IBM 即国际商业机器公司(International Business Machines Corporation)。总部在纽约州阿蒙克市,1911年创立于美国,是全球最大的信息技术和业务解决方案公司。 目前拥有全球雇员 30多万人,业务遍及160多个国家和地区。

VIP内容

题目:

Transfer Learning in Visual and Relational Reasoning

简介:

迁移学习已成为计算机视觉和自然语言处理中的事实上的标准,尤其是在缺少标签数据的地方。通过使用预先训练的模型和微调,可以显着提高准确性。在视觉推理任务(例如图像问答)中,传递学习更加复杂。除了迁移识别视觉特征的功能外,我们还希望迁移系统的推理能力。而且,对于视频数据,时间推理增加了另一个维度。在这项工作中,我们将迁移学习的这些独特方面形式化,并提出了一种视觉推理的理论框架,以完善的CLEVR和COGdatasets为例。此外,我们引入了一种新的,端到端的微分递归模型(SAMNet),该模型在两个数据集上的传输学习中均显示了最新的准确性和更好的性能。改进的SAMNet性能源于其将抽象的多步推理与序列的长度解耦的能力及其选择性的关注能力,使其仅能存储与问题相关的信息外部存储器中的对象。

目录:

成为VIP会员查看完整内容
0
36

最新论文

Machine translation systems are vulnerable to domain mismatch, especially in a low-resource scenario. Out-of-domain translations are often of poor quality and prone to hallucinations, due to exposure bias and the decoder acting as a language model. We adopt two approaches to alleviate this problem: lexical shortlisting restricted by IBM statistical alignments, and hypothesis re-ranking based on similarity. The methods are computationally cheap, widely known, but not extensively experimented on domain adaptation. We demonstrate success on low-resource out-of-domain test sets, however, the methods are ineffective when there is sufficient data or too great domain mismatch. This is due to both the IBM model losing its advantage over the implicitly learned neural alignment, and issues with subword segmentation of out-of-domain words.

0
0
下载
预览
Top