每天推薦一個 GitHub 優質開源項目和一篇精選英文科技或編程文章原文,歡迎關注開源日報。交流QQ群:202790710;電報群 https://t.me/OpeningSourceOrg


今日推薦開源項目:《TensorFlow深度學習模型

推薦理由:關心深度學習的同學,這裡有一串官方整理好的深度學習模型,該存儲庫包含許多在TensorFlow中實現的不同模型,GitHub 地址是 https://github.com/tensorflow/models

官方模型

mnist:對來自MNIST數據集的數字進行分類的基本模型。最開始設計出來的目的是用於識別數字,同時也是深度學習的一個樣例。

 

resnet:一個深度殘差網路,可用於CIFAR-10和ImageNet的1000個類別的數據集進行分類。由於深度學習模型的練習次數達到某一個值時識別準確率以及識別性能會下降,因而開發出了可以提高學習深度的網路。

 

wide_deep:將廣泛的模型和深度網路相結合的模型,用於對人口普查收入數據進行分類。經過學習後,神經網路可以通過其中幾個數據的值推斷出其他數據的值。

 

研究模型(非官方模型,個人使用)

adversarial_crypto:保護與對抗式神經密碼學的通信。

adversarial_text:具有對抗訓練的半監督序列學習。

attention_ocr:圖像識別文本提取模型(用於高干擾的現實場景)。

autoencoder:各種自動編碼器。

brain_coder:帶強化學習的程序綜合器。

cognitive_mapping_and_planning:為視覺導航實現基於空間記憶的映射和規劃體系結構。

compression:使用預先訓練的剩餘GRU網路壓縮和解壓縮圖像。

deeplab:用於語義圖像分割的深度標籤。

delf:用於匹配和檢索圖像的深層局部特徵。

differential_privacy:來自多位教師的學生隱私保護模型。

domain_adaptation:域分離網路。

gan:生成對抗式網路。

im2txt:用於轉換圖像字幕為文本的神經網路。

inception:用於計算機視覺的深度卷積網路。

learning_to_remember_rare_events:用於深度學習的大型終身記憶模塊。

lfads:用於分析神經科學數據的順序變分自動編碼器。

lm_1b:以十億單詞為基準測試的語言建模。

maskgan:用GAN生成文本。

namignizer:識別並生成名稱。

neural_gpu:高度並行的神經計算機。

neural_programmer:用邏輯和數學運算增強的神經網路。

next_frame_prediction:通過交叉卷積網路進行概率性的下一幀合成。

object_detection:定位和識別單個圖像中的多個對象。

pcl_rl:用於幾種強化學習演算法的代碼,包括路徑一致性學習。

ptn:用於三維物體重建的透視變換網。

qa_kg:模塊網路,用於在知識圖上進行問題解答。

real_nvp:使用實值非容量保留(真實NVP)變換的密度估計。

rebar:離散潛變數模型的低方差,無偏差梯度估計。

resnet:深層和廣泛的殘餘網路。

skip_thoughts:遞歸神經網路句 – 矢量編碼器。

slim:TF-Slim中的圖像分類模型。

street:使用深度學習從圖像中識別街道的名稱(僅限於法國)。

swivel:用於生成複合詞的Swivel演算法。

syntaxnet:自然語言語法的神經模型。

tcn:從多視點視頻學習的自我監督表示。

textsum:序列到序列與文本摘要的關注模型。

transformer:空間轉譯網路,可以對網路內的數據進行空間處理。

video_prediction:用神經平流預測未來的視頻幀(類似於next_frame_prediction)。

關於其中的幾個項目:

Mnist實際上是一個簡單的視覺計算數據集,目的大概就是為用機器學習練習對數據進行處理。它本身可能沒有非常有用的一個應用,只是學習機器學習的『陪練』。Mnist主要用來訓練圖像識別相關的機器學習模塊

https://github.com/zalandoresearch/fashion-mnist

這裡有一個很有名也很有趣的mnist數據集fashion-mnist,由60,000個示例的訓練集和10,000個示例的測試集組成。每個示例都是28×28灰度圖像,與10個類別的標籤相關聯。(T恤/上衣,褲子,套頭衫,連衣裙,大衣,涼鞋,襯衫,運動鞋,包,腳踝靴),作者製作這個數據集的本意是用作驗證mnist演算法的基準。

 

wide_deep給的則是用人口普查收入數據預測收入,正如其名字所說的,這是一個深寬模型。也基本是用來驗證tensorflow對深寬模型處理的應用。

 

此外,介紹兩個難度較低的tensorflow的項目:

驗證碼識別:

http://blog.csdn.net/sushiqian/article/details/78305340

驗證碼識別的原理與對mnist手寫數字數據集的處理有異曲同工之妙,也是新手練習的選擇之一

 

五子棋:

https://github.com/junxiaosong/AlphaZero_Gomoku

這個五子棋項目模仿alphago做成的項目,十分有趣,對alphago有興趣的可以了解下


今日推薦英文原文:《Happy 25th birthday Red Hat Linux!》作者: Steven J. Vaughan-Nichols

原文鏈接:https://www.zdnet.com/article/happy-25th-birthday-red-hat-linux/

推薦理由:今天, Linux 在我們身邊隨處可見,不過,25年前呢,可不是這樣,紅帽在開源歷史上有著非常重要的地位,也是迄今為止商業化最成功的開源領域公司,現在,它25歲了。

Happy 25th birthday Red Hat Linux!

oday, Linux and open-source software rule the tech world. Twenty-five years ago? It was an amateur operating system that only geeks knew about. One of the main reasons Linux got from there to here is Red Hat turned a hobby into an IT force.

Red Hat co-founder Bob Young — who had run a rental typewriter business — became interested in Linux. In 1993, he founded ACC Corporation, a catalog business that sold Slackware Linux CDs and open-source software.

Everyone knew, as Young remembered later, “Solaris was much better than Linux, but it was only by using Linux that he could tweak the operating systems to meet their needs.” Young realized that while he couldn’t sell Linux as being better, faster, or having more features than Unix in those days, he could sell one benefit: users could tune it to meet their needs. That would prove to be a key selling point, as it still is today.

So, he joined forces with Linux developer Marc Ewing, and from Young’s wife sewing closet, they launched Red Hat Linux. Like other early Linux businesses, Red Hat started out by selling diskettes, then servers, services, and CDs.

Today, in an interview, Young said, “What I love about the story is that it took many great contributors from the free software/open-source communities including Stallman to Torvalds. To Marc and I and our team-mates to Matthew Szulik, and now Jim and his vast team. None of us could have fundamentally changed the way software is developed and deployed without all the others.”

Young continued, “As my internet software developer son-in-law puts it: he and his colleagues couldn’t do what they do without all the free and open software that Red Hat is both a contributor to and a beneficiary from.” He concluded, “And then there is our families. I would not have been able to make my contribution if my wife Nancy had not been willing to bet our kids’ college education on building a software business on a model never done before.”

First though, Red Hat had to find the magic formula, which would bring it success while so many other of its contemporaries, such as Calera, TurboLinux, and Mandrake, were left in history’s ashbin.

Red Hat’s current CEO Jim Whitehurst told me in an interview, “The real contribution we’ve made, besides open-source software, has been the enterprise business model. It’s obvious now, but it wasn’t obvious at the time.”

I would say so!

In 2003, Paul Cormier, then Red Hat’s vice president of engineering and now Red Hat’s president of Products and Technologies, led the way to leaving behind its early inexpensive distribution, Red Hat Linux, to move to a full business Linux: Red Hat Enterprise Linux (RHEL).

Cormier said later that many “engineers at the time didn’t care about a business model. They wanted to work on Red Hat Linux. We had some level of turmoil inside the company with going to this new model. Some engineers left, but more stayed.”

Many users didn’t like it one darn bit either. They saw Red Hat as abandoning its first customers. Enterprise customers saw it differently.

Whitehurst, who took Red Hat’s reins in 2008, said, “Once RHEL was in the market we had to support it full stop to made it truly consumable for the enterprise.” They did so and the rest is history.

Red Hat grew and grew. In its latest quarter, Red Hat realized $772 million of revenue, which was up 23 percent year over year. Not bad for a company built around an operating system that people back in the day thought of as being only for the lunatic fringe.

Today, Whitehurst, remarked, “Linux is the default choice for open-source companies and enterprises. Ten years ago people still had doubts about open source. Now it’s the default choice for clouds, AI, and big data.” Indeed, “Are there even any important big data or AI projects that aren’t build on open source?” he asked.

The answer, by the by, is no.

It’s not just Red Hat, it’s all of Linux and open-source. “At a Red Hat development site,” Whitehurst said, “an engineer asked me about Microsoft competing with open source.” Whitehurst replied: “Microsoft is not the issue, Windows is a competitor to Linux and we’d love to kill it, but the largest enterprise software company in the world is pro-open source and that’s good for all of us.”

While Red Hat makes the bulk of its money from Linux, Red Hat is no longer just a Linux company. Its eyes are now set on the cloud. Red Hat is determined to use OpenStack to gain a place as big in clouds as the role it already has in Linux.

Red Hat realizes it’s not just the cloud. The company is also heavily invested in containers and container management. Nothing shows that more than its recent acquistion of CoreOS, a leading Kubernetes company.

Linux brought Red Hat where it is today. Moving towards tomorrow it will use open-source software to use the cloud, containers, and container orchestration to rise even further in its next 25 years.


每天推薦一個 GitHub 優質開源項目和一篇精選英文科技或編程文章原文,歡迎關注開源日報。交流QQ群:202790710;電報群 https://t.me/OpeningSourceOrg