Veze, linkovi
Kompjuter biblioteka



Alternativna učenja

Analiza podataka




Apple - MAC OS X





Audio kursevi + knjige

Audio, Multimedia, Video


AutoCad, ArchiCAD, SolidWorks, Catia, Pro/Engineer



Baze podataka


Biološke nauke



C++ Visual C++ C#



Dečije knjige


Digitalna fotografija



Domaće pripovetke

Domaći roman


















Funkcionalno programiranje

Generativna veštačka inteligencija



Git i GitHub



Grafika, Dizajn, Štampa











Istorija i teorija književnosti

Istorija umetnosti

Istorijski roman





Knjiga posle posla - Beletristika i ostala izdanja

Knjižare i naše knjige

Književna kritika

Kuvari, hrana i piće



Ljubavni roman




Mašinsko učenje






Modeliranje podataka



MS Office


Nagrađivanje knjige

Naučna fantastika

Obrada teksta


Operativni sistemi


Organizacione nauke



Pisci u medijima





Popularna medicina

Popularna nauka

Popularna psihologija






Primenjene nauke


Prirodne nauke



Projektovanje softvera




Python programiranje

Računarstvo u oblaku

Raspberry PI






Ruby i Ruby on Rails



Serija Roberta C. Martina



Socijalna mreža - Facebook



Sport i hobi













Unix, Linux



Veštačka inteligencija

Visual Basic .NET, VBA, V. Studio

Web design


Windows 7

Windows 8


Zaštita i sigurnost



Saveti – Mašinsko učenje

Ukupno: 10, strana 1 od 1

18 Machine Learning Platforms For Developers




The following machine learning platforms and tools are available now as resources to seamlessly integrate the power of ML into daily tasks. Machine learning platforms are not the wave of the future. It's happening now. Developers need to know how and when to harness their power. Working within the ML landscape while using the right tools like Filestack can make it easier for developers to create a productive algorithm that taps into its power. The following machine learning platforms and tools — listed in no certain order — are available now as resources to seamlessly integrate the power of ML into daily tasks. 1.  H2O H2O was designed for the Python, R, and Java programming languages by H2O. ai. By using these familiar languages, this open source software makes it easy for developers to apply both predictive analytics and machine learning to a variety of situations. Available on Mac, Windows, and Linux operating systems, H2O provides developers with the tools they need to analyze data sets in the Apache Hadoop file systems as well as those in the cloud. 2.  Apache PredictionIO Developers who are looking for an open-source stack that also has an open-source server for machine learning built on top of it should take a look at Apache PredictionIO as a way to build predictive engines that can meet any artificial intelligence task. In addition to the event server and the platform itself, Apache PredictionIO also includes a template gallery. 3.  Eclipse Deeplearning4j Eclipse Deeplearning4j is an open-source library built for the Java Virtual Machine. With deep learning as its core, this tool is aimed at those developers who need to build deep neural networks within business environments that work on distributed CPUs and GPUs. Scala, Clojure, and Java programmers who work with file systems like Hadoop and who have a DIY bent will appreciate Eclipse Deeplearning4j. Paid support and enterprise distribution are available for this tool, which is a project of the San Francisco-based company Skymind. 4.  Accord. NET Framework Image and audio processing libraries are written in the C# programming language and then combined with the Accord. NET framework. Within it, developers can create a range of apps for commercial use that rely on machine learning such as computer vision, signal processing, pattern recognition, and machine listening, which is also known as computer audition. With multiple options to choose from, developers can utilize image and signal processing, scientific computing, and support libraries. Robust features such as real-time face detection, natural learning algorithms, and more add to the versatility of this framework. 5.  Microsoft During the Ignite conference in September 2017, Microsoft launched three Azure machine learning tools — the Learning Bench, the Learning Model Management service, and the Learning Experimentation service — that allow developers to build their own artificial intelligence models. Three AI tools, Content Moderator, Custom Speech Service, and Bing Speech APIs, were also launched by Microsoft to add to its library of 25 developers' tools that are designed to increase the accessibility of AI. 6.  ai-one Developers can create intelligent assistants that are applicable to nearly any software application by using ai-one. This tool's list of resources includes developer APIs, a document library, and building agents that can be used to turn data into rule sets that support ML and AI structures. 7.  IBM IBM's Watson platform is where both business users and developers can find a range of AI tools. Users of the platform can build virtual agents, cognitive search engines, and chatbots with the use of starter kits, sample code, and other tools that can be accessed via open APIs. 8.  Torch With the Lua programming language as its base, Torch includes a scripting language, a scientific computing framework, and an open-source ML library. Torch supports deep machine learning through an array of algorithms and has been used by DeepMind and the Facebook AI Research Group. 9.  Protege At first blush, it might appear that Protege's focus on enterprises leaves little room for anything else. However, developers can take advantage of Protege's open source tool suite that provides robust app tools for experts and knowledgeable beginners alike. Both groups of developers can modify, create, share, and upload apps as well as take advantage of a supportive community. 10.  TensorFlow Specifically designed for use in projects that rely on machine learning, TensorFlow has the added benefit of being a platform designed using open source software. Aided by a plethora of online resources, documentation, and tutorials, TensorFlow provides a library that contains data flow graphs in the form of numerical computation. The purpose of this approach is that it allows developers to launch frameworks of deep learning across multiple devices including mobile, tablets, and desktops. 11.  DiffBlue DiffBlue is that rather rare developer tool that's an extremely useful yet simple platform dedicated to code automation. DiffBlue has several core purposes — test writing, bug location, refactor code, and the ability to discover and replace weaknesses — that are all accomplished with the use of automation. 12.  Neon The brainchild of Intel and Nervana, Neon is an ML library that is based on Python and is open source to boot. Developers that utilize its tools can take advantage of technologically advanced apps and intelligent agents. Housed within the cloud, Neon supports developers as they launch, build, and train deep learning technologies. 13.  Apache Spark MLlib As a framework that contains in-memory data processing, Apache Spark MLlib features an algorithms database with a focus on clustering, collaborative filtering, classification, and regression. Developers can also find Singa, an open-source framework, that contains a programming tool that can be used across numerous machines and their deep learning networks. 14.  OpenNN A C++ programming library, OpenNN is aimed at those experienced developers who want to implement neural networks. OpenNN includes Neural Designer, a tool that aims to both interpret and simplify data entries with the creation of tables, graphs, and other visual content. Although OpenNN provides its users with an extensive library of tutorials and documentation, it's primarily aimed at those developers who already have lots of AI experience. 15.  Amazon Web Services Developers can take advantage of a number of AI toolkits offered by Amazon Web Services (AWS), which include Amazon Lex, Amazon Rekognition Image, and Amazon Polly. Each is used in a different way by developers to create ML tools. Amazon Polly, for example, takes advantage of AI to automate the process of translating voice to written text. Amazon Lex forms the basis of the brand's chatbots that are used with its personal assistant, Alexa. 16.  Mahout For developers who need to create applications that rely on ML in order to scale, there is Mahout. In addition to resources such as tutorials, Mahout provides beginning developers with the ability to use preconceived algorithms that can then be used with Apache Flink, Apaches Spark, and H2O. 17.  Veles Written in C++ and using Python for node coordination, Veles is Samsung's contribution to the ML landscape. Those developers who already need an API that can be used immediately for data analysis and that is comprised of trained models will find value in Veles. 18.  Caffe Caffe was developed by the Berkeley Vision and Learning Center (BVLC) in collaboration with a developer community. It was designed to provide developers with an automatic inspection tool that is based on images. Caffe is used by some of the biggest brands in the world, including Pinterest and Facebook. Get Started With These Machine Learning Platforms Developers who are just launching their careers, as well as those who are experts, will find a treasure trove of resources as they work their way through the above list. While some are dependent on a specific programming language, others can be used in a variety of instances including in the cloud. Both software and cloud-based offerings allow developers to take advantage of the benefits of each.   https://dzone. com/articles/18-machine-learning-platforms-for-developers

21 korak za dizajniranje mašinskog sistema od nule




Dizajniranje mašinskog sistema od nule uključuje više faza, od definisanja problema do implementacije konačnog modela. U nastavku su navedeni 21 ključni korak koje treba slediti kako bi se kreirao robustan i efikasan mašinski sistem. 1. Definisanje problema Razumevanje poslovnog cilja: Identifikujte problem koji treba rešiti. Da li je to klasifikacija, regresija, klasterisanje ili neki drugi tip problema?Specifikacija ciljeva i ograničenja: Odredite kriterijume uspeha, performanse i sve ograničenja poput vremena, resursa i troškova. 2. Prikupljanje i razumevanje podataka Prikupljanje podataka: Prikupite sirove podatke iz različitih izvora, uključujući baze podataka, API-je ili scraping (sakupljanje podataka sa web stranica). Razumevanje podataka: Sprovođenje istraživačke analize podataka (EDA) za razumevanje distribucije podataka, tipova i preliminarnih uvida. 3. Čišćenje podataka Rukovanje nedostajućim vrednostima: Imputirajte ili uklonite nedostajuće podatke. Uklanjanje izuzetaka: Otkrivanje i rešavanje izuzetaka koji mogu iskriviti rezultate. Ispravljanje grešaka: Popravite bilo kakve neusklađenosti ili greške u podacima. 4. Transformacija podataka Inženjering karakteristika: Kreiranje novih karakteristika koje mogu poboljšati performanse modela. Normalizacija/Standardizacija: Skalirajte karakteristike kako bi doprinosile jednako modelu. 5. Deljenje podataka Trening-Test podela: Podelite podatke na trening i test skupove, obično koristeći odnos 80-20. Kros-validacija: Dalje podelite trening set na manje delove za validaciju performansi modela tokom treninga. 6. Izbor modela Izbor modela: Na osnovu tipa problema, izaberite odgovarajuće algoritme (npr. linearna regresija, decision trees, neuralne mreže). Bazni model: Počnite sa jednostavnim modelom kako biste uspostavili osnovne performanse. 7. Treniranje modela Trening modela: Koristite trening podatke za treniranje modela, podešavajući parametre da biste minimizovali grešku. Podešavanje hiperparametara: Optimizujte hiperparametre koristeći tehnike kao što su grid search ili random search. 8. Evaluacija modela Metodologija performansi: Evaluirajte model koristeći odgovarajuće metrike (npr. tačnost, preciznost, odziv, F1 skor za klasifikaciju; RMSE za regresiju). Validacija: Validirajte model koristeći test set kako biste proverili da li model preterano uči ili nedovoljno uči (overfitting ili underfitting). 9. Interpretacija modela Značaj karakteristika: Identifikujte koje karakteristike su najvažnije za predikcije modela. Vizualizacija: Koristite grafikone i dijagrame za vizualizaciju performansi modela i uvida. 10. Optimizacija modela Iterativno poboljšanje: Na osnovu evaluacije, rafinirajte i ponovo trenirajte model. Podešavanje algoritma: Eksperimentišite sa različitim algoritmima i njihovim postavkama za poboljšanje performansi. 11. Povećanje podataka Sintetički podaci: Kreirajte sintetičke podatke ako je dataset mali da biste poboljšali robustnost modela. Tehnike povećanja: Primena tehnika kao što su rotacija, okretanje ili skaliranje (posebno za slike). 12. Metode ansambla Kombinovanje modela: Koristite tehnike kao što su bagging, boosting ili stacking za kombinovanje više modela radi boljih performansi. Sistemi glasanja: Implementirajte sisteme većinskog glasanja za klasifikacione zadatke. 13. Implementacija modela Priprema za produkciju: Konvertujte model u format spreman za produkciju. Okvir za implementaciju: Koristite okvire kao što su TensorFlow Serving, Flask ili FastAPI za implementaciju. 14. Praćenje modela Praćenje performansi: Kontinuirano pratite performanse modela koristeći metrike i logovanje. Detekcija pomaka: Identifikujte bilo kakav pomak podataka ili degradaciju performansi tokom vremena. 15. Povratna petlja Povratne informacije korisnika: Uključite povratne informacije korisnika za poboljšanje modela. Ponovno treniranje: Periodično ponovo trenirajte model sa novim podacima da biste ga ažurirali. 16. Skalabilnost Horizontalna skalabilnost: Distribuirajte radno opterećenje na više mašina. Cloud usluge: Koristite cloud platforme kao što su AWS, Azure ili GCP za skalabilnu infrastrukturu. 17. Sigurnost i privatnost Sigurnost podataka: Osigurajte da su podaci šifrovani i sigurno pohranjeni. Usklađenost: Pridržavajte se regulativa kao što su GDPR, HIPAA ili CCPA u vezi sa privatnošću podataka. 18. Dokumentacija Dokumentacija koda: Osigurajte da je vaš kod dobro dokumentovan za buduću upotrebu. Dokumentacija modela: Dokumentujte pretpostavke, ograničenja i upotrebu modela. 19. Testiranje Jedinični testovi: Pišite testove za pojedinačne komponente sistema. Integracioni testovi: Testirajte ceo sistem kako biste osigurali da svi delovi rade zajedno glatko. 20. Korisnički interfejs Kontrolna tabla: Kreirajte kontrolne table za netehničke korisnike da interaguju sa modelom. API: Razvijte API-je za druge sisteme da interaguju sa vašim mašinskim modelom. 21. Održavanje Redovna ažuriranja: Održavajte sistem ažuriranim sa najnovijim bibliotekama i okvirima. Popravke grešaka: Brzo rešavajte bilo kakve probleme ili greške koje se pojave u sistemu. Najbolje prakse i saveti: Jasno definišite ciljeve i ograničenja pre nego što započnete bilo koji projekat mašinskog učenja. Koristite istraživačku analizu podataka (EDA) da biste stekli duboko razumevanje podataka. Uvek podelite podatke na trening i test setove kako biste validirali performanse modela. Dokumentujte svaki korak procesa kako bi budući korisnici i developeri mogli lako da razumeju i unaprede sistem. Inovacije: Razmislite o primeni naprednih tehnika kao što su transfer learning ili reinforcement learning za specifične probleme. Istražite mogućnosti automatizacije koraka kao što su podešavanje hiperparametara i izbor modela koristeći AutoML alate.

How to Analyze Tweet Sentiments with PHP Machine Learning




  As of late, it seems everyone and their proverbial grandma is talking about Machine Learning. Your social media feeds are inundated with posts about ML, Python, TensorFlow, Spark, Scala, Go and so on; and if you are anything like me, you might be wondering, what about PHP? Yes, what about Machine Learning and PHP? Fortunately, someone was crazy enough not only to ask that question, but to also develop a generic machine learning library that we can use in our next project. In this post we are going take a look at PHP-ML – a machine learning library for PHP – and we’ll write a sentiment analysis class that we can later reuse for our own chat or tweet bot. The main goals of this post are: Explore the general concepts around Machine learning and Sentiment Analysis Review the capabilities and shortcomings of PHP-ML Define the problem we are going to work on Prove that trying to do Machine learning in PHP isn’t a completely crazy goal (optional) What is Machine Learning? Machine learning is a subset of Artificial Intelligence that focuses on giving “computers the ability to learn without being explicitly programmed”. This is achieved by using generic algorithms that can “learn” from a particular set of data. For example, one common usage of machine learning is classification. Classification algorithms are used to put data into different groups or categories. Some examples of classification applications are: Email spam filters Market segmentation Fraud detection Machine learning is something of an umbrella term that covers many generic algorithms for different tasks, and there are two main algorithm types classified on how they learn – supervised learning and unsupervised learning. Supervised Learning In supervised learning, we train our algorithm using labelled data in the form of an input object (vector) and a desired output value; the algorithm analyzes the training data and produces what is referred to as an inferred function which we can apply to a new, unlabelled dataset. For the remainder of this post we will focus on supervised learning, just because its easier to see and validate the relationship; keep in mind that both algorithms are equally important and interesting; one could argue that unsupervised is more useful because it precludes the labelled data requirements. Unsupervised Learning This type of learning on the other hand works with unlabelled data from the get-go. We don’t know the desired output values of the dataset and we are letting the algorithm draw inferences from datasets; unsupervised learning is especially handy when doing exploratory data analysis to find hidden patterns in the data. PHP-ML Meet PHP-ML, a library that claims to be a fresh approach to Machine Learning in PHP. The library implements algorithms, neural networks, and tools to do data pre-processing, cross validation, and feature extraction. I’ll be the first to admit PHP is an unusual choice for machine learning, as the language’s strengths are not that well suited for Machine Learning applications. That said, not every machine learning application needs to process petabytes of data and do massive calculations – for simple applications, we should be able to get away with using PHP and PHP-ML. The best use case that I can see for this library right now is the implementation of a classifier, be it something like a spam filter or even sentiment analysis. We are going to define a classification problem and build a solution step by step to see how we can use PHP-ML in our projects. The Problem To exemplify the process of implementing PHP-ML and adding some machine learning to our applications, I wanted to find a fun problem to tackle and what better way to showcase a classifier than building a tweet sentiment analysis class. One of the key requirements needed to build successful machine learning projects is a decent starting dataset. Datasets are critical since they will allow us to train our classifier against already classified examples. As there has recently been significant noise in the media around airlines, what better dataset to use than tweets from customers to airlines? Fortunately, a dataset of tweets is already available to us thanks to Kaggle. io. The Twitter US Airline Sentiment database can be downloaded from their site using this link The Solution Let’s begin by taking a look at the dataset we will be working on. The raw dataset has the following columns: tweet_id airline_sentiment airline_sentiment_confidence negativereason negativereason_confidence airline airline_sentiment_gold name negativereason_gold retweet_count text tweet_coord tweet_created tweet_location user_timezone READ ALL

Machine Learning Open Source of the Month (v.Sep 2018)




Machine Learning Open Source of the Month (v. Sep 2018) For the past month, we ranked nearly 250 Machine Learning Open Source Projects to pick the Top 10. We compared projects with new or major release during this period. Mybridge AI ranks projects based on a variety of factors to measure its quality for professionals. Average number of Github stars in this edition: 728 ⭐️ “Watch” Machine Learning Top 10 Open Source on Github and get email once a month. Topics: Research Framework, AutoML library, Deep Learning, PyTorch, TSNE, Algorithm Toolbox, Fairness-ai, Deepdetect, ZOMBIES Open source projects can be useful for programmers. Hope you find an interesting project that inspires you. Rank 1 Dopamine: A research framework for fast prototyping of reinforcement learning algorithms — Google [5316 stars on Github]. Courtesy of Google Rank 2 TransmogrifAI: An AutoML library for building modular, reusable, strongly typed machine learning workflows on Spark with minimal hand tuning [902 stars on Github]. Courtesy of Salesforce Rank 3 Deep-Exemplar-based-Colorization: The source code of “Deep Exemplar-based Colorization”.  [82 stars on Github]. Courtesy of MSRA CVer Rank 4 YOLOv3: Training and inference in PyTorch. A state-of-the-art, real-time object detection system [353 stars on Github]. Courtesy of Ultralytics Rank 5 Mantra: A high-level, rapid development framework for machine learning projects [260 stars on Github]. Courtesy of Ross Taylor Rank 6 FastTSNE: Fast, parallel implementations of tSNE. A visualization of 160,796 single cell trasncriptomes from the mouse nervous system computed in exactly 2 minutes using FFT accelerated interpolation [255 stars on Github]. Courtesy of Pavlin Poličar Rank 7 Evolute: A simple tool for quick experimentation with evolutionary algorithms for numerical optimization.  [111 stars on Github]. Courtesy of Csaba Gór Rank 8 AIF360: An open-source library to help detect and remove bias in machine learning models. This includes metrics for datasets and models to test for biases, explanations for these metrics, and algorithms to mitigate bias in datasets and models.  [159 stars on Github]. Courtesy of IBM Rank 9 DeepSort: AI powered image tagger backed by DeepDetect [84 stars on Github]. Courtesy of Corentin Barreau Rank 10 Zombie-Shooter-Neural-Network: Code that is used in “AI learns to shoot ZOMBIES” [4 stars on Github]. Courtesy of Daporan  

Machine Learning Open Source of the Month May 2018




For the past month, we ranked nearly 250 Machine Learning Open Source Projects to pick the Top 10. We compared projects with new or major release during this period.  Mybridge AI ranks projects based on a variety of factors to measure its quality for professionals. Average number of Github stars in this edition: 979⭐️ “Watch” Machine Learning Top 10 Open Source on Github and get email once a month. Topics: ICLR 2018, Game Research, DeepLearn, Image Translation, Visualization, Text Generation, Self Driving Car, NLP, Video Loader Open source projects can be useful for programmers. Hope you find an interesting project that inspires you. Rank 1 Progressive_growing_of_gans: Progressive Growing of GANs for Improved Quality, Stability, and Variation [2523 stars on Github]. Courtesy of Tero Karras ……[Research Paper] Rank 2 ELF: An Extensive, Lightweight, and Flexible platform for game research. We have used it to build our Go playing bot, ELF OpenGo, which achieved a 14–0 record versus four global top-30 players [1600 stars on Github]. Courtesy of PyTorch ……[Facebook Open Sources ELF OpenGo] Rank 3 DeepLearn: Implementation of research papers on Deep Learning+ NLP+ CV in Python using Keras, Tensorflow and Scikit Learn.  [1277 stars on Github]. Courtesy of Gaurav Bhatt Rank 4 MUNIT: Multimodal Unsupervised Image-to-Image Translation [813 stars on Github]. Courtesy of NVIDIA Research ……[Research Paper] Rank 5 MMdnn: A set of tools to help users inter-operate among different deep learning frameworks. E. g. model conversion and visualization. Convert models between Caffe, Keras, MXNet, Tensorflow [1488 stars on Github]. Courtesy of Microsoft Research Rank 6 Textgenrnn: Python module to easily generate text using a pretrained character-based recurrent neural network.  [620 stars on Github]. Courtesy of Max Woolf Rank 7 Donkey: A modular self driving car library for Python [756 stars on Github]. Courtesy of Will Roscoe Rank 8 TwinGAN: Unpaired Cross-Domain Image Translation with Weight-Sharing GANs [234 stars on Github]. Courtesy of Jerry Li Rank 9 Gluon-nlp: NLP made easy [545 stars on Github]. Courtesy of DMLC Rank 10 Nvvl: A library that uses hardware acceleration to load sequences of video frames to facilitate machine learning training [294 stars on Github]. Courtesy of NVIDIA

Machine Learning Top 10 Articles for May 2017.




Machine Learning Top 10 Articles for May 2017. Between April and May, we’ve ranked nearly 1,900 Machine Learning articles to pick the Top 10 stories that can help advance your career. (0. 5% chance to be picked in the list) Rank 1 Visual Attribute Transfer through Deep Image Analogy. Courtesy of Microsoft Research ………. [ Pdf Paper ] ………. [ Source code on Github ] Rank 2 Why Momentum Really Works. Courtesy of Gabriel Goh Rank 3 Deep Neural Network from scratch. Courtesy of Florian Courtial Rank 4 Big Picture Machine Learning: Classifying Text with Neural Networks and TensorFlow. Courtesy of Déborah Mesquita and freeCodeCamp Rank 5 Phase-Functioned Neural Networks for Character Control. Courtesy of Daniel Holden, Animation Researcher at Ubisoft Montreal ………. [ Paper on Pdf ] ………. [ Video ] Rank 6 A novel approach to neural machine translation. Courtesy of Facebook AI Research ………. [ Sequence-to-Sequence Toolkit, 2,028 stars on Github ] ………. [ Pdf Paper: Convolutional Sequence to Sequence Learning ] Rank 7 Learning Deep Learning with Keras. Courtesy of Piotr Migdał Rank 8 Make your own music with WaveNets: Making a Neural Synthesizer Instrument. Courtesy of Jesse Engel, Researcher at Google Brain Rank 9 Can you improve lung cancer detection? 2nd place solution for the Data Science Bowl 2017. Courtesy of Julian de Wit ………. . [ Official Competition on Kaggle ] Rank 10 OpenAI Baselines: DQN. Reproduce reinforcement learning algorithms with performance on par with published results. ………. [ Open Source Code, 908 stars on Github ] MEDIUM

One model to learh them all




source: https://www. slideshare. net/wearesocialsg/we-are-social-presents-sharing-is-caring/52-ONE_RING_TO_RULE_THEM One model to learn from anything (images, text, speech). I recently stumbled upon this paper by the Google Brain team called “One model to learn them all”. I just had to open it and take a look at the paper because of its great title (best ML paper title ever?). I quickly discovered that it is about a very fascinating and interesting idea. SOURCE WRITTEN BY Sharif Elfouly Follow Machine Learning Engineer @ Vialytics

Šta obuhvata knjiga Algoritmi veštačke inteligencije




Knjiga Algoritmi veštačke inteligencije, iz edicije „Temeljno i intuitivno“ (Grokking), namenjena je programerima softvera i svima u softverskoj industriji koji žele da, kroz praktične primere i vizuelna objašnjenja preko teorijskih dubokih zarona i matematičkih dokaza, otkriju koncepte i algoritme koji stoje iza veštačke inteligencije. Ova knjiga je namenjena svima koji razumeju osnovne koncepte programiranja na računaru, uključujući promenljive, tipove podataka, nizove, uslovne iskaze, iteratore, klase i funkcije (dovoljno je iskustvo u bilo kojem programskom jeziku) i svakome ko razume osnovne matematičke pojmove, kao što su promenljive podataka, predstavljanje funkcija i crtanje podataka i funkcija na grafikonu. Kako je organizovana ova knjiga: smernice Ova knjiga sadrži 10 poglavlja, od kojih se svako fokusira na drugačiji algoritam veštačke inteligencije ili algoritamski pristup. Materijal pokriva osnovne algoritme i koncepte na početku knjige, a oni čine temelj za učenje sofisticiranijih algoritama do kraja knjige. Poglavlje 1, Intuicija veštačke inteligencije, posvećeno je intuiciji i osnovnim konceptima koji obuhvataju podatke, tipove problema, kategorije algoritama i paradigme i slučajeve korišćenja algoritama veštačke inteligencije. U Poglavlju 2, Osnove pretrage, predstavljamo suštinske koncepte struktura podataka i pristupe za primitivne algoritme za pretragu i njihovu upotrebu. U Poglavlju 3, Inteligentno pretraživanje, prevazilazimo primitivne algoritme za pretragu i uvodimo algoritme za pretragu za optimalnije pronalaženje rešenja i pronalaženje rešenja u konkurentnom okruženju. U Poglavlju 4, Evolucioni algoritmi, „zaranjamo“ u rad genetičkih algoritama u kojima se rešenja problema iterativno generišu i poboljšavaju, oponašajući evoluciju u prirodi. Poglavlje 5, Napredni evolutivni pristupi, nastavljamo obradu genetičkih algoritama, započetu u prethodnom poglavlju, ali se bavimo naprednim konceptima koji uključuju kako se koraci u algoritmu mogu prilagoditi za optimalno rešavanje različitih tipova problema. U Poglavlju 6, Inteligencija roja: Mravi, kopamo“ po intuiciji inteligencije roja i razrađujemo kako algoritam za optimizaciju kolonijom mrava koristi teoriju kako mravi žive i kako rešavaju teške probleme. Poglavlje 7, Inteligencija roja: Čestice, nastavljamo obradu algoritama roja, a u isto vreme se koncentrišemo na probleme u vezi sa optimizacijom i kako se rešavaju pomoću optimizacije rojem čestica - jer ona traži dobra rešenja u velikim prostorima za pretragu. Poglavlje 8, Mašinsko učenje, prolazi kroz radni tok procesa mašinskog učenja za pripremu podataka, obradu, modeliranje i testiranje - za rešavanje problema regresije korišćenjem linearne regresije i problema klasifikacije korišćenjem stabala odlučivanja. U Poglavlju 9, Veštačke neuronske mreže, otkrivamo intuiciju, logičke korake i matematičke proračune u treningu i korišćenje veštačke neuronske mreže za pronalaženje obrazaca u podacima i za izradu predviđanja. U Poglavlju 10, Učenja uslovljavanjem pomoću Q-učenja, pokrivamo intuiciju učenja uslovljavanjem iz bihevioralne psihologije i delujemo kroz algoritam Q-učenja da bi agenti naučili da donose dobre odluke u okruženju. Poglavlja treba čitati redom, od početka do kraja. Koncepti i razumevanje se grade usput dok se napreduje kroz poglavlja. Nakon čitanja svakog poglavlja korisno je referencirati se na Python kod u repozitorijumu da bi se eksperimentisalo i stekao praktičan uvid u to kako se može primeniti odgovarajući algoritam. REZERVIŠITE KNJIGU U PRETPLATI, DO 26. 04. 2021. LINK ZA REZERVISANJE

The Essential Artificial Intelligence Glossary for Marketers




Written by Megan Conley @Megan_Conley_ Thank goodness for live chat. If you’re anything like me, you look back at the days of corded phones and 1-800 numbers with anything but fondness. But as you’re chatting with a customer service agent on Facebook Messenger to see if you can change the shipping address on your recent order, sometimes it’s tempting to ask, am I really talking to a human? Or is this kind, speedy agent really just a robot in disguise? Believe it or not, this question is older than you might think. The game of trying to decipher between human and machine goes all the way back to 1950 and a computer scientist named Alan Turing. In his famous paper, Turing proposed a test (now referred to as the Turing Test) to see if a machine’s ability to exhibit intelligent behavior is indistinguishable from that of a human. An interrogator would ask text-based questions to subject A (a computer) and subject B (a person), in hopes of trying to figure out which was which. If the computer successfully fooled the interrogator into thinking it was a human, the computer was said to successfully have artificial intelligence. Since the days of Alan Turing, there’s been decades and decades of debate on if his test really is an accurate method for identifying artificial intelligence. However, the sentiment behind the idea remains: As AI gains traction, will we be able to tell the difference between human and machine? And if AI is already transforming the way we want customer service, how else could it change our jobs as marketers? Why Artificial Intelligence Matters for Marketers As Turing predicted, the concepts behind AI are often hard to grasp, and sometimes even more difficult recognize in our daily lives. By its very nature, AI is designed to flow seamlessly into the tools you already use to make the tasks you do more accurate or efficient. For example, if you’ve enjoyed Netflix movie suggestions or Spotify’s personalized playlists, you’re already encountering AI. In fact, in our recent HubSpot Research Report on the adoption of artificial intelligence, we found that 63% of respondents are already using AI without realizing it. When it comes to marketing, AI is positioned to change nearly every part of marketing -- from our personal productivity to our business’s operations -- over the next few years. Imagine having a to-do list automatically prioritized based on your work habits, or your content personalized based on your target customer writes on social media. These examples are just the beginning of how AI will affect the way marketers work.  No matter how much AI changes our job, we’re not all called to be expert computer scientists. However, it’s still crucial to have a basic understanding how AI works, if only to get a glimpse of the possibilities with this type of technology and to see how it could make you a more efficient, more data-driven marketer. Below we’ll break down the key terms you’ll need to know to be a successful marketer in an AI world. But first, a disclaimer . . . This isn’t meant to be the ultimate resource of artificial intelligence by any means, nor should any 1,500-word blog post. There remains a lot of disagreement around what people consider AI to be and what it’s not. But we do hope these basic definitions will make AI and its related concepts a little easier to grasp and excite you to learn more about the future of marketing. 13 Artificial Intelligence Terms Marketers Need to Know Algorithm An algorithm is a formula that represents the relationship between variables. Social media marketers are likely familiar, as Facebook, Twitter, and Instagram all use algorithms to determine what posts you see in a news feed. SEO marketers focus specifically on search engine algorithmsto get their content ranking on the first page of search results. Even your Netflix home page uses an algorithm to suggest new shows based on past behavior. When you’re talking about artificial intelligence, algorithms are what machine learning programs use to make predictions from the data sets they analyze. For example, if a machine learning program were to analyze the performance of a bunch of Facebook posts, it could create an algorithm to determine which blog titles get the most clicks for future posts. Artificial Intelligence In the most general of terms, artificial intelligence refers to an area of computer science that makes machines do things that would require intelligence if done by a human. This includes tasks such as learning, seeing, talking, socializing, reasoning, or problem solving.  However, it’s not as simple as copying the way the human brain works, neuron by neuron. It’s building flexible computers that can take creative actions that maximize their chances of success to a specific goal. Bots Bots (also known as “chatbots” or “chatterbots”) are text-based programs that humans communicate with to automate specific actions or seek information. Generally, they “live” inside of another messaging application, such as Slack, Facebook Messenger, WhatsApp, or Line. Bots often have a narrow use case because they are programmed to pull from a specific data source, such as a bot that tells you the weather or helps you register to vote. In some cases, they are able to integrate with systems you already use to increase productivity. For example, GrowthBot-- a bot for marketing and sales professionals -- connects with HubSpot, Google Analytics, and more to deliver information on a company’s top-viewed blog post or the PPC keywords a competitor is buying. Some argue that chatbots don’t qualify as AI because they rely heavily on pre-loaded responses or actions and can’t “think” for themselves. However, others see bots’ ability to understand human language as a basic application of AI. Cognitive Science Zoom out from artificial intelligence and you’ve got cognitive science. It’s the interdisciplinary study of the mind and its processes, pulling from the foundations of philosophy, psychology, linguistics, anthropology, and neuroscience.  Artificial intelligence is just one application of cognitive science that looks at how the systems of the mind can be simulated in machines. Computer Vision Computer vision is an application of deep learning (see below) that can “understand” digital images.  For humans, of course, understanding images is one of our more basic functions. You see a ball thrown at you and you catch it. But for a computer to see an image and then describe it makes simulating the way the human eye and brain work together pretty complicated. For example, imagine how a self-driving car would need to recognize and respond to stop lights, pedestrians, and other obstructions to be allowed on the road. However, you don’t have to own a Tesla to experience computer vision. You can put Google’s Quick Draw to the test and see if it recognizes your doodles. Because computer vision uses machine learning that improves over time, you’ll actually help teach the program just by playing. Data Mining Data mining is the process of computers discovering patterns within large data sets. For example, an ecommerce company like Amazon could use data mining to analyze customer data and give product suggestions through the “customers who bought this item also bought” box.   Deep Learning On the far end of the AI spectrum, deep learning is a highly advanced subset of machine learning. It’s unlikely you’ll need to understand the inner workings of deep learning, but know this: Deep learning can find super complex patterns in data sets by using multiple layers of correlations. In the simplest of terms, it does this by mimicking the way neurons are layered in your own brain. That’s why computer scientists refer to this type of machine learning as a “neural network. ” Machine Learning Of all the subdisciplines of AI, some of the most exciting advances have been made within machine learning. In short, machine learning is the ability for a program to absorb huge amounts of data and create predictive algorithms. If you’ve ever heard that AI allows computers to learn over time, you were likely learning about machine learning. Programs with machine learning discover patterns in data sets that help them achieve a goal. As they analyze more data, they adjust their behavior to reach their goal more efficiently. That data could be anything: a marketing software full of email open rates or a database of baseball batting averages. Because machine learning gives computers to learn without being explicitly programmed (like most bots), they are often described as being able to learn like a young child does: by experience. Natural Language Processing Natural language processing (NLS) can make bots a bit more sophisticated by enabling them to understand text or voice commands. For example, when you talk to Siri, she’s transposing your voice into text, conducting the query via a search engine, and responding back in human syntax.  On a basic level, spell check in a Word document or translation services on Google are both examples of NLS. More advanced applications of NLS can learn to pick up on humor or emotion. Semantic Analysis Semantic analysis is, first and foremost, a linguistics term that deals with process of stringing together phrases, clauses, sentences, and paragraphs into coherent writing. But it also refers to building language in the context of culture. So, if a machine that has natural language processing capabilities can also use semantic analysis, that likely means it can understand human language and pick up on the contextual cues needed to understand idioms, metaphors, and other figures of speech. As AI-powered marketing applications advance in areas like content automation, you can imagine the usefulness of semantic analysis to craft blog posts and ebooks that are indistinguishable than that of a content marketer. Supervised Learning Supervised learning is a type of machine learning in which humans input specific data sets and supervise much of the process, hence the name. In supervised learning, the sample data is labeled and the machine learning program is given a clear outcome to work toward. Training Data In machine learning, the training data is the data initially given to the program to “learn” and identify patterns. Afterwards, more test data sets are given to the machine learning program to check the patterns for accuracy.   Unsupervised Learning Unsupervised learning is another type of machine learning that uses very little to no human involvement. The machine learning program is left to find patterns and draw conclusions on its own. ORIGINAL

The Wild Week in AI Issue 70




AI Index 2017: Report Released AIINDEX. ORG – Share The AI Index 2017 Report is a starting point for the conversation about rigorously measuring activity and progress in AI. This report aggregates a diverse set of data, makes that data accessible, and includes discussion about what is provided and what is missing. Facebook rolls out AI feature to detect suicidal posts TECHCRUNCH. COM – Share The new feature will scan all posts for patterns of suicidal thoughts, and when necessary send mental health resources to the user at risk or their friends, or contact local first-responders.  By using AI to flag worrisome posts to human moderators instead of waiting for user reports, Facebook can decrease how long it takes to send help. AWS AI Announcements from re:Invent 2017 AWS. AMAZON. COM – Share AWS announced multiple AI-related services and initiatives at their re:Invent conference, including NLP, video analysis, and model deployment services. It also announced the AWS Machine Learning Research Awards, a new program that funds university departments, faculty, PhD students, and post-docs.   Samsung buys AI company to build out Bixby TECHCRUNCH. COM – Share With the intention of integrating the new technology into its own assistant, Bixby, Samsung acquired the South Korean company Fluenty. The startup offered automated text replies for SMS and third-party social services.   Posts, Articles, Tutorials Sequence Modeling with CTC DISTILL. PUB – Share A visual guide to Connectionist Temporal Classification (CTC), an algorithm used to train deep neural networks in speech recognition, handwriting recognition and other sequence problems.   The impossibility of Intelligence Explosion MEDIUM. COM – Share The concept of an “intelligence explosion” — leading to the sudden rise of “superintelligence” and the accidental end of the human race — has taken hold in the AI community. This post argues that such an event is impossible — that the notion of intelligence explosion comes from a profound misunderstanding of both the nature of intelligence and the behavior of recursively self-augmenting systems.   Specifying AI safety problems in simple environments DEEPMIND. COM – Share As AI systems become more general and more useful in the real world, ensuring they behave safely will become even more important. The post and paper introduce a selection of simple reinforcement learning environments (code here) designed specifically to measure “safe behaviors”.  Read the paper here.   Model-based RL with Neural Network Dynamics BAIR. BERKELEY. EDU – Share The sample inefficiency of deep reinforcement learning methods is one of the main bottlenecks to leveraging learning-based methods in the real world. This post investigates a sample-efficient approach for robot control.  The technique is able to learn locomotion skills of trajectory-following using only minutes of data collected from the robot randomly acting in the environment.   Code, Projects & Data High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs (Stunning Video) TCWANG0509. GITHUB. IO – Share Watch this stunning video. A new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional GANs. The result are stunning 2048x1024 visually results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures.   StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation GITHUB. COM – Share This repository contains the PyTorch implementation of StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation. StarGAN can flexibly translate an input image to any desired target domain using only a single generator and a discriminator.   DeepMind's pycolab Game Engine GITHUB. COM – Share pycolab is a highly-customisable gridworld game engine with some batteries included. It allows you to make your own gridworld games to test reinforcement learning agents. It was used for the recent AI Safety Gridworlds paper.   Deep Image Prior DMITRYULYANOV. GITHUB. IO – Share No pre-trained network or image database needed. A randomly-initialized neural network can be used as a handcrafted prior with excellent results in standard inverse problems such as image denoising, superresolution, and inpainting.  Paper and code are available.   Mozilla’s Speech Recognition Model and Voice Dataset BLOG. MOZILLA. ORG – Share Mozilla’s research team announced the initial release of its open source speech recognition model, as well as the world’s second-largest publicly available voice dataset, which was contributed to by nearly 20,000 people globally.   Highlighted Research Papers [1711. 09534] Neural Text Generation: A Practical Guide ARXIV. ORG – Share In text generation models the decoder can behave in undesired ways, such as by generating truncated or repetitive outputs, outputting bland and generic responses, or in some cases producing ungrammatical gibberish. This paper is intended as a practical guide for resolving such undesired behavior in text generation models, with the aim of helping enable real-world applications.   [1711. 09020] StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation ARXIV. ORG – Share StarGAN is a novel and scalable approach that can perform image-to-image translations for multiple domains using a single model. The unified model architecture of StarGAN allows simultaneous training of multiple datasets with different domains within a single network.   [1711. 09784] Distilling a Neural Network Into a Soft Decision Tree ARXIV. ORG – Share It is hard to explain why a learned network makes a particular classification decision on a particular test case. This is due to their reliance on distributed hierarchical representations.  The authors describe a way of using a trained neural net to create a type of soft decision tree that generalizes better than one learned directly from the training data.   [1711. 10337] Are GANs Created Equal? A Large-Scale Study ARXIV. ORG – Share Despite a very rich research activity leading to numerous interesting GAN algorithms, it is still very hard to assess which algorithm(s) perform better than others. The authors conduct a large-scale empirical study on state-of-the art models and evaluation measures and find that most models can reach similar scores with enough hyperparameter optimization and random restarts.  


Veze, linkovi
Linkedin Twitter Facebook
© Sva prava pridržana, Kompjuter biblioteka, Beograd, Obalskih radnika 4a, Telefon: +381 11 252 0 272