Friday, December 02, 2016

How to Use AWS to Create an End-to-End HLS Streaming Stack

 How to Use AWS to Create an End-to-End HLS Streaming Stack

http://www.streamingmedia.com/Articles/Editorial/Featured-Articles/How-to-Use-AWS-to-Create-an-End-to-End-HLS-Streaming-Stack-94487.aspx

Redis:網路上的電子資源

Redis:網路上的電子資源

PPT_Statistical Analysis of Flow Data Using Pythonand Redis
PPT_Lecture27_Pythonand Redis
Little Redis Book
2016_Spring Data Redis
2013_Redis in Action
2011_Redis Cookbook

MongoDB:網路上的電子資源

MongoDB:網路上的電子資源

The Little MongoDB Book
MongoDB Tutorial
MongoDB Architecture Guide
2014_MongoDB Succinctly -PFC Evolution
2014_MongoDB Cookbook
2014_MongoDB Basics
2013_The Definitive Guide to MongoDB, A Complete Guide to Dealing with Big Data Using MongoDB
2013_Pro iOS and Android Apps for Business with jQuery Mobile, Node.js, and MongoDB
2013_Pro Hibernate and MongoDB
2013_MongoDB, The Definitive Guide
2013_MongoDB Applied Design Patterns
2012_MongoDB in Action
2012_MongoDB An introduction and performance analysis
2011_MongoDB and Python
2010_The definitive guide to MongoDB, the NoSQL database for cloud and desktop computing
2010_MongoDB, The Definitive Guide

-----

Advantages of MongoDB over RDBMS

◎ MongoDB is not trying to be the best at everything, it provides a rich document-orienteddatabase that’s optimized for speedand scalability.

◎ Uses internal memoryfor storing the (windowed) working set, enabling faster access of data.

◎ JSON(Java Script Object Notation)effectively describes all the contentin a given document.

◎ MongoDB uses an open data format called BSON (pronounced Bee-Son), which is short for binary JSON.

-----

Where to Use MongoDB?

◎ Big Data

◎ Content Management and Delivery

◎ Mobileand Social Infrastructure

◎ User Data Management

◎ Data Hub

-----

Additional Drivers
◎ C
◎ C++
◎ C#
◎ Erlang
◎ Go
◎ Java
◎ JavaScript
◎ Node.js
◎ Perl
◎ PHP
◎ Python
◎ Ruby
◎ Scala

AWS雲端企業實戰聖經

AWS雲端企業實戰聖經

2011年初版一刷。

我雖然後知後覺,還好不置於不知不覺。

這本也要好好K一K,因為圖書館有,而且只有這本中文書。

這本清大跟交大都有,而且就這本中文書而已。清大另有一本西書。

還好網路電子資源不算少。

Learning Big Data with Amazon Elastic MapReduce

Learning Big Data with Amazon Elastic MapReduce

接下來這週要看AWS。

剛開始看。

心得是,我很高興之前沒有開始真的創業,因為要做Hadoop,結果不知道AWS,豈不是太好笑了嗎?

這本書的特點,應該是如何在AWS上跑Hadoop的MapReduce。

答案是:EMR。

Seven databases in Seven Weeks

Seven databases in Seven Weeks

最近因為工作需要,我在圖書館找到這本書。

書上介紹了七種目前當紅的資料庫(想當然爾)。

包括:

PostgreSQL
Riak
HBase
MongoDB
CouchDB
Neo4J
Redis

其中HBase是用在Hadoop上,之前已經稍微接觸過,但目前進度幾乎還是零。

MongoDB跟Redis是最近可能會用到的。

MongoDB:適合大數據的資料庫,資料庫容易擴充到很大,比傳統關連式資料庫處理速度快很多。冗餘資料清除後,存放到Redis。

Redis:使用內部記憶體儲存的資料庫,存取超快。配合Spark使用。

希望這不算誤導到人。

打工週記0025:植福‧敬業

打工週記0025:植福‧敬業

2016/12/02

匆匆打工已經一個月了。

前三週plan的是chatbot。第一週是旅遊類,第二週是金融科技類,第三週我把所有最熱的,甚至machine learning跟AI都加進去了。

第四週,局勢忽而轉為big data,我plan了AWS, Hadoop, Spark, Python, Redis, MongoDB, Crawler, Recommender。一週的時間,對這麼多的領域,我能做什麼呢?就是找了些電子書,把對專案有用的資訊找出來,然後擬了一個架構如何開發系統。這樣算是打通了十八銅人陣的第一關。

oh captain my captain

ㄡ,老闆我的老闆

我是個博士,不是神。我的專長在研究,已經不再是年輕的工程師。

老闆威脅利誘,軟硬兼施,好話歹話說盡,要我扛下這個專案,而且可能只有一個半工程師給我。

ㄡ,老闆我的老闆

我是個工讀生,不是你的首席科學家。我對創業啦、首席科學家啦,都沒興趣了。你教我的是:創業第一年先活下去。所以你只要每週付我一點顧問費,我就會幫你plan當週的計畫,銀貨兩訖,各不相欠。

老闆其實是我的大學室友,三十年前幫過我一次,三十年後又幫我一次。只是這一次算互相幫忙。四週內plan出來的計畫,遠超出我之前的水準。

ㄡ超人我簡直是超人了

所以啦,我現在的人生座右銘是「植福」與「敬業」。「拿人錢財,與人消災」。 顧問費不會白拿,表現不佳隨時解雇,這樣我就心滿意足了!

我的人生座右銘是「植福」與「敬業」,不是「救火」與「砲灰」。我是很想幫您,但是考量到你我資本實力的差異,到底是您要幫我,還是我要幫您呢?

互相幫忙吧!我很珍惜上個月跟未來兩個月的共事時間。這幾個月我學到的,不下於過去一年。 接下來,我要開始銅人陣第二關了。


Fig. 1


Fig. 2

-----

死神在新世紀,招手

缺角的風箏,高飛

向何處?未知,且待.....

----- 摘自《獵人》庫洛洛給西索的預言詩

◎◎◎◎ 人工智慧

人工智慧

https://zh.wikipedia.org/wiki/%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD

-----

chatbot=IM+AI

DL=MLP+AE+CNN+RNN
CNN=con.+MP
RNN=VRNN+GRNN

AI=DL∈ML=(cla.+clu.+RS)∈DM
chatbot=NLP+IM+AI, NLP∈AI
chatbot=IM+AI

-----

Chatbot 熱潮方興未艾,然而搜尋資料的過程中,經常看到 machine learning (ML), deep learning (DL), artificial intelligence (AI), data mining (DM) 等關鍵字。大體上,DL [1]-[7] 是AI領域目前最熱的一支。DL的基礎是 artificial neural networks (ANN),是ML的一種。ML主要由三大領域構成,包含分類 (classification)、分群 (clustering),以及推薦系統 (recommendation system) [8]。以上的等式不能說完全正確,但有助於我們瞭解上述學門的關連性。

DL有四種主要的技術,分別是 multilayer perceptron (MLP), autoencoder (AE), convolutional  neural networks (CNN), recurrent neural networks (RNN) [2]。其中以CNN最重要(最熱門)[1]。複雜的方程式不容易看懂,但從以上資料可以了解一些概念,若想更深入一點,在Nature的一篇文章中,有 更仔細的說明 [4]。雖然簡單把DL分成四大類,但你想有可能這麼輕鬆嗎 [3]?

本篇短文簡單說明 ML, DL, DM, AI的定義與相互關係,後續若有新的資訊,會再更新內容。

-----

Abbreviations in alphabetical order

AE: autoencoder
AI: artificial intelligence
ANN: artificial neural networks
cla.: classification
clu.: clustering
CNN: convolutional  neural networks
DL: data mining
DM: data mining
GRNN: Gated Recurrent Neural Network
IM: instant messaging
ML: machine learning
MLP: multilayer perceptron
MP: max-pooling
NLP: natural language processing
RNN: recurrent neural networks
RS: recommendation system
VRNN: Vanilla Recurrent Neural Network

-----

References

[1] Lacey, Griffin, Graham W. Taylor, and Shawki Areibi. "Deep Learning on FPGAs: Past, Present, and Future." arXiv preprint arXiv:1602.04283 (2016).

[2] Wang, Hao, and Dit-Yan Yeung. "Towards Bayesian Deep Learning: A Survey." arXiv preprint arXiv:1604.01662 (2016).

[3] Schmidhuber, Jürgen. "Deep learning in neural networks: An overview." Neural Networks 61 (2015): 85-117.

[4] LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. "Deep learning." Nature 521.7553 (2015): 436-444.

[5] Yu, Dong, Li Deng, and D. Yu. "Deep Learning Methods and Applications." Foundations and Trends in Signal Processing (2014).

[6] Bengio, Yoshua, Aaron C. Courville, and Pascal Vincent. "Unsupervised feature learning and deep learning: A review and new perspectives." CoRR, abs/1206.5538 1 (2012).

[7] Bengio, Yoshua. "Learning deep architectures for AI." Foundations and trends® in Machine Learning 2.1 (2009): 1-127.

[8] Owen, Sean, et al. "Mahout in action." (2012).

◎◎◎ 新方向

新方向

◎ FB社團

-----

◎ 技術類:

BOTS

Chatbots

Chatbot Developers

[Chatbots] Developers (Global)

Chatbot Developers Taiwan

Chatbot UX Taiwan

ChatBot Developer 仙台

Jobs in the Bot Space

Showcase you Chatbot

Messenger Platform Developer Community

日語教學chatbot團隊

線上 chatbot 讀書會

線上 nodejs 讀書會

人工智慧

人工智慧/自然語言處理/搜尋引擎

OpenAI@TW

AI & Deep Learning Enthusiasts Bay Area

HH Machine Learning and Artificial Intelligence

Wit.ai Hackers

GPU Taiwan Facebook

Taiwan Machine Learning for Everything

Artificial Intelligence & Deep Learning


golang

Go程式語言 (Golang Taiwan, Gopher Taipei)

Python Taiwan


Front-End Developers Taiwan

iOS Apple手機App及Swift應用程式開發工程師社團

Android手機App應用程式開發工程師社團

2016年系統軟體課程

開源系統軟體

-----

◎ 語文類:

線上日語讀書會

日文學習同好社

-----

◎ 創業類:

新創x商業-每天20分鐘 思考商業與新創

台灣電子商務創業聯誼會

Tuesday, November 29, 2016

draft

https://github.com/onlinereadbook/office

http://www.investopedia.com/articles/personal-finance/093015/whats-good-profit-margin-new-business.asp 

https://www.azure.cn/cognitive-services

https://medium.com/@robot_MD/when-bias-in-product-design-means-life-or-death-ea3d16e3ddb2#.6w96s9lzq

https://www.crowdflower.com/7-myths-ai/?utm_content=buffer04be5&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer

https://ai100.stanford.edu/2016-report?utm_content=buffer27405&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer

http://www.nytimes.com/2016/09/02/technology/artificial-intelligence-ethics.html?_r=1

https://www.quora.com/Who-is-leading-in-AI-research-among-big-players-like-IBM-Google-Facebook-Apple-and-Microsoft?utm_content=buffer21bca&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer

docker

http://www.rightrelevance.com/search/articles/hero?article=d13323d018cad82379acebe869c954a28eca2c69&query=python&taccount=pythonrr

http://www.infoworld.com/article/3134765/artificial-intelligence/microsoft-adds-python-to-deep-learning-toolkit.html

http://pythonsparkhadoop.blogspot.tw/2016/10/pythonspark_21.html

https://deal.codetrick.net/p/HkxF9uHuSi/introduction-to-machine-learning-and-face-detection-in-python/

http://pythonsparkhadoop.blogspot.tw/2016/10/pythonspark-20-hadoop.html

http://www.rightrelevance.com/search/articles/hero?article=edb1fe634951d9a81108ecfe8010a97238366eaf&query=python&taccount=pythonrr

http://www.rightrelevance.com/search/articles/hero?article=7711c67eaa6142e2762fbe54930f972253b0e8dd&query=python&taccount=pythonrr

https://school.codequs.com/p/SklYYybn/learn-how-to-create-hadoop-mapreduce-jobs-in-python/

https://school.codequs.com/p/BJepvfFs/deep-learning-convolutional-neural-networks-in-python/

http://www.thedevmasters.com/4-essential-skills-to-become-data-scientist/

https://aws.amazon.com/tw/blogs/aws/three-new-aws-reference-architectures-for-e-commerce/

https://www.chalkstreet.com/data-science-with-spark-and-python/?utm_source=Profiles&utm_medium=Facebook&utm_campaign=nm

https://www.analyticsvidhya.com/blog/2015/06/infographic-cheat-sheet-data-exploration-python/?utm_content=bufferb570c&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer

http://www.rightrelevance.com/search/articles/hero?article=0aef589a6bb0f45297cf901a89d69c1bf8fa0c34&query=python&taccount=pythonrr

http://www.rightrelevance.com/search/articles/hero?article=2080cba8aaec64cf7b8423872b978f0ff31314d5&query=python&taccount=pythonrr

http://www.allthingsdistributed.com/2016/11/mxnet-default-framework-deep-learning-aws.html

http://iknow.stpi.narl.org.tw/post/Read.aspx?PostID=12689&utm_content=buffer9f3fe&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer

https://medium.com/intuitionmachine/a-pattern-language-for-deep-learning-30de291434e1#.r9g206x0c

https://medium.com/intuitionmachine/11-biases-why-experts-are-missing-the-train-in-deep-learning-1c092fa9cace#.rcg7ntyti

https://adeshpande3.github.io/adeshpande3.github.io/The-9-Deep-Learning-Papers-You-Need-To-Know-About.html

http://timdettmers.com/2014/08/14/which-gpu-for-deep-learning/

https://techcrunch.com/2016/11/22/telegram-telegraph/

Popular Deep Learning Hardware Tools

Popular Deep Learning Hardware Tools

引用臉書資料:

「金錢也是另一門檻,一般人一定買不起DGX-1,差一點的,像是M40也很難買得起,就算是買TitanX也得花掉不少錢。可是你設備不夠好,你就跑不 快, 跑不快就無法嘗試那些無法解釋的參數組合,那就跑不出結果。跑不出結果,懂再多理論也做不出東西,所以這才是門檻最高的地方。」

「順便問一下樓上 1070 跑得動 VGG16嗎 感覺不太夠力」

「VGG很耗記憶體,但如果把batch弄小一點應該是可以...」

-----

DGX-1

Nvidia發布超級計算機DGX-1 售價超80萬元- 今日頭條 big5.jinri-toutiao.com/id/337559.html 2016年4月7日 - 它配備了7TB固態硬盤,8塊Tesla P100顯卡和2塊英特爾Xeon處理器——如此的配置也給起帶來了超高的處理性能(170萬億次浮點運算/秒), ...

http://big5.jinri-toutiao.com/id/337559.html

NVIDIA在今年的GTC大會上推出了包括GPU晶片Tesla P100(基於全新Pascal平台架構上打造)、DRIVE PX2(用於自動駕駛汽車的開發平台)和用於深度學習研究的超級計算機DGX-1(使用Tesla P100晶片建造,運算速度可達170萬億次)在內的一系列新技術和新產品。

https://read01.com/dxj7Le.html

-----

M40

如果只有深度學習的訓練,NVIDIA的Tesla M40/M4雖然不便宜,但企業或者機構購買還是比較合適的(百度的深度學習研究院就用的這一款),相對於K40單精度浮點運算性能是4.29Tflops,M40可以達到7Tflops。

https://read01.com/dxj7Le.html

-----

Titan X

NVIDIA 新一代卡王,US$1,200 一張的Titan X 現身 - Engadget 中文版 chinese.engadget.com/2016/.../nvidias-new-top-end-graphics-card-is-the-1-200-titan-... 2016年7月22日 - 如果你剛買了張NVIDIA GTX 1080,正在享受擁有地表最快的桌機遊戲顯卡的快感的話,那... 壞消息告訴你,GTX 1080 已經被新一代的Titan X 踢到 ...

http://chinese.engadget.com/2016/07/22/nvidias-new-top-end-graphics-card-is-the-1-200-titan-x/

-----

1070

最強顯示卡現身! NVIDIA 推出GTX 1080、GTX 1070 | 自由電子報3C科技 3c.ltn.com.tw/news/24282 2016年5月7日 - NVIDIA 終於正式推出了新一代顯示卡王者GTX 1080、GTX 1070,作為取代目前GTX 980 和GTX 970 的產品。這兩張顯示卡採用Pascal 架構, ...

http://3c.ltn.com.tw/news/24282

-----

VGG16

關於圖像語義分割的總結和感悟- IT閱讀 www.itread01.com/articles/1476698476.html 2016年10月17日 - 卷積化即是將普通的分類網絡,比如VGG16,ResNet50/101等網絡丟棄全連接層,換上對應的卷積層即可。如下圖: 這裏寫圖片描述 ...

http://www.itread01.com/articles/1476698476.html

前沿| 二值神經網絡:催生手腕上的AlphaGo : 歌穀穀 www.gegugu.com/2016/04/04/5056.html 2016年4月4日 - 後續的VGG-16網絡相比AlexNet提高瞭預測準確率,卻更是一個網絡大小超過500M,單圖片浮.計算量超過280億次的龐然大物。人們對預測準確率 ...

http://www.gegugu.com/2016/04/04/5056.html

Popular Deep Learning Software Tools

 Popular Deep Learning Software Tools

Lacey, Griffin, Graham W. Taylor, and Shawki Areibi. "Deep Learning on FPGAs: Past, Present, and Future." arXiv preprint arXiv:1602.04283 (2016).

-----

‧Caffe, developed by the Berkeley Vision and Learning Center, has unofficial support for OpenCL under the name project GreenTea [2]. There is also an AMD version of Caffe that supports OpenCL [1].

‧Torch, a scientific computing framework written in Lua, is widely used and has unofficial support for OpenCL under the project CLTorch [6].

‧Theano, developed by the University of Montreal, has unofficial support for OpenCL under the work-in-progress gpuarray backend [5].

‧DeepCL is an OpenCL library to train deep convolutional neural networks, developed by Hugh Perkins [3].

-----

Ref.

[1] Caffe-OpenCL.
https://github.com/amd/OpenCL-caffe/wiki, 2015.
[2] Caffe: project greentea.
https://github.com/BVLC/caffe/pull/2195, 2015.
[3] DeepCL.
https://github.com/hughperkins/DeepCL, 2015.
[5] Theano: gpuarray backend.
http://deeplearning.net/software/libgpuarray/index.html, 2015.
[6] Torch: cltorch.
https://github.com/hughperkins/cltorch, 2015.

-----

Table 1: Overview of Deep Learning Frameworks with OpenCL Support

Tool / Core Language / Bindings / OpenCL / User Base
Caffe / C++ / Python, MATLAB / Partial Support / Large
Torch / Lua / - / Partial Support / Large
Theano / Python / - / Minimal Support / Large
DeepCL / C++ / Python, Lua /Full Support / Moderate

What VCs Look For In Bot Investments

What VCs Look For In Bot Investments

http://www.topbots.com/vc-bot-investor-look-bot-investment/?utm_medium=article&utm_source=fb&utm_campaign=vcbotinvestments

這篇文章很棒,我之前想的方向跟它是一致的,也就是AI是chatbot的核心。在chatbot=IM+AI_v002的投影片(bot_v005),一開始有一個統計報告,指出有不少團隊開發自己的framework,而不是拉現成的。

開發者可以花一點時間看一下。英文摘要如下:

Entrepreneurs Underestimate the Complexity of Enterprise Bots

The first challenge is discovery. Each messaging platform solves discovery differently.

Another challenge is that current natural language processing (NLP) and machine learning (ML) technologies are not sophisticated enough to drive excellent user experiences.

跑步(四六):10圈

跑步(四六):10圈

2015/11/29

赤腳。跑1,伸展,跑(5*2),伸展。逆10。

-----

一般來說,沒時間,一星期兩次運動是被建議的。

我現在是有機會就跑,為什麼?萬一有事要忙,或者下雨,就不能跑了。

今天天氣不錯,而最近,感到兩隻腳很輕。

我上癮了!!! :D

Monday, November 28, 2016

跑步(四五):10圈

跑步(四五):10圈

2015/11/28

赤腳。跑1,伸展,跑(5*2),伸展。逆10。

-----


冬天真的到了,即使有陽光,溫度還是低。

我穿著排汗衫、棉質運動衣跟薄運動夾克跑步,褲子還是短褲,

這樣說來,冬天真的還沒到。


靠壘球場的跑道滿是樹枝,赤腳的我,只好靠著內側跑。

由於最近工作進展非常順利,跑步後,我也有充足的休息時間復原了!

Saturday, November 26, 2016

跑步(四四):10圈

跑步(四四):10圈

2015/11/26

赤腳。跑1,伸展,跑(5*2),伸展。逆10。

-----

冬天到了,不會太早起床,也不會太早跑步.....

今天星期六,早上天氣不錯,還出大太陽,可以打赤膊跑步。

有陽光,所以是傳統的逆時針跑。

印度人玩板球,板球有什麼好玩呢?印度人不一定知道,但肯定覺得好玩。

膝蓋OK,腳踝OK,心臟OK,

什麼不OK呢?早餐前跑步好像不大OK了,跑到十圈就差不多了。

我該思考跑步與補給的關係了!

Friday, November 25, 2016

使用 Python與Spark建立推薦引擎

使用 Python與Spark建立推薦引擎

http://pythonsparkhadoop.blogspot.tw/2016/10/pythonspark.html

NVIDIA Is Not Just Accelerating AI, It Aims To Reshape Computing

NVIDIA Is Not Just Accelerating AI, It Aims To Reshape Computing

NVIDIA Is Not Just Accelerating AI, It Aims To Reshape Computing www.forbes.com/.../nvidia-is-not-just-accelerating-ai-it-aims-to-reshape... 翻譯這個網頁 2016年11月17日 - NVIDIA has announced new partnerships and capabilities this week at SuperComputing '16 to further the company's vision of the accelerated ...

http://www.forbes.com/sites/moorinsights/2016/11/17/nvidia-is-not-just-accelerating-ai-it-aims-to-reshape-computing/?partner=yahootix&yptr=yahoo&utm_content=buffer5953d&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer#779ee2f32732

Key trends in machine learning and AI

Key trends in machine learning and AI

Key trends in machine learning and AI | TechCrunch https://techcrunch.com/2016/07/.../key-trends-in-machine-learning-and... 翻譯這個網頁 2016年7月6日 - You can hardly talk to a technology executive or developer today without talking about artificial intelligence, machine learning or bots.

https://techcrunch.com/2016/07/06/key-trends-in-machine-learning-and-ai/

Why nature is our best guide for understanding artificial intelligence

Why nature is our best guide for understanding artificial intelligence

Why nature is our best guide for understanding artificial intelligence ... https://techcrunch.com/.../why-nature-is-our-best-guide-for-understandi... 翻譯這個網頁 4 天前 - In living organisms, evolution is a multi-generational process where mutations in genes are dropped and added. Well-adapted organisms ...

https://techcrunch.com/2016/11/20/why-nature-is-our-best-guide-for-understanding-artificial-intelligence/

打工週記0024:古寺‧愁城

打工週記0024:古寺‧愁城

2016/11/25

局勢還不能讓我完全放鬆。
局勢已經讓我開始有點放鬆了.....

匆匆由創業變成打工的狀態也已三週。
這八個月來常跑台北,最後漸漸就剩大直。
這幾週,上工前我總喜歡先在劍南路捷運站旁的石椅上,脫掉襪子先坐一陣,觀察一下來來去去的人群,聽一下不甚喧擾的汽機車噪音,隔了有點遠,所以聽起來有點遠,不甚吵。

上工後頭腦要全速運轉,上工前先放空一下。

也許是上週跟老闆去過植福宮拜拜了,也許放鬆了,本週,我決定到附近先溜達溜達。
劍潭古寺是最近的。為何在大直卻叫劍潭古寺,因為古寺是日據時因擴建神社被遷過來的。

原本是觀音廟,佛光山接管後,改成供奉釋迦牟尼佛,正殿兩側還有十八羅漢。左側是辦公區,我看了結緣的善書,有豪華精裝的妙法蓮花經,也有英漢對照的地藏菩薩本願經。惡習不改,又想請回家看一遍。本來打算下班後再過來拿,但是下班總是不會再過來了,當下即知。


Fig. 1

旁邊碑林保留有原址一些文物。比較有意思的是還有一間小廟,叫多寶寺,還被鎖起來。我喵了喵裡面,有神像。查了一下網路,中國跟日本都有叫多寶寺的廟。但不知這個多寶寺有何來歷?


Fig. 2


Fig. 3

廟旁的高樓,在前人旅記的文章上尚未見到。孤峰獨聳,面相上是孤剋的象徵。風水上,我冒充一下專家好了:若有年輕女性獨住,身體不好,不利婚姻,但事業有成,可昇為高階主管。


Fig. 4


Fig. 5

離開石椅的前一週,我領悟到,創業不離「植福」與「敬業」,人生亦復如是。為什麼呢?因為要到公司,得從植福路轉敬業路。


Fig. 6


Fig. 7

https://hbr.org/2016/11/hiring-your-first-chief-ai-officer 

總經理跟董事長都很客氣,要我名片上面印上首席科學家,然而,在真正拿到一紙待遇優厚的合約之前,我不過是個兼職的博士後工讀生罷了。這也要公司先增資成功,才有舞台讓我發揮吧。

凡此種種,都不是我能掌握的。唯一確定的是:遊畢劍潭古寺,不再坐困愁城!

chatbot=IM+AI_v002

chatbot=IM+AI_v002

-----

v001:簡介machine learning與deep learning

2016/11/23

我查了一些資料,把一些困擾已久的名詞(machine learning, deep learning, data mining等)搞清楚,我想這對chatbot的開發者也是有用的,貼出來讓大家參考一下。

-----

v002:增加chatbot、machine learning、deep learning相關的links

2016/11/5

我這三個禮拜,蒐集了將近一百篇不錯的文章的links,分別關於chatbot、machine learning、deep learning。資料有分類,但尚無每篇的閱讀心得(估計也沒時間寫了XD)。缺點是platform和framework尚未加入(應該會找時間 做)。不管是對chatbot、data science、AI有興趣的開發者,應該都會有些幫助。

-----

DL=MLP+AE+CNN+RNN
CNN=con.+MP
RNN=VRNN+GRNN

AI=DL∈ML=(cla.+clu.+RS)∈DM
chatbot=NLP+IM+AI, NLP∈AI
chatbot=IM+AI

-----

Chatbot 熱潮方興未艾,然而搜尋資料的過程中,經常看到 machine learning (ML), deep learning (DL), artificial intelligence (AI), data mining (DM) 等關鍵字。大體上,DL [1]-[7] 是AI領域目前最熱的一支。DL的基礎是 artificial neural networks (ANN),是ML的一種。ML主要由三大領域構成,包含分類 (classification)、分群 (clustering),以及推薦系統 (recommendation system) [8]。以上的等式不能說完全正確,但有助於我們瞭解上述學門的關連性。

DL有四種主要的技術,分別是 multilayer perceptron (MLP), autoencoder (AE), convolutional  neural networks (CNN), recurrent neural networks (RNN) [2]。其中以CNN最重要(最熱門)[1]。複雜的方程式不容易看懂,但從以上資料可以了解一些概念,若想更深入一點,在Nature的一篇文章中,有 更仔細的說明 [4]。雖然簡單把DL分成四大類,但你想有可能這麼輕鬆嗎 [3]?

本篇短文簡單說明 ML, DL, DM, AI的定義與相互關係,後續若有新的資訊,會再更新內容。

-----

Abbreviations in alphabetical order

AE: autoencoder
AI: artificial intelligence
ANN: artificial neural networks
cla.: classification
clu.: clustering
CNN: convolutional  neural networks
DL: data mining
DM: data mining
GRNN: Gated Recurrent Neural Network
IM: instant messaging
ML: machine learning
MLP: multilayer perceptron
MP: max-pooling
NLP: natural language processing
RNN: recurrent neural networks
RS: recommendation system
VRNN: Vanilla Recurrent Neural Network

-----

References

[1] Lacey, Griffin, Graham W. Taylor, and Shawki Areibi. "Deep Learning on FPGAs: Past, Present, and Future." arXiv preprint arXiv:1602.04283 (2016).

[2] Wang, Hao, and Dit-Yan Yeung. "Towards Bayesian Deep Learning: A Survey." arXiv preprint arXiv:1604.01662 (2016).

[3] Schmidhuber, Jürgen. "Deep learning in neural networks: An overview." Neural Networks 61 (2015): 85-117.

[4] LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. "Deep learning." Nature 521.7553 (2015): 436-444.

[5] Yu, Dong, Li Deng, and D. Yu. "Deep Learning Methods and Applications." Foundations and Trends in Signal Processing (2014).

[6] Bengio, Yoshua, Aaron C. Courville, and Pascal Vincent. "Unsupervised feature learning and deep learning: A review and new perspectives." CoRR, abs/1206.5538 1 (2012).

[7] Bengio, Yoshua. "Learning deep architectures for AI." Foundations and trends® in Machine Learning 2.1 (2009): 1-127.

[8] Owen, Sean, et al. "Mahout in action." (2012).

-----

Appendix

[1] chatbot、machine learning、deep learning相關的links

Wednesday, November 23, 2016

盛世之下的科大訊飛 能否避免全球語音巨頭Nuance的悲劇?

盛世之下的科大訊飛 能否避免全球語音巨頭Nuance的悲劇?

盛世之下的科大訊飛能否避免全球語音巨頭Nuance的悲劇? - 每日頭條 https://kknews.cc/tech/pzkmnz.html 2016年10月20日 - 來這裡找志同道合的小夥伴!這兩天,因為錘子手機M1而大火的科大訊飛,關注度大有超過錘子的勢頭,股價更是連續上漲。不管怎麼樣,一場發布 ...

https://kknews.cc/tech/pzkmnz.html

[摘要] 2015_Deep learning

Deep learning

LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. "Deep learning." Nature 521.7553 (2015): 436-444.

-----

Deep convolutional nets have brought about breakthroughs in processing images, video, speech and audio, whereas recurrent nets have shone light on sequential data such as text and speech.

-----


Representation learning is a set of methods that allows a machine to be fed with raw data and to automatically discover the representations needed for detection or classification.

Deep-learning methods are representation-learning methods with multiple levels of representation.

An image, for example, comes in the form of an array of pixel values, and the learned features in the first layer of representation typically represent the presence or absence of edges at particular orientations and locations in the image.

The second layer typically detects motifs by spotting particular arrangements of edges, regardless of small variations in the edge positions.

The third layer may assemble motifs into larger combinations that correspond to parts of familiar objects, and subsequent layers would detect objects as combinations of these parts.

The key aspect of deep learning is that these layers of features are not designed by human engineers: they are learned from data using a general-purpose learning procedure.

-----

Supervised learning

Imagine that we want to build a system that can classify images as containing, say, a house, a car, a person or a pet.

In a typical deep-learning system, there may be hundreds of millions of these adjustable weights, and hundreds of millions of labelled examples with which to train the machine.

In practice, most practitioners use a procedure called stochastic gradient descent (SGD).

It is called stochastic because each small set of examples gives a noisy estimate of the average gradient over all examples.

After training, the performance of the system is measured on a different set of examples called a test set.

A linear classifier, or any other ‘shallow’ classifier operating on raw pixels could not possibly distinguish the latter two, while putting the former two in the same category.

The conventional option is to hand design good feature extractors, which requires a considerable amount of engineering skill and domain expertise.

But this can all be avoided if good features can be learned automatically using a general-purpose learning procedure. This is the key advantage of deep learning.

With multiple non-linear layers, say a depth of 5 to 20, a system can implement extremely intricate functions of its inputs that are simultaneously sensitive to minute details — distinguishing Samoyeds from white wolves — and insensitive to large irrelevant variations such as the background, pose, lighting and surrounding objects.

-----

Backpropagation to train multilayer architectures

-----

由於沒有時間,所以我很快地讀完這篇。

Deep learning的強大,比起傳統信號處理,就不說了。

我認為,重點有一段時間將會是硬體,這也不是我說的,晶片大廠都已磨刀霍霍了。

這離我還太遠,先把CNN跟RNN搞清楚再說。

Deep learning from Jason Tsai (FB)

Deep learning from Jason Tsai (FB)

如果自行閱讀 MIT Press 甫印行出版的這本「Deep Learning」深度學習大作有極大困難的話,下面是昨天「Deep Learning 101」meetup 最後提到的一些相對初階的教材:
1. Michael Nielsen 寫的「Neural Networks and Deep Learning」線上電子書
http://neuralnetworksanddeeplearning.com/

GitBook 上有簡體版的翻譯(不全)和閱讀筆記
https://www.gitbook.com/…/neural-networks-and-deep-…/details
https://www.gitbook.com/…/neural-networks-and-deep-…/details
https://www.gitbook.com/…/neural-networks-and-deep-…/details
2. Quoc V. Le (音譯黎國越) 的 「A Tutorial on Deep Learning」 Part I & II
http://www.trivedigaurav.com/…/quoc-les-lectures-on-deep-l…/
3. Stanford 大學的 CS231n (CNN) 和 CS224d (NLP) 課程
http://cs231n.stanford.edu/
http://cs224d.stanford.edu/
另外,如果您是專攻 NLP,特別是 Speech Recognition 領域,應該知道這一本也很艱深的大作:
「Automatic Speech Recognition: A Deep Learning Approach」(2015), by Dong Yu (俞棟) & Li Deng (鄧力)
http://www.springer.com/us/book/9781447157786
大陸有出簡體中譯本,書名「解析深度學習:語音識別實踐」,電子工業出版社出版。
如果您只是想要可以速成的 AI 教材,我只能告訴你 There is no such thing! 天下沒有白吃的午餐。
推薦看一下兩位年輕傑出學者,台大李宏毅教授和清大孫民教授的觀點
https://www.facebook.com/groups/Taiwan.AI.Group/permalink/1708217909500643/

蛋白質

蛋白質

你到底需要多少蛋白質?你攝取夠多了嗎?確認的方法之一就是評估你的身體和精神健康狀況。如果以下的指標出現在你身上,你的飲食裡你可能就需要更多的蛋白質:

經常感冒或喉嚨痛
練習之後恢復緩慢
脾氣暴躁
對訓練的反應不佳(很難達到良好體能)
手指甲生長速度緩慢以及指甲易斷
稀疏的頭髮或異常的掉髮
習慣性疲勞
精神不集中
嗜吃甜食
臉色蒼白
月經中止

要注意的是,光靠這些指標並無法證實你需要更多的蛋白質,因為每個指標的癥狀,背後都可能會有其他的成因。

-- 摘自《鐵人三項訓練聖經》p.230

chatbot=IM+AI_v001

chatbot=IM+AI_v001

DL=MLP+AE+CNN+RNN
CNN=con.+MP
RNN=VRNN+GRNN

AI=DL∈ML=(cla.+clu.+RS)∈DM
chatbot=NLP+IM+AI, NLP∈AI
chatbot=IM+AI

-----

Chatbot 熱潮方興未艾,然而搜尋資料的過程中,經常看到 machine learning (ML), deep learning (DL), artificial intelligence (AI), data mining (DM) 等關鍵字。大體上,DL [1]-[7] 是AI領域目前最熱的一支。DL的基礎是 artificial neural networks (ANN),是ML的一種。ML主要由三大領域構成,包含分類 (classification)、分群 (clustering),以及推薦系統 (recommendation system) [8]。以上的等式不能說完全正確,但有助於我們瞭解上述學門的關連性。

DL有四種主要的技術,分別是 multilayer perceptron (MLP), autoencoder (AE), convolutional  neural networks (CNN), recurrent neural networks (RNN) [2]。其中以CNN最重要(最熱門)[1]。複雜的方程式不容易看懂,但從以上資料可以了解一些概念,若想更深入一點,在Nature的一篇文章中,有更仔細的說明 [4]。雖然簡單把DL分成四大類,但你想有可能這麼輕鬆嗎 [3]?

本篇短文簡單說明 ML, DL, DM, AI的定義與相互關係,後續若有新的資訊,會再更新內容。

-----

Abbreviations in alphabetical order

AE: autoencoder
AI: artificial intelligence
ANN: artificial neural networks
cla.: classification
clu.: clustering
CNN: convolutional  neural networks
DL: data mining
DM: data mining
GRNN: Gated Recurrent Neural Network
IM: instant messaging
ML: machine learning
MLP: multilayer perceptron
MP: max-pooling
NLP: natural language processing
RNN: recurrent neural networks
RS: recommendation system
VRNN: Vanilla Recurrent Neural Network

-----

References

[1] Lacey, Griffin, Graham W. Taylor, and Shawki Areibi. "Deep Learning on FPGAs: Past, Present, and Future." arXiv preprint arXiv:1602.04283 (2016).

[2] Wang, Hao, and Dit-Yan Yeung. "Towards Bayesian Deep Learning: A Survey." arXiv preprint arXiv:1604.01662 (2016).

[3] Schmidhuber, Jürgen. "Deep learning in neural networks: An overview." Neural Networks 61 (2015): 85-117.

[4] LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. "Deep learning." Nature 521.7553 (2015): 436-444.

[5] Yu, Dong, Li Deng, and D. Yu. "Deep Learning Methods and Applications." Foundations and Trends in Signal Processing (2014).

[6] Bengio, Yoshua, Aaron C. Courville, and Pascal Vincent. "Unsupervised feature learning and deep learning: A review and new perspectives." CoRR, abs/1206.5538 1 (2012).

[7] Bengio, Yoshua. "Learning deep architectures for AI." Foundations and trends® in Machine Learning 2.1 (2009): 1-127.

[8] Owen, Sean, et al. "Mahout in action." (2012).

Tuesday, November 22, 2016

Telegram Bot 開發起手式

Telegram Bot 開發起手式

2016-07-19 on JavaScript, Telegram, Bot

https://neighborhood999.github.io/2016/07/19/Develop-telegram-bot/

跑步(四三):10圈

跑步(四三):10圈

2015/11/22

赤腳。跑1,伸展,跑(5*2),伸展。順10。

-----

上週跑步不順,星期三清大運動會,星期六幼稚園運動會,星期天醫師協會運動會。

要不然就下雨。

今天跑完時也下起雨,雨不大,但也可以把人趕回家,還好剛好跑完。

休息了這麼多天,有影響嗎?

主要不是休息的關係,今天跑起來很順。

可以用比較快的速度跑,雖然是有點喘,意外的是並不累,表示可以用較高的心臟輸出運動了。

跑姿還是繼續調整。持續感覺到四頭肌跟臀大肌的作用。

到最後一圈,雨滴變大了,於是我邁開大步,全然地享受跑步。

今天是一個新的里程碑。

我還是專注在五千,但是可能可以開始跑快一點了!

FinTech百強第一的螞蟻金服,用人工智慧驅動「普惠金融」願景

FinTech百強第一的螞蟻金服,用人工智慧驅動「普惠金融」願景

FinTech百強第一的螞蟻金服,用人工智慧驅動「普惠金融」願景| iThome www.ithome.com.tw/news/109602 3 天前 - 在KPMG發布的2016全球FinTech百強排行榜,拿下第一名寶座的,是來自中國阿里巴巴集團的螞蟻金服。從行動網路、雲端運算、大數據以至人工 ...

http://www.ithome.com.tw/news/109602

9 Key Deep Learning Papers, Explained

9 Key Deep Learning Papers, Explained

 - KDnuggets www.kdnuggets.com/2016/.../9-key-deep-learning-papers-explained.ht... 翻譯這個網頁 2016年5月18日 - If you are interested in understanding the current state of deep learning, this post outlines and thoroughly summarizes 9 of the most influential ...

http://www.kdnuggets.com/2016/09/9-key-deep-learning-papers-explained.html

A Visual Introduction to Machine Learning

A Visual Introduction to Machine Learning

http://www.r2d3.us/visual-intro-to-machine-learning-part-1/

Concise Visual Summary of Deep Learning Architectures

Concise Visual Summary of Deep Learning Architectures

Concise Visual Summary of Deep Learning Architectures - Data ... www.datasciencecentral.com/.../concise-visual-summary-of-deep-learni... 翻譯這個網頁 2016年9月21日 - This article was written by Fjodor Van Veen. With new neural network architectures popping up every now and then, it's hard to keep track of ...

http://www.datasciencecentral.com/profiles/blogs/concise-visual-summary-of-deep-learning-architectures

Must Read Books on Machine Learning Artificial Intelligence

Must Read Books on Machine Learning Artificial Intelligence

https://www.analyticsvidhya.com › Business Analytics 翻譯這個網頁 2015年10月25日 - Must Read Books for Beginners on Machine Learning and Artificial Intelligence. Must Read Books for Beginners on Machine Learning and ...

https://www.analyticsvidhya.com/blog/2015/10/read-books-for-beginners-machine-learning-artificial-intelligence/

Deep Learning Frameworks Compared

Deep Learning Frameworks Compared

發佈日期:2016年9月30日 In this video, I compare 5 of the most popular deep learning frameworks (SciKit Learn, TensorFlow, Theano, Keras, and Caffe). We go through the pros and cons of each, as well as some code samples, eventually coming to a definitive conclusion.

https://www.youtube.com/watch?v=MDP9FfsNx60&feature=youtu.be

Understanding Convolutional Neural Networks for NLP

Understanding Convolutional Neural Networks for NLP

– WildML www.wildml.com/.../understanding-convolutional-neural-networks-for-... 翻譯這個網頁 2015年11月7日 - When we hear about Convolutional Neural Network (CNNs), we typically think of Computer Vision. CNNs were responsible for major ...

http://www.wildml.com/2015/11/understanding-convolutional-neural-networks-for-nlp/

Model evaluation, model selection, and algorithm selection in machine learning

Model evaluation, model selection, and algorithm selection in machine learning

Model evaluation, model selection, and algorithm selection in machine ... sebastianraschka.com/blog/2016/model-evaluation-selection-part1.html 翻譯這個網頁 2016年6月11日 - And how do we select a good model in the first place? Maybe a different learning algorithm could be better-suited for the problem at hand?

http://sebastianraschka.com/blog/2016/model-evaluation-selection-part1.html 

http://sebastianraschka.com/blog/2016/model-evaluation-selection-part2.html 

http://sebastianraschka.com/blog/2016/model-evaluation-selection-part3.html

16 Free Machine Learning Books

16 Free Machine Learning Books

https://hackerlists.com/free-machine-learning-books/?utm_campaign=Data%2BElixir&utm_medium=web&utm_source=Data_Elixir_84

An absolute beginner’s guide to machine learning, deep learning, and AI

An absolute beginner’s guide to machine learning, deep learning, and AI

2016/08/20

http://www.datasciencecentral.com/m/blogpost?id=6448529%3ABlogPost%3A460807

The 10 Algorithms Machine Learning Engineers Need to Know

The 10 Algorithms Machine Learning Engineers Need to Know

http://www.iamwire.com/2016/10/the-10-algorithms-machine-learning-engineers-need-to-know/142223

A Short Talk on AI Ethics

A Short Talk on AI Ethics

A Short Talk on AI Ethics—Stephen Wolfram Blog blog.stephenwolfram.com/2016/10/a-short-talk-on-ai-ethics/ 翻譯這個網頁 2016年10月17日 - Video and transcript from Stephen Wolfram's talk on the ethics of artificial intelligence at the NYU Philosophy Department's Center for Mind, ...

http://blog.stephenwolfram.com/2016/10/a-short-talk-on-ai-ethics/

5 algorithms to train a neural network

5 algorithms to train a neural network

5 algorithms to train a neural network | Neural Designer https://www.neuraldesigner.com/.../5_algorithms_to_train_a_neural_net... 翻譯這個網頁 The procedure used to carry out the learning process in a neural network is called the training algorithm. There are many different training algorithms, whith ...

https://www.neuraldesigner.com/blog/5_algorithms_to_train_a_neural_network.html

Deep Learning Papers Reading Roadmap

Deep Learning Papers Reading Roadmap

https://github.com/songrotek/Deep-Learning-Papers-Reading-Roadmap/blob/master/README.md

10 Lessons Learned from Building Deep Learning Systems

10 Lessons Learned from Building Deep Learning Systems

10 Lessons Learned from Building Deep Learning Systems – Intuition ... blog.alluviate.com/?p=233 翻譯這個網頁 2016年10月24日 - Deep Learning is a sub-field of Machine Learning that has its own peculiar ways of doing things. Here are 10 lessons that we've uncovered ...

https://medium.com/intuitionmachine/10-lessons-learned-from-building-deep-learning-systems-d611ab16ef66#.nwx6mdeth

Machine Learning Becomes Mainstream: How to Increase Your Competitive Advantage

Machine Learning Becomes Mainstream: How to Increase Your Competitive Advantage

Machine Learning Becomes Mainstream: How to Increase Your ... https://itpeernetwork.intel.com/machine-learning-becomes-mainstream-i... - 翻譯這個網頁 2016年8月17日 - Machine Learning Becomes Mainstream: How to Increase Your Competitive Advantage. Written by Nidhi Chappell | August 17, 2016.

https://itpeernetwork.intel.com/machine-learning-becomes-mainstream-increase-competitive-advantage/

How AI Is Shaking Up the Chip Market

How AI Is Shaking Up the Chip Market

 | WIRED https://www.wired.com/2016/10/ai-changing-market-computer-chips/ 翻譯這個網頁 2016年10月28日 - How AI Is Shaking Up the Chip Market. Clayton Cotterell for WIRED. In less than 12 hours, three different people offered to pay me if I'd spend ...

https://www.wired.com/2016/10/ai-changing-market-computer-chips/

The role of UX in AI

The role of UX in AI

- Chatbots Life https://chatbotslife.com/the-role-of-ux-in-ai-1b1a234ef6f3?... 翻譯這個網頁 2016年10月31日 - With the interest and demand for Artificial Intelligence on the rise, I often hear the question, “What will the role be of the UX Designer in the ...

https://chatbotslife.com/the-role-of-ux-in-ai-1b1a234ef6f3#.zi5jh2ed3

A Survey of Selected Papers on Deep Learning at ICML 2016

A Survey of Selected Papers on Deep Learning at ICML 2016
By two sigma ON September 26, 2016

https://www.twosigma.com/insights/a-survey-of-selected-papers-on-deep-learning-at-icml-2016

Deep Learning Articles

Deep Learning Articles

2016_Deep Learning on FPGAs, Past, Present, and Future

2016_Towards Bayesian Deep Learning, A Survey

2015_Deep learning in neural networks, An overview

2015_Deep learning

2014_Deep Learning, Methods and Applications

2012_Unsupervised feature learning and deep learning, A review and new perspectives

2009_Learning deep architectures for AI