一、
二、
三、
四、
五、
月份：2019年9月
【ACL-2019】A Novel Bi-directional Interrelated Model for Joint Intent Detection and Slot Filling
一、Background
-
- Joint model奠基者：
- 其他Model的缺点
- Attention-based recurrent neural network models for joint intent detection(Liu and Lane 2016,INTERSPEECH2016) just applied a joint loss function to link the two tasks implicitly
- Multi-domain joint semantic frame parsing using bi-directional rnn-lstm(Hakkani-T¨ur et al. 2016,INTERSPEECH2016) did not establish the explicit relationships between the slots and intent.
- Slot-gated modeling for joint slot filling and intent prediction(Goo et al. (2018),NAACL-HLT-2018). Slot information is not used in intent detection task(Slot-gate利用ID信息只对SF任务加成) .The bi-directional direct connections are still not established.
二、Contributions
-
- A novel ID subnet is proposed to apply the slot information to intent detection task;（最大的贡献）
- A novel iteration mechanism inside the SF-ID network in order to enhance the connections between the intent and slots;
- The experiments on two benchmark datasets show the effectiveness and superiority of the proposed model.
三、Approaches
-
- 1.SF-First Mode：We apply the intent context vector Cinte and slot context vector Cslot in the SF subnet and generate the slot reinforce vector Rslot. Then, the newly formed vector Rslot is fed to the ID subnet to bring the slot information.(将SF信息带入ID中).
- SF subnet
Cislot和Cinte就和Slot-Gated 里面的一样，分每个隐藏状态的attention和整个隐藏状态的attention。 - ID subnet
- Iteration Mechanism
- SF subnet
- 2.ID-First Mode
- 3.CRF layer：加在the SF subnet outputs上
- 1.SF-First Mode：We apply the intent context vector Cinte and slot context vector Cslot in the SF subnet and generate the slot reinforce vector Rslot. Then, the newly formed vector Rslot is fed to the ID subnet to bring the slot information.(将SF信息带入ID中).
四、Experiment
五、Refs
ConfigParser/configparser
- ConfigParser：在Python2中的包名
- configparser: 在Python3中的包名
按节（section）、键（option）、值来读写配置文件。
【AAAI-2019】Unsupervised Transfer Learning for Spoken Language Understanding in Intelligent Agents
一、Contributions
- 应用ELMo从原始ASR文本上做非监督的知识迁移，展示了SLU准确性的提升
- 提出了EMLo-Light（ELMoL），非常适合商业的设置，与ELMo对比减少了耗时和内存需求
- 结合非监督迁移学习（UT）和监督迁移学习（ST），展示了二者结合的效果。
- 在多种资源条件下，我们在benchmark SLU数据集和来自Alexa数的据集上评估了本算法的效果。
二、Methods for Unsupervised Transfer
- Embeddings from Language Model (ELMo)
首先使用unlabeled data训练EMLo（CNN-BIG-LSTM，CNN得到与上下文无关的词表达，BiLSTM得到与上下文有关的词表达），
其中使用了两层BiLSTM，训练过程中BiLSTM的参数保持不变，然后在ET和IC联合任务上来优化参数y和si。
- ELMo-Light (ELMoL) for SLU tasks
移除CNN，使用单层BiLSTM在unlabeled data训练词的表示，然后只需要训练三个ELMO的参数(y; s0; s1)，因为BiLSTM只有一层。最终在labeled data上fine-tune BiLSTM 的参数。
这边提出了一些训练ELMoL的技巧：（由于在labeled data上fine-tune BiLSTM的参数会导致catastrophic forgetting，忘掉之前在unlabeled data上的信息）
-
- gradual unfreezing (guf )：updating only the top layer for a few epochs keeping lower layers frozen, and then progressively updating bottom layer
- discriminative fine-tuning (discr)：combined with discr, e.g, using different learning rates across network layers
- slanted triangular learning rates (tlr )：learning rates that are initially slow which discourages drastic updates in early stages of learning, then it rapidly increase allowing more exploration of the parameter space and then slowly decrease enabling learned parameters to stabilize
-
Combining Supervised Transfer (ST) with UT
- ELMo+ST：1.在unlabeled data训练LM model得到ELMo embedding；2.在labeled data上训练IC/ET任务；3.fine-tune。特点是第2步和第3步中LM中的参数（主要是BiLSTM的参数）不变。
- ELMoL+ST：1.在unlabeled data训练LM model得到ELMoL embedding；2.在labeled data上训练IC/ET任务，训练上层网络时，保持底层网络参数不变，一旦上层网络稳定，就开始训练整个网络，此时底层网络使用比上层更小的学习率；3.使用上述的三个技巧（guf、discr、tlr）fine-tune所有的网络。
三、Experiment Result
四、Future Work
- apply the transfer techniques across different languages.
- experiment with alternative architectures such as transformer and adversarial networks
五、Refs
Transformer
一、概述
上图的左半边用 NX 框出来的，就代表一层 encoder，其中论文里面的 encoder 一共有6层这样的结构。上图的右半边用 NX 框出来的，则代表一层 decoder，同样也有6层。
二、流程
定义输入序列首先经过 word embedding，再和 positional encoding 相加后，输入到 encoder 中。输出序列经过的处理和输入序列一样，然后输入到 decoder。
最后，decoder 的输出经过一个线性层，再接 Softmax。