Highlight

Những điều thú vị khi dùng Trí tuệ nhân tạo của Viettel

Những người dùng Internet tại Việt Nam thường lấy “chị Google” ra để… giải trí. Khi “chị” đọc văn bản hay chỉ đường cho người tham gia gi...

Tuesday, December 6, 2016

Industrielle Revolution: Wie Industrie 4.0 die Welt verändert

Roboter die jeder mit einem Computer oder Tablet steuern kann. Quelle: Franka

Die Digitalisierung hält Einzug in die Industrie. Wie der Megatrend den Arbeitsmarkt, die Fertigung und die Globalisierung verändert.
Man nennt ihn die vierte industrielle Revolution oder kurz „Industrie 4.0“: Der Megatrend krempelt gerade die Industrie um. Nach Dampfmaschine, Fließband und Elektronik sind diesmal die Vernetzung der Geräte, die Analyse riesiger Datenmengen und der Einsatz von smarten Robotern die wichtigsten Treiber.
So wie Elektrizität vor hundert Jahren in die Fabriken einzog, sind es nun die Datennetze. In den kommenden Jahren werden sie die Art verändern, wie wir Dinge herstellen und konsumieren. Sogar die globale Arbeitsteilung steht auf dem Prüfstand. Warum das so ist und was Experten für die Zukunft der Industrie prognostizieren.

Die Maschinen übernehmen

Ein Gerät, das Dinge fabriziert: Davon träumen Technik-Nerds seit der Replikator in der Science-Fiction-Serie Stark Trek das Konzept populär gemacht hat. Inzwischen sieht es danach aus, dass die Fiktion Schritt für Schritt Realität wird.
Die 3-D-Drucker erobern den Massenmarkt. Bis zum Jahr 2020 wird der Umsatz mit der Technik weltweit auf 35 Milliarden Dollar steigen. Und das dürfte nur der Anfang sein. Denn die Maschinen werden Jahr für Jahr schneller. Und sie können inzwischen dutzende Materialien verarbeiten und kombinieren - Keramik, Kunststoff, ja sogar Holz.
Erstens ermöglicht das die Produktion on demand: Autohersteller wie Daimler sind an der Technik interessiert, weil sie damit unter anderem die Produktion von Ersatzteilen umkrempeln können. Statt jahrelang für jedes Modell Teile teuer zu lagern, können sie nun vor Ort und nach Bedarf herstellen. Sogar Oldtimer lassen sich so mit passenden Teilen renovieren.

Zweitens lassen sich Produkte per 3-D-Druck maßschneidern. Anbieter wie Shapeways drucken Schmuck und vieles mehr auf individuelle Bestellung. Mit spezieller Software kann jeder seine 3-D-Modelle erstellen und an den Druckdienstleister schicken. Das Isala-Krankenhaus in der niederländischen Stadt Zwolle druckt demnächst sogar die Mahlzeiten seiner Patienten - um jedem die perfekte Nährstoffmischung zu bieten.
Drittens macht 3-D-Druck die Produktion hyperlokal. Anstelle großer Fabriken mit Industriefräsen reicht eine Maschine von der Größe einer Kühltruhe, die praktisch überall Platz findet. Allein in New York gibt es mehr als 500 3-D-Druck-Dienstleister. Statt in China bestellen wir also künftig Spielzeug und Co. beim Laden neben dem Chinaimbiss.

Aufmarsch der Roboter

Auch Roboter halten zunehmend Einzug in die Fabriken. Und das ist vielseitiger Form: Als Lieferroboter, die Transportaufgaben übernehmen. Als helfende Hand, die Produkte in Kartons verpackt. Oder als Roboterarm, der Maschinen bestückt und langweilige, schmutzige und gefährliche Jobs übernimmt.
Jede vierte Aufgabe in der Industrie wird im Jahr 2025 von cleveren Maschinen übernommen, so die Prognose der Boston Consulting Group (BCG). Mit vernommen Konsequenzen: Die Arbeitskosten werden in Deutschland um 21 Prozent sinken. Die Industrie steht vor einem enormen Produktivitätsschub.

Deutsche Fabriken werden produktiver

Industrie 4.0, die schlaue Vernetzung der Maschinen, wird einen massiven Produktivitätsschub auslösen. Davon gehen Forscher der DZ Bank aus. Schon bis 2025 wird die deutsche Industrie dank vernetzter Maschinen, Robotern und 3-D-Druck im Schnitt 11,5 Prozent produktiver fabrizieren.
Am meisten profitieren die Chemieindustrie und der Maschinenbau. In diesen Branchen wird der Produktivitätszuwachs laut der Prognose sogar 30 Prozent betragen. Die Autofertigung wird immerhin um 20 Prozent effizienter, die IT-Branche um 15 Prozent. Gut für den Standort - und gut für die Konsumenten, denn Produkte können dadurch preiswerter werden.
Was aber bedeutet das für Arbeitsplätze? Immer öfter warnen Experten vor einer neuen Welle der Arbeitslosigkeit, ausgelöst durch künstliche Intelligenz und Roboter. Die sollen uns zunehmend Aufgaben abnehmen, weil sie im Eiltempo schlauer werden. Bis zu 47 Prozent aller Stellen in den USA seien mittelfristig gefährdet, sagt eine Studie der Universität Oxford voraus.
Für das verarbeitende Gewerbe allerdings sind solcher Horrorszenarien vermutlich verfrüht. In den vergangenen Jahren sind die Stellen in dem Sektor spürbar gestiegen. Der Trend zur Industrie 4.0 wird daran nicht viel ändern. Davon gehen jedenfalls Experten des Instituts für Arbeitsmarkt- und Berufsforschung in Nürnberg aus.

Wo die Maschine den Mensch ersetzt

  • Über die Liste
  • Transportwesen/Personenverkehr
  • Transportwesen/Güterverkehr
  • Finanzwesen
  • Industrie und Produktion
  • Gesundheitswesen
  • Dienstleistung allgemein
  • Landwirtschaft
  • andere Bereiche
Ihre Prognose: Bis 2030 werden zwar 490.000 Stellen in der Industrie wegfallen. Aber dafür werden auch 430.000 neue Stellen geschaffen: Jobs, die nur in Zusammenarbeit mit neuen Robotern möglich sind. Gabelstapler-Fahrer werden zum Beispiel IT-Fachleuten weichen, die in Warenlagern intelligente Flurförderfahrzeuge programmieren.
Netto fallen also nur 60000 Arbeitsplätze weg. Allerdings: Vor allem gering qualifizierte Jobs sind in Gefahr, die neuen Jobs sind vor allem für Fachkräfte. Und die sind in Deutschland heute schon knapp. Bildung und Weiterbildung werden also umso wichtiger.

Fabriken wandern zurück in die Heimat

Made in USA - das ist ein gewaltiger Slogan, der in den USA viele Menschen bewegt. Allein die Handelskette Walmart will bis 2023 Waren im Wert von 250 Milliarden Dollar aus den USA ordern statt bisher aus dem Ausland. Die Hoffnung: Dank Automatisierung wird der Standort wieder wettbewerbsfähig - und Fabriken, die nach Asien abgewandert sind, kehren zurück. Mit ihnen hoffentlich auch die Jobs.
In einer Befragung der Boston Consulting Group unter Konzernen in den USA mit mindesten einer Milliarde Dollar Umsatz gaben 17 Prozent der Befragten Unternehmenschefs an, heute schon Produktion zurück in die Heimat zu verlagern. 53 Prozent gaben an, darüber nachzudenken oder sich vorstellen zu können, diesen Schritt in näher Zukunft zu erwägen.
Forscher des Fraunhofer Instituts für System- und Innovationsforschung in Karlsruhe beobachteten bei einer großen Umfrage unter Industriebetrieben in Deutschland schon einen Trend zur Rückverlagerung. Steigende Löhne sind nur ein Grund von vielen. Viel wichtiger ist, dass sich die Art, wie wir Dinge herstellen, komplett verändern wird. 
Die Fabrik der Zukunft, so das Versprechen gleicht mehr einem Roboter-Park, der Produkte individuell für jeden Kunden fertigen kann - und dabei weniger Energie verbraucht und kaum noch Schmutz erzeugt. Fabriken wandern sogar wieder mitten in die Stadt. 

Monday, December 5, 2016

Intelligence May Stem From a Basic Algorithm in the Human Brain

human brain connectome

The human brain is the most sophisticated organ in the human body. The things that the brain can do, and how it does them, have even inspired a model of artificial intelligence (AI). Now, a recent study published in the journal Frontiers in Systems Neuroscience shows how human intelligence may be a product of a basic algorithm.
This algorithm is found in the Theory of Connectivity, a "relatively simple mathematical logic underlies our complex brain computations," according to researcher and author Joe Tsien, neuroscientist at the Medical College of Georgia at Augusta University, co-director of the Augusta University Brain and Behavior Discovery Institute and Georgia Research Alliance Eminent Scholar in Cognitive and Systems Neurobiology. He first proposed the theory in October 2015.
Basically, it's a theory about how the acquisition of knowledge, as well as our ability to generalize and draw conclusions from them, is a function of billions of neurons assembling and aligning. "We present evidence that the brain may operate on an amazingly simple mathematical logic," Tsien said.

THE BRAIN'S FORMULA

The theory describes how groups of similar neurons form a complexity of cliques to handle basic ideas or information. These groups cluster into functional connectivity motifs (FCM), which handles every possible combinations of ideas. More cliques are involved in more complex thoughts.
In order to test it, Tsien and his team monitored and documented how the algorithm works in seven different brain regions, each involved in handling basics like food and fear in mice and hamsters. The algorithm represented how many cliques are necessary for an FCM, a power-of-two-based permutation logic (N=2i—1), according to the study.
They gave the animals various combinations of four different foods (rodent biscuits, pellets, rice, and milk). Using electrodes placed at specific areas of the brain, they were able to "listen" to the neurons' response. The scientists were able to identify all 15 different combinations of neurons or cliques that responded to the assortment of food combinations, as the Theory of Connectivity would predict. Furthermore, these neural cliques seem prewired in the brain, as they appeared immediately as soon as the food choices did.
If the intelligence in the human brain, in all its complexity, can be summed up by a particular algorithm, imagine what it means for AI. It is possible, then, for the same algorithm to be applied to how AI neural networks work, as these already mimic the brain's structural wiring.
Read the original article on Futurism. Follow Futurism on Facebook. Copyright 2016. Follow Futurism on Twitter.

Chỉ bằng một công thức, lý thuyết này có thể giải thích được trí tuệ của con người


Điều này có thể sẽ trở thành nền tảng giúp trí tuệ nhân tạo mô phỏng não người chính xác hơn nữa.
Bộ não là cơ quan phức tạp nhất trong cơ thể con người. Mọi thứ bộ não có thể làm, cũng như cách nó làm, thậm chí còn trở thành nguồn cảm hứng cho một mô hình trí tuệ nhân tạo (AI) của máy tính. Nhưng giờ, một nghiên cứu mới được công bố gần đây trên tạp chí Frontiers in System Neuroscience (Những biên giới trong các hệ thống về khoa học thần kinh) cho thấy, trí tuệ con người có thể là sản phẩm của một thuật toán cơ bản.
Theo nhà nghiên cứu Joe Tsien, thuật toán này được tìm thấy trong một lý thuyết do ông làm tác giả, Lý thuyết về Kết nối (Theory of Connectivity), một “logic toán học tương đối đơn giản làm cơ sở cho các tính toán phức tạp trong não bộ của chúng ta.” Ông Joe Tsien hiện là nhà thần kinh học tại trường Y Dược Georgia thuộc Đại học Augusta, và là đồng giám đốc của Viện nghiên cứu về Hành vi và Não bộ thuộc Đại học Augusta, là học giả xuất sắc về nhận thức các hệ thống thần kinh thuộc tổ chức phi lợi nhuận Georgia Research Alliance. Ông đưa ra lý thuyết của mình lần đầu tiên vào tháng Mười năm 2015.

Về cơ bản, đó là lý thuyết về việc chúng ta thu nhận kiến thức như thế nào, cũng như khả năng của chúng ta nhằm khái quát và rút ra kết luận từ chúng, một chức năng có được do sự sắp xếp và ghép nối với nhau bởi hàng tỷ các tế bào thần kinh. “Chúng tôi trình bày các bằng chứng cho thấy, bộ não có thể hoạt động theo một logic toán học đơn giản đến tuyệt vời.” Ông Tsien cho biết.
Công thức của não bộ
Lý thuyết này mô tả cách các nhóm tế bào thần kinh tương tự nhau hình thành một cụm (neural clique) phức tạp như thế nào để xử lý các ý tưởng hoặc thông tin cơ bản. Nhóm các cụm tế bào thần kinh này sau đó tập hợp lại thành các motif kết nối chức năng (functional connectivity motifs: FCM), để xử lý mọi khả năng kết hợp có thể xảy ra giữa những ý tưởng. Càng nhiều cụm tham gia vào những nhóm này, các suy nghĩ sẽ càng phức tạp hơn.
Theo nghiên cứu, thuật toán này sẽ biểu thị được số lượng các cụm tế bào thần kinh cần thiết cho một FCM, bằng một công thức logic hoán vị: (N=2i – 1).
Theo giải thích của ông Tsien, N là số lượng các cụm tế bào thần kinh kết nối theo nhiều cách khác nhau, số 2 ở đây có nghĩa là các tế bào thần kinh trong những cụm này có đang nhận thông tin đầu vào hay không, và i là thông tin mà chúng đang nhận vào, còn “– 1” chỉ là một phần của công thức toán học cho phép bạn có thể giải thích cho mọi khả năng.

Để kiểm tra lý thuyết của mình, Tsien và nhóm của ông đã theo dõi và ghi chép lại cách thuật toán này hoạt động trong 7 vùng bộ não khác nhau, mỗi vùng lại liên quan đến việc xử lý những ý tưởng cơ bản như thức ăn, hay nỗi sợ chuột cũng như hamster.
Họ thí nghiệm bằng cách cho những loài động vật các tập hợp khác nhau cảu bốn loại thực phẩm (bánh quy cho loài gặm nhấm, bột viên, gạo và sữa). Sử dụng các điện cực đặt trong những vùng cụ thể của não bộ những loài động vật này, họ có thể “nghe” được phản hồi của tế bào thần kinh.
Vì vậy, các nhà khoa học có thể xác định được tất 15 cách kết hợp khác nhau của các cụm tế bào thần kinh để phản hồi lại các kiểu sắp xếp thực phẩm, đúng như Lý thuyết về Kết nối đã dự đoán. Hơn nữa, các cụm tế bào thần kinh này dường như còn kết hợp sẵn với nhau trong bộ não, vì chúng có thể xuất hiện ngay lập tức mỗi khi những lựa chọn cụ thể về thực phẩm được đưa ra.
Vậy điều này có gì quan trọng? Nếu trí tuệ trong bộ não phức tạp của con người có thể được xác định bằng một thuật toán cụ thể, hãy tưởng tượng điều này sẽ ý nghĩa thế nào với trí tuệ nhân tạo. Vì rất có thể, một thuật toán tương tự như vậy cũng có thể được áp dụng cho cách mạng lưới thần kinh của AI hoạt động, khi chúng đang bắt chước cấu trúc kết nối của bộ não người.
Theo Business Insider

Shazam for furniture has an eye for style

Researchers have developed a deep learning neural network that can analyze photos of furniture and bring...

Researchers have developed a deep learning neural network that can analyze photos of furniture and bring up its make, model and availability(Credit: Cornell University)
Say you're flicking through a lifestyle magazine, or looking at a photo of someone's living room, and you spot a lamp or a chair that would go great in your own home. But how do you find that exact one? A new system, developed by researchers at Cornell University, can run photos of furniture through a deep learning neural network and identify the manufacturer, model and where to buy it.
Googling a vague description of a piece of furniture, like "red leather chair", is likely to return a slew of unwanted results, so Dr. Sean Bell and Prof. Kavita Bala designed a neural network in the vein of Shazam, the app which listens to a snippet of music and in seconds can bring up the song's title, artist, album and other details.
The system developed at Cornell scans photos of objects, and compares their shape, color and features to a database of "iconic images" taken from manufacturer catalogs and websites. From there, it returns the best match, along with details on who makes it and where it's available.
"It seems a lot of people want to buy things they see in someone else's home or in a photo, but they don't know where to look," says Sean Bell, one of the paper's authors.
Online communities like Houzz, Pinterest and LikeThatDecor allow users to share information on products, including where to buy them, but the researchers wanted to automate and streamline that process into a service.
Images of household scenes, with boxes drawn around items of interest, were crowdsourced, and fed into a neural network along with their iconic images to train the system to recognize objects. Rather than waiting for it to trawl through the entire database, the range of matches is narrowed down by first broadly analyzing the image for its edges and lines, to weed out the most obviously wrong answers. Then it looks at more specific parts and shapes, before searching for entire objects in a now much smaller list.
The researchers have formed a startup company, Grokstyle, to get the service up and running on a subscription basis for retailers and designers, and in future, similar systems could be developed for other areas like clothing.

Facebook Is Helping Students to Learn How Artificial Intelligence Works.

Artificial Intelligence at Facebook

Artificial Intelligence is going to be the new future of Technology. It is definitely going to show a huge impact on our digital lives. AI Technology may soon be equal to or greater than that of human brain. It is finding its ways to evolve as Super Intelligence.
Scientists are researching AI for the past 20 to 30 years but suddenly the topic gets its value because, the big names Google, Facebook, Microsoft, Tesla and many more are investing heavily on Artificial Intelligence. They are working hard to design intelligent products that rely on AI. This would definitely transform our digital lives.
These major Tech players are playing a good game in catering the fruits of AI. Whether it be Driver less cars or it be Super smart Digital Assistants like Apple’s Siri, Google’s Google Now and Microsoft’s Cortana or it might be Some super smart chat bots from Facebook.
Your smartphone, house, bank, and car already use AI on a daily basis. Sometimes it’s obvious, like when you ask Siri to get you directions to the nearest gas station, or Facebook suggests a friend for you to tag in an image you posted online.
All these futuristic super smart products are making their foot steps in the process of evolution to make a better future. The best thing about AI is, it’s not as complex as it sounds.
So Facebook wants students to learn How Artificial Intelligence works? In order to help students outline the basics. So the Social Network Company, Facebook has initiated a campaign to understand the benefits of AI and demystify the myths about Artificial Intelligence. The campaign calls it “the backbone of many of the most innovative apps and services of tomorrow.”
As a part of this campaign it launched a series of educational videos that help students to visualize AI in a simple manner. In the videos, Facebook’s Yann LeCun and Joaquin Quiñonero Candela tell that interested students must focus on Calc I, Calc II, Calc III, Linear Algebra, Probability, and Statistics. These videos also explain How Neural Networks work?
Here are these videos by Facebook on basics of AI:
We suggest you to go through these simple explanatory videos by Yann LeCun.

AI Education – Introduction to AI:


AI Education – Gradient Descent:


AI Education – Convolutional Neural Nets:


AI Education – Machine Learning:


Has these videos helped you understand the basics of AI, Tell us your opinions in the comments below. Also ask any questions that are juggling in your mind, our community might may help you to get clear.
Interested in AI? if you want to dig deep and know what Facebook is doing at it’s Facebook Artificial Intelligence Research (FAIR) lab, read the Full Post.

Also Don’t forget to reach us on social media. Stay in touch with us on Facebook, Twitter and Google+.

Deep Learning : What, Why and Applications

Deep Learning

Big companies like Google, Facebook, Intel, IBM, etc. are investing huge on Artificial Intelligence and Machine Learning. Deep Learning (DL) is a specialised type of machine learning. Deep Learning is about learning multiple levels of representation and abstraction that help to make sense of data such as images, sound, and text. People might be excited to learn about, What is DL? Why DL? and What are the different applications of DL? We thought to write an article which answers these questions.

Deep Learning

Deep Learning is based on Deep Neural Network (DNN). DNN is an Artificial Neural Network (ANN) with more number of hidden layers of units between the input and output layers. A simple architecture of Deep Neural Network is given in Fig.1. It is same as Conventional Neural Network but it has more number of hidden layer.
The deep in deep learning refers to having many levels of hidden layers in a neural network. Generally, if there are more than 2 hidden layers (or number of layers is more than 3), it is considered deep learning. Learning is similar to ANN but with more hidden layers. How DNN learn patterns is nicely explained in “Why DL?” section of this article.
DNNs can model complex non-linear relationships in a better way than ANN. Deep Learning with Deep Neural Network Provides best solution to many problems in speech recognition, image recognition and in natural image processing.

Types of Deep Neural Networks

1. Stacked Autoencoder Network

As per the deep neural network (DNN), it consist of multiple hidden layers of units between the input and output layers. This hidden layer working as a Stacked Autoencoders. Autoencoder is a nonlinear feature extraction method without using class labels. This network mostly learns using unsupervised manner. It is trained using greedy layer-wise training procedure.

2. Deep Belief Network

Deep belief network is a probabilistic generative model contains many layer of hidden variable. Each layer captures higher order correlation between the activities of hidden feature present in a layers below in Deep belief network. Mostly, Top two layer in Deep belief network (DBN) are undirected bipartition graph, which is called Restricted Boltzmann Machine. The Lower layer it typically contains a directed sigmoid belief network.

3. Deep Convolutional Network

Convolutional Neural Networks (CNN) is biologically-inspired variants of MLPs. It is a special class of feed forward network that are very suitable for image processing related task. A Convolutional Neural Network (CNN) is comprised of one or more convolutional layers with a sub-sampling layer and then followed by one or more fully connected layers as in a standard multilayer neural network. One more advantage is training of CNN is easier to another network.

Why Deep Learning?

In the past decades, neural network has been used for classification or recognition of object present in an image or video. It was using back-propagation or gradient descent learning for weight adjustment in neural network. But sometimes it is very difficult to learn very complex information in an image or video. So it’s could not give good result for various image processing task.
DL is a new area for machine learning. It uses representation learning for multiple level of representation. It transforms representation at one level (Raw input) into a representation at higher (more abstract) level. Deep learning learns a network level by level. So it can learn more complex function very easily.
For example, an image contain 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.
Deep Learning
Fig.2 – Performance Vs. Volume of Data
The second layer typically detects particular arrangements of edges, regardless of small variations in the edge positions. The third layer may assemble 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.
With recent improvements in GPU technology a learning of network can be done very efficiently in parallel. Therefore, training a deep network is not as time consuming. This is one of the reasons why deep learning is gaining important. Figure-2 shows that performance of deep learning is much better than non-deep learning algorithm. In addition, it is automatically do the feature extraction. Therefore Deep Learning is so much popular in many important applications.

Applications of Deep Learning

Deep learning is impacting everything from healthcare to transportation to manufacturing, and more. Companies are turning to deep learning to solve hard problems, like speech recognition, object recognition, and machine translation. Major application areas are:

Language Modelling

Speech Recognition, Machine Translation, Mobile authentication

Acoustic Modelling

Speech Recognition, Music modeling

NLP Syntactic/Semantic Tagging

Part‐Of-Speech, Chunking, Named Entity Recognition, Semantic Role Labeling, Parsing, Text Understanding

NLP Applications

Sentiment analysis, Paraphrasing, Question-answering, Word‐Sense Disambiguation

Object Recognition

Photo search and image search, handwriting recognition, document analysis, handwriting synthesis, superhuman traffic sign classification, street View house numbers, emotion detection from facial images, Scene Labeling, age estimation, hand segmentation and pose estimation, gesture recognition, Video Classification, action recognition from video, Grasp Detection.

Useful Resources

  1. Deep Learning by Yoshua Bengio, Ian Goodfellow and Aaron Courville
  2. Neural Networks and Deep Learning by Michael Nielsen
  3. Deep Learning Tutorial by LISA lab, University of Montreal
  4. Deep Learning in Neural Networks: An Overview
  5. Theano ( Python CPU / GPU ) mathematical and deep learning library, http://deeplearning.net/soLware/theano
  6. Torch ML Library (C++) http://www.torch.ch
  7. http://www.kdnuggets.com/2015/11/crazy-deep-learning-topological-data-analysis.html