Se ha denunciado esta presentación.
Se está descargando tu SlideShare. ×

Data science e machine learning


Eche un vistazo a continuación

1 de 92 Anuncio

Data science e machine learning

Descargar para leer sin conexión

Una breve introduzione alla data science e al machine learning con un'enfasi sugli scenari applicativi, da quelli tradizionali a quelli più innovativi. La overview copre la definizione di base di data science, una overview del machine learning e esempi su scenari tradizionali, Recommender systems e Social Network Analysis, IoT e Deep Learning

Una breve introduzione alla data science e al machine learning con un'enfasi sugli scenari applicativi, da quelli tradizionali a quelli più innovativi. La overview copre la definizione di base di data science, una overview del machine learning e esempi su scenari tradizionali, Recommender systems e Social Network Analysis, IoT e Deep Learning


Más Contenido Relacionado

Presentaciones para usted (20)

Similares a Data science e machine learning (20)


Más reciente (20)

Data science e machine learning

  1. 1. Data Science e Machine Learning Giuseppe Manco Workshop “Open Data e Data Science” 6 luglio 2017 Distretto Cyber Security di Poste Italiane
  2. 2. Giuseppe Manco • Senior researcher at Institute for high performance computing and networking of the National Research Council of Italy • Head of the Laboratory of Advanced Analytics on Complex Data
  3. 3. Outline • Definizioni di base • Cos’è la data science • Il ruolo del data scientist • Data Science, Technically • Le fasi del processo di analisi • Machine Learning • Scenari applicativi • Applicazioni tradizionali • Recommendation, Social Network Analysis • IoT e Sensor Networks • Challenging Scenarios: Deep Learning • Opportunità • Data science e open data
  4. 4. Definizioni di Base
  5. 5. Premessa: Il ruolo dei Big Data Volume Variety Velocity IDC Report 2001 8 billion in 2015 20 billion in 2020 Relational, graph, Time series, sensors, Audio, video, Text, geo, scientific Storage – Transport - processing Mostly unstructured Facebook 500 tb/day Twitter 300k tweets/min Real time - stream
  6. 6. Big data is here Storing and analyzing it is problematic Good News Bad News
  7. 7. Un po’ di storia… 1962 John Tukey’s “The Future of Data Analysis” 29/07/2016 A Very Short History Of Data Science ­ Forbes 1974 Peter Naur publishes Concise Survey of Computer Methods in Sweden and the United States. The book is a survey of contemporary data processing methods that are used in a wide range of applications. It is organized around the concept of data as defined in the IFIP Guide to Concepts and Terms in Data Processing: “[Data is] a representation of facts or ideas in a formalized manner capable of being communicated or manipulated by some process.“ The Preface to the book tells the reader that a course plan was presented at the IFIP Congress in 1968, titled “Datalogy, the science of data and of data processes and its place in education,“ and “For a long time I thought I was a statistician, interested in inferences from the particular to the general. But as I have watched mathematical statistics evolve, I have had cause to wonder and doubt... I have come to feel that my central interest is in data analysis... Data analysis, and the parts of statistics which adhere to it, must...take on the characteristics of science rather than those of mathematics... data analysis is intrinsically an empirical science... How vital and how important... is the rise of the stored program electronic computer? In many instances the answer may surprise many by being ‘important but not vital,’ although in others there is no doubt but what the computer has been ‘vital.’” A very short History of Data Science – Forbes -
  8. 8. Un po’ di storia… 1974 Peter Naur pubblica “Concise Survey of Computer Methods” In the text of the book, ”the term ‘data science’ has been used freely: “The science of dealing with data, once they have been established, while the relation of the data to what they represent is delegated to other fields and sciences.” A very short History of Data Science – Forbes -
  9. 9. Un po’ di storia… 1977 Viene fondata la International Association for Statistical Computing “It is the mission of the IASC to link traditional statistical methodology, modern computer technology, and the knowledge of domain experts in order to convert data into information and knowledge.” A very short History of Data Science – Forbes -
  10. 10. Un po’ di storia… 1989 Gregory Piatesky-Shapiro organizza il ”First Knowledge Discovery in Databases (KDD) Workshop” Nel 1995 diventa la ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) A very short History of Data Science – Forbes -
  11. 11. Un po’ di storia… 1994 BusinessWeek pubblica una cover story su ”Database Marketing” “Companies are collecting mountains of information about you, crunching it to predict how likely you are to buy a product, and using that knowledge to craft a marketing message precisely calibrated to get you to do so... An earlier flush of enthusiasm prompted by the spread of checkout scanners in the 1980s ended in widespread disappointment: Many companies were too overwhelmed by the sheer quantity of data to do anything useful with the information... Still, many companies believe they have no choice but to brave the databasemarketing frontier.” A very short History of Data Science – Forbes -
  12. 12. Un po’ di storia… 1996 Biennial Conference of the International Federationof Classification Societies (IFCS) the term “data science” is included in the title of the conference A very short History of Data Science – Forbes -
  13. 13. Un po’ di storia… 1996 U. Fayyad, G. Piatetsky-Shapiro, P. Smyth pubblicano “From Data Mining to Knowledge Discovery in Databases” “Historically, the notion of finding useful patterns in data has been given a variety of names, including data mining, knowledge extraction, information discovery, information harvesting, data archeology, and data pattern processing... In our view, Knowledge Discovery in Databases refers to the overall process of discovering useful knowledge from data, and data mining refers to a particular step in this process. Data mining is the application of specific algorithms for extracting patterns from data... the additional steps in the KDD process, such as data preparation, data selection, data cleaning, incorporation of appropriate prior knowledge, and proper interpretation of the results of mining, are essential to ensure that useful knowledge is derived from the data. Blind application of data mining methods (rightly criticized as data dredging in the statistical literature) can be a dangerous activity, easily leading to the discovery of meaningless and invalid patterns.” A very short History of Data Science – Forbes - ■ Data mining and knowledge discovery in databases have been attracting a significant amount of research, industry, and media atten- tion of late. What is all the excitement about? This article provides an overview of this emerging field, clarifying how data mining and knowledge discovery in databases are related both to each other and to related fields, such as machine learning, statistics, and databases. The article mentions particular real-world applications, specific data-mining techniques, challenges in- volved in real-world applications of knowledge discovery, and current and future research direc- tions in the field. A cross a wide variety of fields, data are being collected and accumulated at a dramatic pace. There is an urgent need for a new generation of computational theo- ries and tools to assist humans in extracting useful information (knowledge) from the rapidly growing volumes of digital data. These theories and tools are the subject of the emerging field of knowledge discovery in databases (KDD). At an abstract level, the KDD field is con- cerned with the development of methods and techniques for making sense of data. The basic problem addressed by the KDD process is one of mapping low-level data (which are typically too voluminous to understand and digest easi- ly) into other forms that might be more com- pact (for example, a short report), more ab- stract (for example, a descriptive approximation or model of the process that generated the data), or more useful (for exam- ple, a predictive model for estimating the val- ue of future cases). At the core of the process is the application of specific data-mining meth- ods for pattern discovery and extraction.1 This article begins by discussing the histori- cal context of KDD and data mining and their intersection with other related fields. A brief summary of recent KDD real-world applica- tions is provided. Definitions of KDD and da- ta mining are provided, and the general mul- tistep KDD process is outlined. This multistep process has the application of data-mining al- gorithms as one particular step in the process. The data-mining step is discussed in more de- tail in the context of specific data-mining al- gorithms and their application. Real-world practical application issues are also outlined. Finally, the article enumerates challenges for future research and development and in par- ticular discusses potential opportunities for AI technology in KDD systems. Why Do We Need KDD? The traditional method of turning data into knowledge relies on manual analysis and in- terpretation. For example, in the health-care industry, it is common for specialists to peri- odically analyze current trends and changes in health-care data, say, on a quarterly basis. The specialists then provide a report detailing the analysis to the sponsoring health-care or- ganization; this report becomes the basis for future decision making and planning for health-care management. In a totally differ- ent type of application, planetary geologists sift through remotely sensed images of plan- ets and asteroids, carefully locating and cata- loging such geologic objects of interest as im- pact craters. Be it science, marketing, finance, health care, retail, or any other field, the clas- sical approach to data analysis relies funda- mentally on one or more analysts becoming Articles FALL 1996 37 From Data Mining to Knowledge Discovery in Databases Usama Fayyad, Gregory Piatetsky-Shapiro, and Padhraic Smyth Copyright © 1996, American Association for Artificial Intelligence. All rights reserved. 0738-4602-1996 / $2.00 AI Magazine Volume 17 Number 3 (1996) (© AAAI)
  14. 14. Un po’ di storia… 1997 Professor C. F. Jeff Wu calls for statistics to be renamed data science and statisticians to be renamed data scientists. A very short History of Data Science – Forbes - Lancio del Journal of Data Mining and Knowledge Discovery
  15. 15. Un po’ di storia… 1999 Jacob Zohavi in “Mining Data for Nuggets of Knowledge” “Conventional statistical methods work well with small data sets. Today’s databases, however, can involve millions of rows and scores of columns of data... Scalability is a huge issue in data mining. Another technical challenge is developing models that can do a better job analyzing data, detecting non linear relationships and interaction between elements... Special data mining tools may have to be developed to address website decisions.” A very short History of Data Science – Forbes -
  16. 16. Un po’ di storia… 2001 William S. Cleveland “Data Science: An Action Plan for Expanding the Technical Areas of the Field of Statistics.” “Statisticians should look to computing for knowledge today just as data science looked to mathematics in the past. ... departments of data science should contain faculty members who devote their careers to advances in computing with data and who form partnership with computer scientists.” A very short History of Data Science – Forbes - 2003 Lancio del “Journal of Data Science” “By ‘Data Science’ we mean almost everything that has something to do with data: Collecting, analyzing, modeling...... yet the most important part is its applications–all sorts of applications. This journal is devoted to applications of statistical methods at large....” Leo Breiman’s “Statistical Modeling: The Two Cultures" “There are two cultures in the use of statistical modeling to reach conclusions from data. One assumes that the data are generated by a given stochastic data model. The other uses algorithmic models and treats the data mechanism as unknown. Algorithmic modeling, both in theory and practice, has developed rapidly in fields outside statistics. It can be used both on large complex data sets and as a more accurate and informative alternative to data modeling on smaller data sets.
  17. 17. Un po’ di storia… 2005 T.H. Davenport, D. Cohen, and Al Jacobson “Competing on Analytics” A very short History of Data Science – Forbes - “Long-lived Digital Data Collections: Enabling Research and Education in the 21st Century” Skills, Role & Career Structure of Data Scientists & Curators: Assessment of Current Practice & Future Needs Data scientists as “people who work where the research is carried out–or, in the case of data centre personnel, in close collaboration with the creators of the data–and may be involved in creative enquiry and analysis, enabling others to work with digital data, and developments in data base technology.” 2008 “a new form of competition based on the extensive use of analytics, data, and fact-based decision making... Instead of competing on traditional factors, companies are beginning to employ statistical and quantitative analysis and predictive modeling as primary elements of competition. ” “data scientists as the information and computer scientists, database and software engineers and programmers, disciplinary experts, curators and expert annotators, librarians, archivists, and others, who are crucial to the successful management of a digital data collection.”
  18. 18. Un po’ di storia… 2009 Hal Varian, Google’s Chief Economist “I keep saying the sexy job in the next ten years will be statisticians. The ability to take data—to be able to understand it, to process it, to extract value from it, to visualize it, to communicate it—that’s going to be a hugely important skill in the next decades... Because now we really do have essentially free and ubiquitous data. So the complimentary scarce factor is the ability to understand that data and extract value from it...” A very short History of Data Science – Forbes - Nathan Yau in “Rise of the Data Scientist” “by statisticians, he actually meant it as a general title for someone who is able to extract information from large datasets and then present something of use to non-data experts Fry... argues for an entirely new field that combines the skills and talents from often disjoint areas of expertise... [computer science; mathematics, statistics, and data mining; graphic design; infovis and human-computer interaction].
  19. 19. Un po’ di storia… 2010 Hillary Mason, Chris Wiggins “A taxonomy of Data Science” “...we thought it would be useful to propose one possible taxonomy... of what a data scientist does, in roughly chronological order: Obtain, Scrub, Explore, Model, and iNterpret.... Data science is clearly a blend of the hackers’ arts... statistics and machine learning... and the expertise in mathematics and the domain of the data for the analysis to be interpretable... It requires creative decisions and open- mindedness in a scientific context.” A very short History of Data Science – Forbes -
  20. 20. Un po’ di storia… 2011 Drew Conway’s “The Data Science Venn Diagram” “ needs to learn a lot as they aspire to become a fully competent data scientist. Unfortunately, simply enumerating texts and tutorials does not untangle the knots. Therefore, in an effort to simplify the discussion, and add my own thoughts to what is already a crowded market of ideas, I present the Data Science Venn Diagram... hacking skills, math and stats knowledge, and substantive expertise.” A very short History of Data Science – Forbes - David Smith’s “The Data Science: What’s in a name?” “The terms ‘Data Science’ and ‘Data Scientist’ have only been in common usage for a little over a year, but they’ve really taken off since then. But despite the widespread adoption, some have resisted the change from the more traditional terms like ‘statistician’ or ‘quant’ or ‘data analyst’.... I think ‘Data Science’ better describes what we actually do: a combination of computer hacking, data analysis, and problem solving.”
  21. 21. Un po’ di storia… 2012 Tom Davenport and D.J. Patil’s “Data Scientist: The Sexiest Job of the 21st Century” A very short History of Data Science – Forbes -
  22. 22. Cos’è un data scientist?
  23. 23. Cos’è un data scientist?
  24. 24. Cos’è un data scientist?
  25. 25. Cos’è un data scientist?
  26. 26. You’re not a data scientist
  27. 27. Skills e competenze • Problem solver • Creativo, collaborativo, proattivo • Matematica e statistica • Machine Learning, Statistical modeling • Experimental design • Optimization • Computer Science • Statistical Computing, artificial intelligence • Databases • Distributed computing • Coding skills • Communication • Storytelling • Conoscenza di tools di visualizzazione • Tradurre data-driven insights in azioni
  28. 28. Data Scientist, Data Analyst Data Scientist Familiar with databases Analytical functions and how to use them on data: median, rank, etc. Perfection in mathematics, statistics, correlation, predictive analytics Deep statistical insights and machine learning: Bayesian modeling, clustering, etc. Data Analyst Familiar with data warehousing and business intelligence concepts Strong understanding of Big Data tools (Hbase, Hive, Mapreduce, etc) Skill on data storind and retrieving tools Familiar with ETL tools
  29. 29. Cosa fa un data scientist? Scrive algoritmi super-sofisticati che hanno un impatto notevole sul business
  30. 30. Cosa fa un data scientist? Scrive algoritmi super-sofisticati che hanno un impatto notevole sul business
  31. 31. Cosa fa un data scientist? Scrive algoritmi super-sofisticati che hanno un impatto notevole sul business Spende l’80% del suo tempo a: • Raccogliere i dati • Studiare i dati e capirne la semantica • Fare assessment della qualità e la validità dei dati • Pulire i dati • Trasformare i dati facendo feature engineering e feature modeling Il restante 20% è dedicato agli algoritmi di analisi
  32. 32. Data Science, Technically
  33. 33. Data Science e Data-driven decision making Figure 1-1. Data science in the context of various data-related processes in the organization. F. Provost, T. Fawcett, Data Science for Business, O’Reilly, 2013
  34. 34. The Data Science Workflow
  35. 35. The Data Science Workflow Validate Deploy Explore Understand Model Communicate Results Clean and Transform Collect
  36. 36. Il ruolo del Machine Learning • Un campo dell’intelligenza artificiale nel quale le macchine imparano senza programmazione esplicita Training Data Learning Algorithm Data Predictions Learned Model Training Prediction
  37. 37. Machine Learning, Formalmente • Dati • Un insieme ! = #$, … , #' di (proprietà di) entità • (opzionale) Un insieme di responsi / = 0(#$), … , 0(#') • ! ammette una funzione 0 che descrive una proprietà delle entità • Problema • Formulare una ipotesi ℎ relativa alla funzione 0 • ℎ #4 ≡ 0 #4 per ogni entità #4 • ℎ rappresenta il modello appreso • Approccio • Algoritmo di apprendimento sull’insieme di training
  38. 38. Machine Learning • Supervised Learning • Apprendimento dal un insieme di training etichettato • Esempio: email spam detector • Unsupervised Learning • Identificazione di patterns su dati non etichettati • Esempio: raggruppare documenti simili in base al contenuto • Reinforcement Learning • Apprendimento sulla base di feedback • Esempio: Imparare a giocare a scacchi tramite trial & error
  39. 39. Tipologie di problemi • Prediction • Regression, classification • Outlier detection • Description • Association rules • clustering analysis x1 x2 f(x) x if age > 35 and income < $35k then ... Regression (supervised – predictive) Classification (supervised – predictive) Anomaly Detection (unsupervised– descriptive) Clustering (unsupervised – descriptive) Machine Learning - Basics Problem Types Regression (supervised – predictive) Classification (supervised – predictive) Anomaly Detection (unsupervised– descriptive) Clustering (unsupervised – descriptive)
  40. 40. Esempio: Boston housing • 506 case campionate da vari quartieri intorno a Boston • Descritte tramite un insieme di caratteristiche • Obiettivo • Predire il valore di MEDV sulla base degli altri parametri
  41. 41. Esempio: Boston housing • Obiettivo • Predire il valore di MEDV sulla base degli altri parametri • Dati
  42. 42. Esempio: Boston housing • Obiettivo • Predire il valore di MEDV sulla base degli altri parametri • Modello • MEDV = 12,2*log(LSTAT) + 51,78 #scatterplot of lstat vs. medv plot(log(lstat), medv, xlab = "Log Transform of % of Houshold with Low Socioeconomic Income", ylab = "Median House Value", col = "red", pch = 20) # Make the line color blue, and the line s width =3 (play with the width!) abline(model, col = "blue", lwd =3) 0.5 1.0 1.5 2.0 2.5 3.0 3.5 1020304050 Log Transform of % of Houshold with Low Socioeconomic Income MedianHouseValue
  43. 43. Esempio: Iris Data • 150 fiori misurati per lunghezza e ampiezza del petalo e del sepalo • Obiettivo: • Quali sono le regolarità che possiamo osservare sui fiori?
  44. 44. Esempio: Iris Data • Obiettivo • Quali sono le regolarità che possiamo osservare sui fiori? • Dati
  45. 45. Esempio: Iris Data • Obiettivo • Quali sono le regolarità che possiamo osservare sui fiori? • Modello • Clustering dei campioni
  46. 46. Quali algoritmi per quali problemi?
  47. 47. Perché il machine learning si sta affermando? • Abbondanza di dati (especially with the advent of the Internet) • Potere computazionale • Teoria e algoritmi in continuo sviluppo negli ultimi anni • Supporto da parte delle industrie
  48. 48. Gli scenari applicativi
  49. 49. Svariati scenari INPUT RESPONSE APPLICATION Loan application Will they repay the loan? Loan approval Patient data Will the patient be responsive to treatment X? Medicine Ad + user information Will user click on ad? Targeted online ads Audio clip transcript Speech recognition Sensor data Is it about to fail? Preventive maintenance Car camera and other sensors Position of other cars Self-driving cars
  50. 50. Fraud detection • Insurance subrogation • Evasione fiscale
  51. 51. Intrusion detection • Cyber Security
  52. 52. Churn Analysis • Identificare/segnalare possibili utenti churning • Comprendere le ragioni dell’abbandono Campo State Account length Area code Phone number International Plan VoiceMail Plan Number of voice mail messages Number of calls to customer service Total day minutes Total day calls Total day charge …
  53. 53. • Autonomous and self- aware Energy Management Systems based on Intelligent Data Analysis of time series Energy management
  54. 54. Text Analysis: Entity Recognition
  55. 55. Recommender Systems Highlights (I)
  56. 56. Recommender Systems
  57. 57. Preference data
  58. 58. Recommender systems, formalmente • Problema di predizione • Data la cella (u,i), predire il valore associato • Usare la predizione per generare una lista di preferenze • Campi applicativi • (user, movie) (collaborative filtering) • (user, user) (link prediction) • … (item response, political science)
  59. 59. Social Network Analysis Highlights (II)
  60. 60. Social Networks • Gli utenti condividono e collaborano sui contenuti, esprimono opinioni, costruiscono legami • Perché è rilevante: • Human behavior • Marketing analytics • Product sentiment Any user can share and contribute content, express opinions, link to others This means: Can data-mine opinions and behaviors of millions of users to gain insights into: Human behavior Marketing analytics Product sentiment Jure Leskovec:Social Media Analytics (KDD '11 tutorial) 68/21/2011 6
  61. 61. Actionable Intelligence Consumer Generated, Not Edited, Not Authenticated 8/21/2011 Jure Leskovec:Social Media Analytics (KDD '11 tutorial) 7
  62. 62. Applicazioni: Reputation management • Consumer Brand Analytics • Che opinione c’è sul mio brand? • Marketing Communications • Qual’è l’effetto della campagna di marketing • Product reviews • Come viene percepito il mio prodotto? • Facile da usare, comodo, prezzo troppo alto, …
  63. 63. Applicazioni: Responsiveness • Feedbacks su temi politici • Campagne • Perché un candidato viene votato? • Law enforcement • Minority report
  64. 64. Applicazioni: Viral Marketing • Viral marketing: • Raccomandazioni personalizzate • Il ruolo dei forum online: • Il 79.2% dei membri dei forum aiuta altri utenti a prendere decisioni sui prodotti da acquistare • Il 65% degli utenti condividono opinioni che sviluppano sui forum tematici
  65. 65. IoT e Sensor Networks Highlights (III)
  66. 66. Internet of Things • Internet of Things refers to the concept that the Internet is no longer a global network for people to communicate with one another using computers , but it is also a platform for devices to communicate electronically with the world around them
  67. 67.
  68. 68.
  69. 69. Machine Learning & IoTs (1)
  70. 70. Machine Learning & IoTs (2)
  71. 71. Un esempio: IoT & Predictive Maintenance • Predizione dei guasti alle porte nei treni della metropolitana di Londra data sources recording historical diagnostic data. In Section 3 we provide the methodology that has inspired our work. First, data abstraction and problem formulation are given. Then, data pre-processing is described. Fi- nally, both fault prediction and fault explanation techniques are provided. In Section 4 the experimental evaluation work is presented. In Section 5 a description of the architecture supporting the described methodology that was set up for the purposes of the project is given. Finally, in Section 7 we draw our conclusions. 2. Scenario: Door Train Failures Operational Analytics Per Trigger train maintenance team may decide to raise an incident Sensors in trains continuously create Events and Environmental data (measures) Teams that operate and maintain trains indentify new problems from time to time The operational analytics tools continuously watch for the occurrence of patterns (defined in Alerts) Patterns are codified as automated Alerts, and deployed on the Operational Analytics tool Analytical team try detect patters in historical data that ca be used to identify or predict failures that caused the problem Alerts are triggered automatically when the pattern is observed in the actual event /measure data sets Problems trigger analytical teams to conductroot cause analysis Alert Triggers Events Measures Alert Pattern Problem Investigational Analytics Develop Alerts Data created continuously Watch for pattern Investigate and validate for new patterns Act on Alert Trigger Train Maintenance Train in Operation 1 2 3 4 56 7 8 Figure 1: Overview of the Overall Predictive Maintenance Process.
  72. 72. Deep Learning Highlights (IV)
  73. 73. Deep Learning • Idea: Imitare il processo di apprendimento delle cellule neuronali • Utilizzando diversi livelli di astrazione
  74. 74. Dal feature engineering al feature learning Input Data Feature engineering Traditional Learning algorithm Input Data Deep Learning algorithm
  75. 75. Y LeCun MA Ranzato The Mammalian Visual Cortex is Hierarchical [picture from Simon Thorpe] The ventral (recognition) pathway in the visual cortex has multiple stages Retina - LGN - V1 - V2 - V4 - PIT - AIT .... Lots of intermediate representations • Il processo di riconoscimento visuale lungo la corteccia è fatto di stadi multipli • Retina - LGN - V1 - V2 - V4 - PIT - AIT .... • Molte rappresentazioni intermedie
  76. 76. Astrazione • Hierarchical Learning • Una progressione naturale verso livelli di astrazione più generali Esempio: Image Recognition 3o livello “Oggetti” 2o livello “parti di oggetti” 1o livello “contorni” Pixels (Raw data)
  77. 77. Deep Learning - Basics Architecture A deep neural network consists of a hierarchy of layers, whereby each layer transforms the input data into more abstract representations (e.g. edge -> nose -> face). The output layer combines those features to make predictions.
  78. 78.
  79. 79. Colorization
  80. 80. Object Recognition
  81. 81. Presidential campaign 2016 Mit Technology Review, march 2016
  82. 82. Self-driving cars End to End Learning for Self-Driving Cars Mariusz Bojarski NVIDIA Corporation Holmdel, NJ 07735 Davide Del Testa NVIDIA Corporation Holmdel, NJ 07735 Daniel Dworakowski NVIDIA Corporation Holmdel, NJ 07735 Bernhard Firner NVIDIA Corporation Holmdel, NJ 07735 Beat Flepp NVIDIA Corporation Holmdel, NJ 07735 Prasoon Goyal NVIDIA Corporation Holmdel, NJ 07735 Lawrence D. Jackel NVIDIA Corporation Holmdel, NJ 07735 Mathew Monfort NVIDIA Corporation Holmdel, NJ 07735 Urs Muller NVIDIA Corporation Holmdel, NJ 07735 Jiakai Zhang NVIDIA Corporation Holmdel, NJ 07735 Xin Zhang NVIDIA Corporation Holmdel, NJ 07735 Jake Zhao NVIDIA Corporation Holmdel, NJ 07735 Karol Zieba NVIDIA Corporation Holmdel, NJ 07735 Abstract We trained a convolutional neural network (CNN) to map raw pixels from a sin- gle front-facing camera directly to steering commands. This end-to-end approach proved surprisingly powerful. With minimum training data from humans the sys- tem learns to drive in traffic on local roads with or without lane markings and on highways. It also operates in areas with unclear visual guidance such as in parking lots and on unpaved roads. The system automatically learns internal representations of the necessary process- ing steps such as detecting useful road features with only the human steering angle as the training signal. We never explicitly trained it to detect, for example, the out- line of roads. Compared to explicit decomposition of the problem, such as lane marking detec- tion, path planning, and control, our end-to-end system optimizes all processing steps simultaneously. We argue that this will eventually lead to better perfor- mance and smaller systems. Better performance will result because the internal components self-optimize to maximize overall system performance, instead of op- timizing human-selected intermediate criteria, e. g., lane detection. Such criteria understandably are selected for ease of human interpretation which doesn’t auto- matically guarantee maximum system performance. Smaller networks are possi- ble because the system learns to solve the problem with the minimal number of processing steps. We used an NVIDIA DevBox and Torch 7 for training and an NVIDIA DRIVETM PX self-driving car computer also running Torch 7 for determining where to drive. The system operates at 30 frames per second (FPS). arXiv:1604.07316v1[cs.CV]25Apr2016
  83. 83. Molto di più… • Speech Recognition • Automatic translation • Chatbots • AlphaGo • …
  84. 84. Le opportunità
  85. 85. Open Data
  86. 86. Dove posso prendere i dati?
  87. 87. Cosa posso fare con gli open data?
  88. 88. learning-to-open-data-sets-to-find-new-customers
  89. 89. Conclusioni • Data Science • “sexiest job of 21st century” • Insights sui processi che governano i fenomeni, sulla base delle tracce che tali fenomeni lasciano nei dati • Un ampio spettro di applicazioni • Il ruolo del machine learning • Nuovi modelli • Nuovo hardware • Il ruolo degli open data