will introduce the applications of sparse representation in hyperspectral image classification, as well as the latest research progress of the presenter. Finally, future research direction in the field will be discussed.
报告人3：丁玉连 21:30-22:15
题目：基于机器学习方法的生物分子与疾病之间的关系预测
Predicting biomolecule-disease associations based on machine learning approaches
摘要：生物分子，例如小分子核糖核酸（miRNA）， 环状核糖核酸（circRNA），和长非编码核糖核酸（lncRNA）之间相互作用彼此调节以维持生物体丰富多彩的生命活动。一种生物分子的调节异常可能会影响整个生物分子调控系统的稳定性，并最终导致疾病。识别生物分子与疾病之间的相关性（BDAs）有助于揭示复杂的疾病发病机理，对疾病的诊断、治疗、预后和预防有很大的帮助。由于使用生物实验方法探索生物分子疾病关系耗时耗力，通过计算方法挑选高质量疾病相关的候选生物分子可以提升生物分子与疾病之间关系识别的效率。
通常，一个计算模型的性能与两个因素有关：模型中涉及的生物数据多样性和模型中使用的预测算法。组学数据的日益增加和机器学习方法的不断发展为基于机器学习的生物分子与疾病之间的关系预测提供了前所未有的机遇。本研究中我们结合不同组学数据的特征以及不同机器学习算法的优势构建一些新的基于机器学习的预测模型以探索深层复杂的生物分子与疾病之间的相关性。构建的预测模型包括基于深度置信网络的非负矩阵分解模型，基于图的变自动编码机模型，和多视图拉普拉斯正则化的DeepFM模型。这些模型有着他们不同的优势，并且与过去的一些模型比较，都一定程度上提升了预测性能。
Biomolecules, such as microRNAs (miRNAs), circRNAs, and long non-coding RNAs (lncRNAs) interact and regulate each other to sustain a rich variety of life activities. The dysregulation of one kind of biomolecule may affect the stability of the whole biomolecule regulation system, which eventually causes diseases. Identifying the biomolecule-disease associations (BDAs) can uncover the complex disease pathogenesis to greatly facilitate the diagnosis, treatment, prognosis, and prevention of diseases. Due to the low efficiency of exploring BDAs by biological experiments, prioritizing disease-related biomolecule candidates by computational approaches is imperative which can serve as an auxiliary tool to select high probability candidates.
The performance of a computational model usually relies on two factors: the biological data evidence involved in the model and the computational algorithm used in the model. The increasing of omics data and the development of machine learning methods provide an unprecedented opportunity for the improvement of machine learning-based BDAs prediction. We combine the characteristics of omics datasets and the advantages of different machine learning methods to construct some novel machine learning-based BDA prediction models to explore the deep underlying complex nonlinear associations. The developed models include deep belief network-based matrix factorization, variational graph autoencoder model, and multi-view Laplacian regularized DeepFM model. Those models have different advantages and improve the BDA prediction performance compared with previous prediction models.