Slide 1

Size: px

Start display at page:

Download "Slide 1"

湫毕
7 years ago
Views:

1 13PRESENTATION BY TOMASO POGGIO AT THE FORUM: INTELLIGENCE ON THE WORLD, EUROPE, AND ITALY Villa d'este, Cernobbio - September 2, 3 and 4, 2011 Reproduced by The European House-Ambrosetti for internal use only.

2 13~ 2010 ~ Villa deste, 2011 Today the World of Tomorrow C Scientific Developments tomaso poggio, McGovern Institute BCS, CSAIL MIT

Institute BCS, CSAIL MIT 1717171717171717171717171717171717171717171717171717171717

1717171717171717171717171717171717171717 171717171717171717171717171717171717171717171717 171717171717 1717171717171717 171717171717171717171717171717171717 171717171717171717 171717171717

5 Since the introduction of learning techniques in AIvision, there have been significant (and not well known) advances in a few domains: 61 Games (Chess, Go...) 61 Search (Google,...) 61 Speech recognition (Nuance, Google,...) 61 Natural Language/Knowledge retrieval (Watson and Jeopardy) 61 Vision (see later) Poggio, T. and F. Girosi. Networks for Approximation and Learning, Proceedings of the IEEE 1990) also Science, 1990 Poggio, T. and S.Smale , Notices American Mathematical Society (AMS), 2003 Poggio, T., R. Rifkin, S. Mukherjee and P. Niyogi. General Conditions for Predictivity in Learning Theory, Nature, 2004 Mathematics ENGINEERING APPLICATIONS Beymer, D. and T. Poggio , Science, 272., , 1996 Brunelli, R. and T. Poggio. Face Recognition: Features Versus Templates, IEEE PAMI, Engineering Ezzat, T., G. Geiger and T. Poggio. Trainable Videorealistic Speech Animation, ACM SIGGRAPH 2002 modelsexperiments Freedman, D.J., M. Riesenhuber, T. Poggio and E.K. Miller , Science, 291, , Riesenhuber, M. and T. Poggio , Nature Neuroscience, Science 2, , Serre, T., A. Oliva and T. Poggio , (PNAS), Vol. 104, No. 15, , Poggio, T. and E. Bizzi , Nature, Vol. 431, , 2004.

..) 61 Natural Language/Knowledge retrieval (Watson and Jeopardy) 61 Vision (see later) 1717171717171717171717171717171717171717171717171717171717171717 Poggio, T. and F. Girosi.

6 Sung & Poggio 1995, also Kanade& Baluja... modelsexperiments Sung & Poggio 1995 modelsexperiments

7 Sung & Poggio 1995 modelsexperiments Face detection is now available in digital cameras (commercial systems) modelsexperiments

8 Papageorgiou&Poggio, 1997, 2000 also Kanade&Scheiderman modelsexperiments Papageorgiou&Poggio, 1997, 2000 also Kanade&Scheiderman modelsexperiments

9 Papageorgiou&Poggio, 1997, 2000 also Kanade&Scheiderman modelsexperiments Pedestrian and car detection are also solved (commercial systems, MobilEye) modelsexperiments

10 Since the introduction of learning/statistics in the 90s, computer vision has made significant (and not well known) advances in a few problem areas: 61 Face identification under controlled conditions is solved (commercial systems) at ~ human level performance 61 Face detection is available in cheap digital cameras (commercial systems) 61 Pedestrian and car detection are also solved (commercial systems)

learning/statistics in the 90s, computer vision has made significant (and not well known) advances in a few problem areas: 61 Face identification under controlled conditions

11 Each of the systems above can pass its narrow version of the Turing test. However, the full problem of vision (and intelligence!) is still not solved. I bet that we may need to understand visual cortex (and the brain!) to achieve scene understanding at human level, and develop systems that pass a broad Turing test Learning: Math, Engineering, Neuroscience (now) Models of cortex lead to better systems for action recognition in videos: automatic phenotyping of mice Serre* Jhuang* Garrote Poggio Steele, 2009

I bet that we may need to understand visual cortex (and the brain!

12 13Learning: Math, Engineering, Neuroscience (now) Models of cortex lead to better systems for action recognition in videos: automatic phenotyping of mice Serre* Jhuang* Garrote Poggio Steele, Performance Models of cortex lead to better systems for action recognition in videos: automatic phenotyping of mice human agreement 72% proposed system 77% commercial system 61% chance 12%

cortex lead to better systems for action recognition in videos: automatic phenotyping of mice human agreement 72% proposed system 77% commercial system 61% chance 12%

13 Performance Models of cortex lead to better systems for action recognition in videos: automatic phenotyping of mice human agreement 72% proposed system 77% commercial system 61% chance 12%

14 Most areas of business and commerce in the Western world are loosing jobs...but, not surprisingly, designers of computers and robots are in high demand (startups, engineers, scientists, MIT graduates...) Apple is now the largest company in the world (> Exxon); Google is a machine learning company; top Universities are one of the very few booming enterprises in the US Countries which invest in research are doing well: Switzerland, Nordic countries, Israel, Singapore, South Korea, Austria, Germany... Italy is at the bottom (1% vs 3% or more)

..) Apple is now the largest company in the world (> Exxon); Google is a machine learning company; top Universities are one of the very few booming enterprises in the US Countries which invest in

15 The first and last attempt in history was ~ fifty years ago -- at the beginning of Artificial Intelligence (Dartmouth workshop,1956) and then ML 61 It is time to try again because C explosive progress in the neurosciences C several intelligent apps from the first wave of MLAI: search vision speech language Go... C...but these are not intelligent systems and we do not yet understand intelligence... C...we need to integrate computation with neuroscience and cognition for the next jump in understanding 61 Thus we need a new effort in basic research:

家长讲师团案例之一

家长讲师团案例之一书香的气息哲思的气韵高远的气质主编 : 费建华题字 : 朱永新 2011-2012 学年度第一学期第 5 期 ( 总 181 期 ) 教师论文简论中学历史教学中的学生自学历史组周昆内容摘要 : 众所周知, 论从史出只有掌握了一定的历史信息, 才能分析