NLG enables machines to convert structured data into human text or speech.
It's objective is machine creation of natural language so that they can do something useful. Challenge is the understanding and representing the meaning of the word. NLG system input can be defined as a four-tuple such as knowledge source, communicative goal, user model, discourse history.
Let see the NLG pipeline as the simple view. First, the purpose of communication motivation. Second, What do we want to say? Content or message. Third, How are we going to say it? Finally, Produce communication formulating the output.
The goal, knowledge of domain and grammar used to generate the document planning.
Why is NLG difficult? Mapping, choice, planning problems.
NLG application contains Machine translation, authoring, text summarization, Question answering in the knowledge base.
To deploy the nlg, you have to think what the objective is? And what is the data to process? What is the output? Will the benefit offer a return on investment?
The computer must behave intelligently that act like humans to achieve human-level performance in all cognitive tasks. It includes NLP, knowledge representation, automated reasoning, machine learning.
Alan Turing and the Imitation Game
Turing Test - Still valid?
Measures of intelligence
The problem with the Turing Test
NLG article: a detailed survey of the field
Marco planning - what do you want to say?
: Content determination Discourse planning
Micro planning - how will you say it?
: Lexicalization, Aggregation, referring expressions
Realization - what will you actually say?
NLG Application
- Summarization
- Content Creation
- Conversational Interfaces(Chabots)
NLG has two approaches. For example GOFAI techniques such as rules and templates, and the second using a type of Neural Network called a Recurrent Neural Network.
Yufeng interviews Google Research engineer Justin Zhao to talk NLG and their use of recurrent neural networks
Templates vs. NLG (where NLG means machine learning) by Ehud Reiter
this article by Andrej Karpathy: simple RNN
excellent article on Github by Christopher Olah of Google Brain of the issues of RNNs
2018년 11월 27일 화요일
NLP week2
NLP module2 is Natural Language Processing.
We have many text data to improve through the NLP technologies. And NLP technologies have many algorithms and representations. There are Good Old Passion AI with rule and template. And Deep Learning algorithms.
NLG Objective is machine understanding of natural language so that they can do something useful for people. Semantic interpretation and Syntactic Analysis is so challenging to implement.
A sequence of steps that converts unstructured text into a data structure an algorithm can work with.
First, text to Numeric data (Bag of Words)
There are two problems such as order and importance.
We can solve order problem by bigram. and Importance problem can be explained by TF/IDF
Second Linguistic inquiry and word count.
It uses the Based on the classification of the word.
And then A feature space based on the classes.
NLP problems commonly were divided by classification and regression.
Regression problems is evaluated by Root mean squared Error.
One of NLP classification algorithms is SVM, support vector machine that find the best line that divides the classes. Then It maximizes the distance between the nearest data point and the track. Svm contain a kernel function to classify data that is not linearly separable.
And It's robust to noise.
Evaluation has three factors such as accuracy (prediction is correct), recall (positive occurrences are predicted), precision (positive prediction is accurate).
Precision is essential to be right (Expensive interventions, Safe to miss out, eg. placing a Bet)
The recall is necessary to be complete (Inexpensive interventions, Dangerous to miss out, eg, Cancer screening)
You carefully think about evaluation methods. Is it worth it for customers using the evaluation?
Overfitting is when a model learns a dataset to well
To overcome this problem, remove unnecessary features. Regularization.
TF-IDF: FreeCodeCamp medium website
LIWC: 2010 paper by
Tausczik and Pennebaker
example of using Naive Bayes for text classification on the MonleyLearn blog
We have many text data to improve through the NLP technologies. And NLP technologies have many algorithms and representations. There are Good Old Passion AI with rule and template. And Deep Learning algorithms.
NLG Objective is machine understanding of natural language so that they can do something useful for people. Semantic interpretation and Syntactic Analysis is so challenging to implement.
A sequence of steps that converts unstructured text into a data structure an algorithm can work with.
First, text to Numeric data (Bag of Words)
There are two problems such as order and importance.
We can solve order problem by bigram. and Importance problem can be explained by TF/IDF
Second Linguistic inquiry and word count.
It uses the Based on the classification of the word.
And then A feature space based on the classes.
NLP problems commonly were divided by classification and regression.
Regression problems is evaluated by Root mean squared Error.
One of NLP classification algorithms is SVM, support vector machine that find the best line that divides the classes. Then It maximizes the distance between the nearest data point and the track. Svm contain a kernel function to classify data that is not linearly separable.
And It's robust to noise.
Evaluation has three factors such as accuracy (prediction is correct), recall (positive occurrences are predicted), precision (positive prediction is accurate).
Precision is essential to be right (Expensive interventions, Safe to miss out, eg. placing a Bet)
The recall is necessary to be complete (Inexpensive interventions, Dangerous to miss out, eg, Cancer screening)
You carefully think about evaluation methods. Is it worth it for customers using the evaluation?
Overfitting is when a model learns a dataset to well
To overcome this problem, remove unnecessary features. Regularization.
TF-IDF: FreeCodeCamp medium website
LIWC: 2010 paper by
Tausczik and Pennebaker
2018년 11월 22일 목요일
AI intelligence for the business week1
I recently studied the AI intelligence for the business of South Hampton. It's an excellent opportunity to think about the whole AI Business processes. I thought I just use the AI platform and utilize some of the algorithms. But AI is not the perfect tool for solving the problem. They maybe aggregate the issue more and more, if you can use correctly and suitably. First, you can find the problem. That's the start of the AI process. There are many problems and AI algorithms. We have to select what we can improve by using AI.
1. Artificial intelligence definition: this article from TechEmergence
2. The philosophy of AI :
a very detailed response to his points
3.
3.
- Deep Learning: A Critical Appraisal.pdf
- Deep Learning
- The Dark Secret at the Heart of AI
- The Case Against Deep Learning Hype
- Thoughts on Gary Marcus’ Critique of Deep Learning
- Gary Marcus’ Deep Learning Critique Triggers Backlash
- Wise up, deep learning may never create a general purpose AI
- The Boogeyman Argument that Deep Learning will be Stopped by a Wall
- The Revolutionary Technique That Quietly Changed Machine Vision Forever
- Is All The Excitement Around Deep Learning Justified – Not According To Deep Learning’s Biggest Skeptic Gary Marcus
4.
2016년 6월 15일 수요일
deep learning useful materials
Deep learning lecture in Stanfordhttp://cs231n.stanford.edu/
Caffe Tutorial slides in Weiszmann institutehttp://www.wisdom.weizmann.ac.il/~vision/courses/2016_1/DNN/files/TA_lecture.pptx
NVIDIA’s open Deep Learning Courseshttps://developer.nvidia.com/deep-learningcourses
Caffe Tutorial in CVPR 2015http://tutorial.caffe.berkeleyvision.org/
Lecture notes (11, 12) in CS231n:
Convolutional Neural Networks for Visual Recognition (Stanford Univ.)
http://deeplearning.net/software_links/
https://github.com/soumith/convnetbenchmarks
YouTube: CS231n Winter 2016: Lecture 12: Deep Learning libraries
Caffe Tutorial slides in Weiszmann institutehttp://www.wisdom.weizmann.ac.il/~vision/courses/2016_1/DNN/files/TA_lecture.pptx
NVIDIA’s open Deep Learning Courseshttps://developer.nvidia.com/deep-learningcourses
Caffe Tutorial in CVPR 2015http://tutorial.caffe.berkeleyvision.org/
Lecture notes (11, 12) in CS231n:
Convolutional Neural Networks for Visual Recognition (Stanford Univ.)
– A comparison of Caffe, Torch, Theano and Tensorflow
Deep Learning Tutorials
http://deeplearning.net/tutorial/
Deep Learning Tutorials
Deep Learning is a new area of Machine Learning research, which has been introduced with the objective of moving Machine Learning closer to one of its original goals: Artificial Intelligence. See these course notes for a brief introduction to Machine Learning for AI and an introduction to Deep Learning algorithms.
Deep Learning is about learning multiple levels of representation and abstraction that help to make sense of data such as images, sound, and text. For more about deep learning algorithms, see for example:
- The monograph or review paper Learning Deep Architectures for AI (Foundations & Trends in Machine Learning, 2009).
- The ICML 2009 Workshop on Learning Feature Hierarchies webpage has a list of references.
- The LISA public wiki has a reading list and a bibliography.
- Geoff Hinton has readings from 2009’s NIPS tutorial.
The tutorials presented here will introduce you to some of the most important deep learning algorithms and will also show you how to run them using Theano. Theano is a python library that makes writing deep learning models easy, and gives the option of training them on a GPU.
The algorithm tutorials have some prerequisites. You should know some python, and be familiar with numpy. Since this tutorial is about using Theano, you should read over the Theano basic tutorial first. Once you’ve done that, read through our Getting Started chapter – it introduces the notation, and [downloadable] datasets used in the algorithm tutorials, and the way we do optimization by stochastic gradient descent.
The purely supervised learning algorithms are meant to be read in order:
- Logistic Regression - using Theano for something simple
- Multilayer perceptron - introduction to layers
- Deep Convolutional Network - a simplified version of LeNet5
The unsupervised and semi-supervised learning algorithms can be read in any order (the auto-encoders can be read independently of the RBM/DBN thread):
- Auto Encoders, Denoising Autoencoders - description of autoencoders
- Stacked Denoising Auto-Encoders - easy steps into unsupervised pre-training for deep nets
- Restricted Boltzmann Machines - single layer generative RBM model
- Deep Belief Networks - unsupervised generative pre-training of stacked RBMs followed by supervised fine-tuning
Building towards including the mcRBM model, we have a new tutorial on sampling from energy models:
- HMC Sampling - hybrid (aka Hamiltonian) Monte-Carlo sampling with scan()
- Building towards including the Contractive auto-encoders tutorial, we have the code for now:
- Contractive auto-encoders code - There is some basic doc in the code.
- Recurrent neural networks with word embeddings and context window:
- LSTM network for sentiment analysis:
- Energy-based recurrent neural network (RNN-RBM):
Note that the tutorials here are all compatible with Python 2 and 3, with the exception of Modeling and generating sequences of polyphonic music with the RNN-RBMwhich is only available for Python 2.
Silicon Valley AI Lab
http://svail.github.io/
Optimizing RNNs with Differentiable Graphs
Part II: Optimizing RNN performance
Date: June 14th, 2016
Authors: Jesse Engel
Differentiable graph notation provides an easy way to visually infer the gradients for complex neural networks. We also show several useful rules of thumb for optimizing graphs of new algorithms.
Date: June 14th, 2016
Authors: Jesse Engel
Differentiable graph notation provides an easy way to visually infer the gradients for complex neural networks. We also show several useful rules of thumb for optimizing graphs of new algorithms.
Persistent RNNs: 30 times faster RNN layers at small mini-batch sizes
Date: March 25th, 2016
Authors: Greg Diamos
YouTube: SVAIL Tech Notes: Accelerating RNNs by Stashing Weights On-Chip
At SVAIL, our mission is to create AI technology that lets us have a significant impact on hundreds of millions of people. We believe that a good way to do this is to improve the accuracy of speech recognition by scaling up deep learning algorithms on larger datasets than what has been done in the past.
Authors: Greg Diamos
YouTube: SVAIL Tech Notes: Accelerating RNNs by Stashing Weights On-Chip
At SVAIL, our mission is to create AI technology that lets us have a significant impact on hundreds of millions of people. We believe that a good way to do this is to improve the accuracy of speech recognition by scaling up deep learning algorithms on larger datasets than what has been done in the past.
Around the World in 60 Days: Getting Deep Speech to Work on Mandarin
Date: February 9th, 2016
Authors: Tony Han, Ryan Prenger
YouTube: SVAIL Tech Notes: Recognizing both English and Mandarin
In our recent paper Deep Speech 2, we showed our results in Mandarin. In just a few months, we had produced a Mandarin speech recognition system with a recognition rate better than native Mandarin speakers. Here we want to discuss what we did to adapt the system to Mandarin and how the end-to-end learning approach made the whole project easier.
Authors: Tony Han, Ryan Prenger
YouTube: SVAIL Tech Notes: Recognizing both English and Mandarin
In our recent paper Deep Speech 2, we showed our results in Mandarin. In just a few months, we had produced a Mandarin speech recognition system with a recognition rate better than native Mandarin speakers. Here we want to discuss what we did to adapt the system to Mandarin and how the end-to-end learning approach made the whole project easier.
Fast Open Source CPU/GPU Implementation of CTC
Date: January 14th, 2016
Contact: svail-questions@baidu.com
YouTube: SVAIL Tech Notes: Warp CTC
Warp-CTC from Baidu Research's Silicon Valley AI Lab is a fast parallel implementation of CTC, on both CPU and GPU. Warp-CTC can be used to solve supervised problems that map an input sequence to an output sequence, such as speech recognition. To get Warp-CTC follow the link above. If you are interested in integrating Warp-CTC into a machine learning framework reach out to us. We are happy to accept pull requests.
Contact: svail-questions@baidu.com
YouTube: SVAIL Tech Notes: Warp CTC
Warp-CTC from Baidu Research's Silicon Valley AI Lab is a fast parallel implementation of CTC, on both CPU and GPU. Warp-CTC can be used to solve supervised problems that map an input sequence to an output sequence, such as speech recognition. To get Warp-CTC follow the link above. If you are interested in integrating Warp-CTC into a machine learning framework reach out to us. We are happy to accept pull requests.
Investigating performance of GPU BLAS Libraries
Part I: Optimizing RNN performance
Date: November 17th, 2015
Author: Erich Elsen
Most researchers engaging in Neural Network research have been using GPUs for training for some time now due to the speed advantage they have over CPUs. GPUs from NVIDIA are almost universally preferred because they come with high quality BLAS (cuBLAS) and convolution (cuDNN) libraries.
Date: November 17th, 2015
Author: Erich Elsen
Most researchers engaging in Neural Network research have been using GPUs for training for some time now due to the speed advantage they have over CPUs. GPUs from NVIDIA are almost universally preferred because they come with high quality BLAS (cuBLAS) and convolution (cuDNN) libraries.
2016년 4월 25일 월요일
피드 구독하기:
글 (Atom)