MixTaiwan 20170222 清大電機 孫民 AI The Next Big Thing

Post on 07-Apr-2017

234 views 0 download

Transcript of MixTaiwan 20170222 清大電機 孫民 AI The Next Big Thing

Artificial Intelligence: The Next Big Thing

from a computer vision perspective

VSLab 清大電機

孫民

AlphaGo

2016 by Google DeepMind

Are these what AI all about?

2014 Subfields of AI

2015

Artifical General Intelligence (AGI)

2016

Deep Learning (DL)

• Data

• GPU Computing

• Talents

Data:

• 開始於 2007 @ Princeton

• 初登場於 2009 @ CVPR

• 照片停止搜集於 2010

總共類別:21841

總共圖片:1千4百萬

• ILSVR Challenge 從2010到現今

Jia Deng Fei-Fei Li

Info from http://www.image-net.org/

1K Image Classification

Figure from Olga Russakovsky ECCV'14 workshop

Deep Learning 深度學習

Label = f(Image)

GPU: NVIDIA CUDA

Tesla P100 With Over 20 TFLOPS Of FP16 Read more: http://wccftech.com/nvidia-pascal-gpu-gtc-2016/#ixzz456KT75Jf

Talents: DNNresearch acquired by Google

Geoffrey Hinton (right: Professor) Alex Krizhevsky (middle; PhD student), and Ilya Sutskever (left; Postdoc)

A story in computer Vision!

DL Fuses AI-subfields • Vision and Language

• Vision and Control

http://mscoco.org/

Atari Breakout game & AlphaGo, DeepMind.

-> AGI

• Multiple Encoding and Decoding

Image Captioning

f( ) = The man at bat is

ready to swing at the pitch

Vision Language

Recurrent Neuron Network (RNN) credit: Nature

convolutions

Convolution Neuron Network (CNN) credit: wiki

Video Captioning/Titling

Zhen et al. ECCV 2016 from VSLab and Stanford AI Lab

Big Video Data with Titles • Pairs of

Raw Video

CNN CNN CNN CNN

Title

Viral Videos

Huge Video Repository

Currently 28740 videos and keep growing

Vision and Control

https://gym.openai.com/

• Learning to play game with weak supervision:

Reinforcement Learning (RL)

Where It All Begins …

by DeepMind in NIPS 2013 Deep Learning Wrokshop

Playing Atari with

Deep Reinforcement Learning

slides by Yen-Chen Lin

Self-driving Car: Trigger Accident Warning

VSLab Under Submission

Fusing Multiple Sensors

Ke# le%

Medium+wrap%

Ke# le%

Medium+wrap%

thumb+4+finger%

Manipula7on%Region%

Side+view%

Chan et al. ECCV 2015 from VSLab

Real-time Wearable Demo

Fisheye camera NVIDIA TK1

Real-time Wearable Demo cellphone, bottle, keyboard, mouse, free hand

Deep Learning (DL)

• Data

• GPU Computing

• Talents

Talents

• Teach as many/early as possible

• Open! Open! Open!

• Critical mass

How to Find Talents

• Our students know deep learning is HOT!

[ 2015 Deep Learning Workshop 中研院 ] 500 位參加者

Teach As Early As Possible

Case Study: YenChen Lin NTHU Undergraduate

https://github.com/yenchenlin1994/DeepLearningFlappyBird

http://www.victoria.ac.nz/design/about/staff/tom-white

Start Doing Research Early!

Case Study: UNIST@Korean Undergraduate

Arxiv

• http://arxiv-sanity.com/

Critical Mass

• Google Brain

• Google Deepmind

• Facebook AI Lab

• Microsoft Research

• Baidu Research

A Team of Talents

Most of them fresh PhDs

1 Billion Pledged USD

A Team of Talents

A Team of Talents

Taiwan Issues

• Critical mass

• Collaboration

• Not open

Taiwan’s Opportunities

• Factory Automation

– Manufacture Data

• Intelligent of Things (IoT)

– Sensors: AI for sensor fusion

• Smart Cities

– Government Open Data (http://index.okfn.org/place/taiwan/)

• Health Care

– Causality

• VR

– Content Generation

Thanks!