Pointer networks Tensorflow2

Overview

Pointer networks Tensorflow2

原文:https://arxiv.org/abs/1506.03134
仅供参考与学习,内含代码备注

环境

tensorflow==2.6.0
tqdm
matplotlib
numpy

《pointer networks》阅读笔记

应用场景:

文本摘要,凸包问题,Roundelay 三角剖分,旅行商问题

其中包括一些Latex,github无法渲染,所以建议clone下来用Typora查看。

abstract

本文提出一种新的网络结构:输出序列的元素是与输入序列中的位置相对应的离散标记。

an output sequence with elements that are discrete tokens corresponding to positions in an input sequence.

这种问题目前可以被一些现有的方法解决:sequence-to-sequence, neural turing machines。但是这些方法不是特别适用。

本文解决的问题是sorting variable sized sequences,以及各种组合优化问题。本模型使用attention机制来解决变化尺寸的输出。

intro

RNN模型的输出维度是固定的,sequence-to-sequence模型移除了这一个限制,通过用一个RNN把输入映射为一个embedding,又用一个RNN把embedding映射到输出序列。

但是这些sequence-to-sequence 方法都是固定大小的词汇表。

例如词汇表中只存在A,B,C。那么输入

1,2,3 ----> A,B,C

1,2,3,4 ----> A,B,C,A

本文提出的框架适用于输出的词汇表大小取决于输入问题的大小

image-20211105133740833

image-20211105134312635

左图:seq-2-seq

蓝色RNN,输出一个向量。

紫色RNN,利用概率的链式法则,输出一个固定维度。

本文的贡献如下:

  1. 提出一种新的结构,称为指针网路。简单且高效
  2. 良好的泛化性能
  3. 一个TSP近似求解器

Models

sequence-to-sequence 模型

训练数据为: $$ (P,C^P) $$ 其中,$\mathcal{P}=\left{P_{1}, \ldots, P_{n}\right}$,是n个向量。$\mathcal{C}^{\mathcal{P}}=\left{C_{1}, \ldots, C_{m(\mathcal{P})}\right}$ ,n个对应的结果,$m(\mathcal{P})\in [1,n]$ 。传统的sequence-to-sequence的$\mathcal{C}^{\mathcal{P}}$是固定大小的,但是要提前给定。本文的$\mathcal{C}^{\mathcal{P}}$为n,根据输入改变。

如果模型的参数记为$\theta$,神经网络模型表达为: $$ p(C^P|P,\theta) $$ 使用链式法则,写为: $$ p\left(\mathcal{C}^{\mathcal{P}} \mid \mathcal{P} ; \theta\right)=\prod_{i=1}^{m(\mathcal{P})} p_{\theta}\left(C_{i} \mid C_{1}, \ldots, C_{i-1}, \mathcal{P} ; \theta\right) $$ 训练阶段,最大似然概率: $$ \theta^{*}=\underset{\theta}{\arg \max } \sum_{\mathcal{P}, \mathcal{C}^{\mathcal{P}}} \log p\left(\mathcal{C}^{\mathcal{P}} \mid \mathcal{P} ; \theta\right) $$ input sequence的末端加一个$\Rightarrow$,代表进入生成阶段,$\Leftarrow$代表结束生成阶段。

推断: $$ \hat{\mathcal{C}}^{\mathcal{P}}=\underset{\mathcal{C}^{\mathcal{P}}}{\arg \max } p\left(\mathcal{C}^{\mathcal{P}} \mid \mathcal{P} ; \theta^{*}\right) $$

content based input attention

对于attention机制,请查看《Neural Machine Translation By Jointly Learning To Align And Translate》阅读笔记。

对于LSTM RNN $$ \begin{aligned} u_{j}^{i} &=v^{T} \tanh \left(W_{1} e_{j}+W_{2} d_{i}\right) & j \in(1, \ldots, n) \ a_{j}^{i} &=\operatorname{softmax}\left(u_{j}^{i}\right) & j \in(1, \ldots, n) \ d_{i}^{\prime} &=\sum_{j=1}^{n} a_{j}^{i} e_{j} & \end{aligned} $$ 对于这个传统的attention机制,可以看到$u^{i}$, 是一个长度为$n$的向量。

这样的话,在解码器的每一个时间步迭代都会得到一个 n 长度的向量,可以作为指针,用于指向之前的 n 长度的序列。

Ptr-Net

所以Ptr-Net计算公式写为: $$ \begin{aligned} u_{j}^{i} &=v^{T} \tanh \left(W_{1} e_{j}+W_{2} d_{i}\right) \quad j \in(1, \ldots, n) \ p\left(C_{i} \mid C_{1}, \ldots, C_{i-1}, \mathcal{P}\right) &=\operatorname{softmax}\left(u^{i}\right) \end{aligned} $$ image-20211111103159924

image-20211111110334755

数据以 [Batch, time_steps, feature] 的形式进入编码器LSTM(绿色部分),在时间步上迭代$n$次以后,得到:

  • n 个 e [batch, units], 可以合并写为 [batch, n, units]

  • 最后一个时间步输出的 c [batch, units]

进入到解码器LSTM(蓝色部分),输入为:

  • 上次得到解码得到的的pointer,如果是第一次则为initial pointer
  • 上次的状态d,c

pointer 如何得到?计算公式如下: $$ \begin{aligned} u_{j}^{i} &=v^{T} \tanh \left(W_{1} e_{j}+W_{2} d_{i}\right) \quad j \in(1, \ldots, n) \ p\left(C_{i} \mid C_{1}, \ldots, C_{i-1}, \mathcal{P}\right) &=\operatorname{softmax}\left(u^{i}\right) \end{aligned} $$

motivation and datasets structure

文章是为了解决三种问题,凸包,Delaunay Triangulation,旅行商问题。在此只对旅行商问题进行探讨。

travelling salesman problem

给定一个城市列表,我们希望找到一条最短的路线,每个城市只访问一次,然后返回起点。此外,假设两个城市之间的距离在正反方向上是相同的。这是一个NP难问题,测试模型的能力和局限性。

数据生成:

卡迪尔坐标系(二维),$[0,1] \times[0,1]$

使用 Held-Karp algorithm 得到准确解,n最多为20。

A1,A2,A3为三种其他算法。A1,A2时间复杂度为$O\left(n^{2}\right)$,A3时间复杂度为$O\left(n^{3}\right)$。A3,Christofides algorithm 算法保证在距离最佳长度1.5倍的范围内找到解,详细信息查看原文参考文献。生成1M个数据进行训练。

image-20211111111416012

分析表格:

  1. n=5的时候,性能都很好
  2. n=10,ptr-net的性能比A1好
  3. n=50的时候,无法超过数据集性能(因为ptr-net使用不准确的答案进行训练的)
  4. 只用n少的训练,推广到大n情况,性能不太好。

对于n=30的情况,Ptr-net算法复杂度为$O(n \log n)$,远低于A1,A2,A3。却有相似的性能,说明可发展空间还是很大的。

You might also like...
Complex-Valued Neural Networks (CVNN)Complex-Valued Neural Networks (CVNN)

Complex-Valued Neural Networks (CVNN) Done by @NEGU93 - J. Agustin Barrachina Using this library, the only difference with a Tensorflow code is that y

A framework that constructs deep neural networks, autoencoders, logistic regressors, and linear networks

A framework that constructs deep neural networks, autoencoders, logistic regressors, and linear networks without the use of any outside machine learning libraries - all from scratch.

Tensors and Dynamic neural networks in Python with strong GPU acceleration
Tensors and Dynamic neural networks in Python with strong GPU acceleration

PyTorch is a Python package that provides two high-level features: Tensor computation (like NumPy) with strong GPU acceleration Deep neural networks b

Lightweight library to build and train neural networks in Theano

Lasagne Lasagne is a lightweight library to build and train neural networks in Theano. Its main features are: Supports feed-forward networks such as C

A flexible framework of neural networks for deep learning
A flexible framework of neural networks for deep learning

Chainer: A deep learning framework Website | Docs | Install Guide | Tutorials (ja) | Examples (Official, External) | Concepts | ChainerX Forum (en, ja

Fast, flexible and fun neural networks.

Brainstorm Discontinuation Notice Brainstorm is no longer being maintained, so we recommend using one of the many other,available frameworks, such as

Image-to-Image Translation with Conditional Adversarial Networks (Pix2pix) implementation in keras

pix2pix-keras Pix2pix implementation in keras. Original paper: Image-to-Image Translation with Conditional Adversarial Networks (pix2pix) Paper Author

Code samples for my book "Neural Networks and Deep Learning"

Code samples for "Neural Networks and Deep Learning" This repository contains code samples for my book on "Neural Networks and Deep Learning". The cod

Python Library for learning (Structure and Parameter) and inference (Statistical and Causal) in Bayesian Networks.

pgmpy pgmpy is a python library for working with Probabilistic Graphical Models. Documentation and list of algorithms supported is at our official sit

Releases(v0)
Owner
HUANG HAO
Program = Algorithm + Data structure
HUANG HAO
Ranking Models in Unlabeled New Environments (iccv21)

Ranking Models in Unlabeled New Environments Prerequisites This code uses the following libraries Python 3.7 NumPy PyTorch 1.7.0 + torchivision 0.8.1

14 Dec 17, 2021
On Evaluation Metrics for Graph Generative Models

On Evaluation Metrics for Graph Generative Models Authors: Rylee Thompson, Boris Knyazev, Elahe Ghalebi, Jungtaek Kim, Graham Taylor This is the offic

13 Jan 07, 2023
Codes for realizing theories learned from Data Mining, Machine Learning, Deep Learning without using the present Python packages.

Codes-for-Algorithms Codes for realizing theories learned from Data Mining, Machine Learning, Deep Learning without using the present Python packages.

Tracy (Shengmin) Tao 1 Apr 12, 2022
Matthew Colbrook 1 Apr 08, 2022
Pytorch implementation for Semantic Segmentation/Scene Parsing on MIT ADE20K dataset

Semantic Segmentation on MIT ADE20K dataset in PyTorch This is a PyTorch implementation of semantic segmentation models on MIT ADE20K scene parsing da

MIT CSAIL Computer Vision 4.5k Jan 08, 2023
Keyhole Imaging: Non-Line-of-Sight Imaging and Tracking of Moving Objects Along a Single Optical Path

Keyhole Imaging Code & Dataset Code associated with the paper "Keyhole Imaging: Non-Line-of-Sight Imaging and Tracking of Moving Objects Along a Singl

Stanford Computational Imaging Lab 20 Feb 03, 2022
Brain tumor detection using CNN (InceptionResNetV2 Model)

Brain-Tumor-Detection Building a detection model using a convolutional neural network in Tensorflow & Keras. Used brain MRI images. InceptionResNetV2

1 Feb 13, 2022
Deep learning model for EEG artifact removal

DeepSeparator Introduction Electroencephalogram (EEG) recordings are often contaminated with artifacts. Various methods have been developed to elimina

23 Dec 21, 2022
The code for our paper "AutoSF: Searching Scoring Functions for Knowledge Graph Embedding"

AutoSF The code for our paper "AutoSF: Searching Scoring Functions for Knowledge Graph Embedding" and this paper has been accepted by ICDE2020. News:

AutoML Research 64 Dec 17, 2022
Augmented CLIP - Training simple models to predict CLIP image embeddings from text embeddings, and vice versa.

Train aug_clip against laion400m-embeddings found here: https://laion.ai/laion-400-open-dataset/ - note that this used the base ViT-B/32 CLIP model. S

Peter Baylies 55 Sep 13, 2022
Code release for "Self-Tuning for Data-Efficient Deep Learning" (ICML 2021)

Self-Tuning for Data-Efficient Deep Learning This repository contains the implementation code for paper: Self-Tuning for Data-Efficient Deep Learning

THUML @ Tsinghua University 101 Dec 11, 2022
Multivariate Time Series Transformer, public version

Multivariate Time Series Transformer Framework This code corresponds to the paper: George Zerveas et al. A Transformer-based Framework for Multivariat

363 Jan 03, 2023
Directed Greybox Fuzzing with AFL

AFLGo: Directed Greybox Fuzzing AFLGo is an extension of American Fuzzy Lop (AFL). Given a set of target locations (e.g., folder/file.c:582), AFLGo ge

380 Nov 24, 2022
Automatically erase objects in the video, such as logo, text, etc.

Video-Auto-Wipe Read English Introduction:Here   本人不定期的基于生成技术制作一些好玩有趣的算法模型,这次带来的作品是“视频擦除”方向的应用模型,它实现的功能是自动感知到视频中我们不想看见的部分(譬如广告、水印、字幕、图标等等)然后进行擦除。由于图标擦

seeprettyface.com 141 Dec 26, 2022
Character Controllers using Motion VAEs

Character Controllers using Motion VAEs This repo is the codebase for the SIGGRAPH 2020 paper with the title above. Please find the paper and demo at

Electronic Arts 165 Jan 03, 2023
[SIGMETRICS 2022] One Proxy Device Is Enough for Hardware-Aware Neural Architecture Search

One Proxy Device Is Enough for Hardware-Aware Neural Architecture Search paper | website One Proxy Device Is Enough for Hardware-Aware Neural Architec

10 Dec 16, 2022
Reproduces the results of the paper "Finite Basis Physics-Informed Neural Networks (FBPINNs): a scalable domain decomposition approach for solving differential equations".

Finite basis physics-informed neural networks (FBPINNs) This repository reproduces the results of the paper Finite Basis Physics-Informed Neural Netwo

Ben Moseley 65 Dec 28, 2022
Enhancing Aspect-Based Sentiment Analysis with Supervised Contrastive Learning.

Enhancing Aspect-Based Sentiment Analysis with Supervised Contrastive Learning. Enhancing Aspect-Based Sentiment Analysis with Supervised Contrastive

<a href=[email protected](SZ)"> 7 Dec 16, 2021
GenshinMapAutoMarkTools - Tools To add/delete/refresh resources mark in Genshin Impact Map

使用说明 适配 windows7以上 64位 原神1920x1080窗口(其他分辨率后续适配) 待更新渊下宫 English version is to be

Zero_Circle 209 Dec 28, 2022
code for "Feature Importance-aware Transferable Adversarial Attacks"

Feature Importance-aware Attack(FIA) This repository contains the code for the paper: Feature Importance-aware Transferable Adversarial Attacks (ICCV

Hengchang Guo 44 Nov 24, 2022