我们遵循《 A contrastive framework for neural text generation》和《Neural text generation with unlikelihood training》中的实验设置。在exploration阶段,为了有一组具有不同重复级别的代表性模型输出,我们将贪婪解码和核采样按照50%-50%的比例混合,因为当使用贪婪解码时重复会更频繁。我们使用多样性度量作为奖励,以...
Controlable Text Generation,可控制的文本生成,就是能够在传统的文本生成的基础上,增加对生成文本一些属性、风格、关键信息等等的控制,从而使得生成的文本符合我们的某种预期。如果给一个明确的定义的话,文中引用了另一篇更早的综述的定义: Controllable text generation (CTG)refers to the task of generating text ac...
Controlable Text Generation,可控制的文本生成,就是能够在传统的文本生成的基础上,增加对生成文本一些属性、风格、关键信息等等的控制,从而使得生成的文本符合我们的某种预期。如果给一个明确的定义的话,文中引用了另一篇更早的综述的定义: Controllable text generation (CTG)refers to the task of generating text ac...
A controllable grounded response generation framework includes a machine learning model, a grounding interface, and a control interface. The machine learning model is trained to output computer-generated text based on input text. The grounding interface is useable by the machine learning model to ...
Causal Framework for Controllable Text Generation# Structual Causal Model# 这里假设treatmentaa是一个二元变量,可以是a∈{0,1}a∈{0,1},也可以是一个prompt。 由于在现实中高纬混淆变量zz可能难于估计,所以我们引入一个代理变量cc来间接估计zz。
Controllable Text-to-Image Generation with GPT-4 http://t.cn/A6pUphTs ChatPaper综述:本文阐述了当前的文本到图像生成模型在按照文本指示生成图像方面存在困难,特别是涉及到空间推理的指令。而使用大型语言...
In recent years, methods using large-scale pre-trained language models (PLMs), in particular the widely used transformer-based PLMs, have become a new paradigm of NLG, allowing generation of more diverse and fluent text. However, due to the lower level of interpretability of deep neural ...
Diffusion-LM improves controllable text generation 扩散语言模型改善可控文本生成(翻译) 论文作者:Xiang Lisa Li …… Stanford University 论文链接:https://arxiv.org/abs/2205.14217
Controllable Image Generation with Semi-supervised Deep Learning and Deformable-Mean-Template Based Geometry-Appearance Disentanglement * Typical deep-neural-network (DNN) based generative image models often (i) show limited ability to learn a disentangled latent representation, (ii) show lim... KASP ...
因为是套的方法,所以可以参考原始论文(传送门:The Curious Case of Neural Text Degeneration) 四、训练&推理过程解耦(Control Code Sampling) 上述都是训练过程,到了推理阶段,作者直接定死几个控制流信息就行,分别为 客观性控制流信息:<no-first-person>内容上保证模型客观; 词法精准度控制流信息:<high-prec>...