目錄

?

(1) 傳統(tǒng)機(jī)器學(xué)習(xí)的主要假設(shè)之一

(2) 遷移學(xué)習(xí)的目的

應(yīng)用

(1) 腦電圖分類

(2) 事件抽取-transfer learning and shared semantic space

---

(1) 傳統(tǒng)機(jī)器學(xué)習(xí)的主要假設(shè)之一

是用于訓(xùn)練分類器的訓(xùn)練數(shù)據(jù)和用于評估分類器的測試數(shù)據(jù)屬于相同的特征空間，并且遵循相同的概率分布。但是，由于人的可變性，在許多應(yīng)用中經(jīng)常違反這一假設(shè)[55]。

reference:?

A. M. Azab, J. Toth, L. S. Mihaylova, and M. Arvaneh, “A review on transfer learning approaches in brain–computer interface,” inSignal Processing and Machine Learning for Brain-Machine Interfaces. Institution of Engineering and Technology, 2018, pp. 81–101.

例如，腦電圖EEG(Electroencephalography)數(shù)據(jù)分布中的變化通常發(fā)生在從不同主題或跨主題內(nèi)的會話和時間中獲取數(shù)據(jù)時。而且，由于EEG信號是變化的而不是靜態(tài)的(EEG has a high temporal resolution)，因此廣泛的BCI會話表現(xiàn)出獨(dú)特的一致性一致性分類問題[56]。

??while fMRI(functional magnetic resonance imaging)功能磁共振成像的一項基本挑戰(zhàn)是將觀察到的反應(yīng)映射到皮層下結(jié)構(gòu)。

reference:?mapping of observed responses to subcortical structures is a fundamental challenge in contrast to other neuroimaging approaches such as functional magnetic resonance imaging (fMRI), cf. Glover, 2011)

(2) 遷移學(xué)習(xí)的目的

??因此，遷移學(xué)習(xí)的目的是通過利用在學(xué)習(xí)給定任務(wù)時獲得的知識來解決不同但相關(guān)的任務(wù)來應(yīng)對違反該假設(shè)的數(shù)據(jù)。換句話說，遷移學(xué)習(xí)考慮用基于在學(xué)習(xí)另一個任務(wù)的同時獲得的信息來增強(qiáng)在一個任務(wù)（也擴(kuò)展到一個會話或主題）上學(xué)習(xí)的分類器的性能的一組方法。遷移學(xué)習(xí)的進(jìn)步可以放寬BCI的局限性，因為它不需要從開始就進(jìn)行校準(zhǔn)，遷移信息的噪音更少，并且無需依賴先前可用的數(shù)據(jù)來增加數(shù)據(jù)集的大小。

應(yīng)用

(1) 腦電圖分類

?當(dāng)前大多數(shù)機(jī)器學(xué)習(xí)研究都集中在靜態(tài)數(shù)據(jù)上，這不是對快速變化的大腦信號進(jìn)行準(zhǔn)確分類的最佳方法[34]

reference:

X. Zhang, L. Yao, X. Wang, J. Monaghan, and D. Mcalpine, “A survey on deep learning based brain computer interface: Recent advances and new frontiers,”?arXiv preprint arXiv:1905.04149, 2019.

我們將介紹具有CNN架構(gòu)的自發(fā)EEG應(yīng)用程序，近期研究中GAN的利用，RNN的過程和應(yīng)用程序，尤其是長短期記憶（LSTM）。我們還說明了從深度學(xué)習(xí)算法和轉(zhuǎn)移學(xué)習(xí)方法擴(kuò)展而來的深度轉(zhuǎn)移學(xué)習(xí)，然后以對抗攻擊為例，對用于系統(tǒng)測試的深度學(xué)習(xí)模型進(jìn)行了舉例說明.

(2) 事件抽取-transfer learning and shared semantic space

我們提出了一種可轉(zhuǎn)移的神經(jīng)體系結(jié)構(gòu)，該體系結(jié)構(gòu)利用現(xiàn)有的人為構(gòu)造的事件模式和少量手動注釋的可見類型集，并將現(xiàn)有類型的知識轉(zhuǎn)移到未見類型的提取中，以提高事件提取的可擴(kuò)展性，并節(jié)省人力。 在沒有任何注釋的情況下，我們的方法可以達(dá)到從大量標(biāo)記數(shù)據(jù)中訓(xùn)練而來的最新監(jiān)督模型可比的性能。

將來，我們將通過合并事件定義和參數(shù)描述來擴(kuò)展我們的框架，以提高事件提取性能。

We propose a transferable neural architecture, which leverages existing human constructed event schemas and manual annotations for a small set of seen types, and transfers the knowledge from the existing types to the extraction of unseen types, to improve the scalability of event extraction as well as save human effort. Without any annotation, our approach can achieve comparable performance with state-of-the-art supervised models trained from a large amount of labeled data. In the future, we will extend our framework by incorporating event definitions and argument descriptions to improve the event extraction performance。

This transferable neural architecture jointly learns and maps the structural representations of both event metions and types into a shared semantic space by minimizing the distance between each event mention and its corresponding type. For event mentions with unseen types, their structures will be projected into the same semantic space using the same framework and assigned types with top-ranked similarity values.

• candidate trigger
• construct event mention structute <--based on--candidate arguments, AMR parsing ---->每個事件mention元組數(shù)據(jù)：2AMR concepts, 1 AMP relation

? ? ? ? event type structure, each type y <---- incorporating its predefined roles and take the type as the root ---->事件類型元組數(shù)據(jù)：1 type name, 1 argument role name

• we apply a weight-sharing CNN to each input structure to jointly learn event mention and type structual representations, which will be later used to learn the ranking function for zero-shot event extraction.

? ? ? ? 1) input layer, dimensionality d x?2h* and d x 2p* respectively represent mention structure and type structure.

? ? ? ? 2) convolution layer, input matrix of St is a feature map of dimensionality d x 2h*

? ? ? ? 3) max-pooling, 所有的元組表示 all tuple representations Ci --max-pooling-->input sequence

? ? ? ? 4) learning

? ? ? ? 5) the linge loss, design a new loss function L1(t, y)

? ? ? ? 6) By minimizing L1, we can learn the optimized model ----> can compose structure representations

? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ----> map both event mention and types into a shared semantic space, where the positive type ranks the highest for each mention.

Reference:

Zero-Shot Transfer Learning for Event Extraction, Lifu Huang and Huai-zhong Ji and Kyunghyun Cho and Clare R. Voss, ACL, year=?2018

?

?

?

?

?

?

?

總結(jié)

以上是生活随笔為你收集整理的机器学习高阶认识(一): 机器学习假设与迁移学习的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。