Predictive Modeling and Deep Learning

Slide 1

Slide 1 text

Predictive Modeling & Deep Learning Olivier Grisel - ENMI - Paris 2015

Slide 2

Slide 2 text

Outline • Predictive Modeling & Artiﬁcial Intelligence • Deep Learning • Computer Vision • Natural Language Understanding and Machine Translation • Learning to Reason and Answer Questions

Slide 3

Slide 3 text

Predictive Modeling

Slide 4

Slide 4 text

type (category) # rooms (int) surface (ﬂoat m2) public trans (boolean) Apartment 3 50 TRUE House 5 254 FALSE Duplex 4 68 TRUE Apartment 2 32 TRUE sold (ﬂoat k€) 450 430 712 234 features target samples (train)

Slide 5

Slide 5 text

Slide 6

Slide 6 text

Training text docs images sounds transactions Labels Machine Learning Algorithm Model Predictive Modeling Data Flow Feature vectors

Slide 7

Slide 7 text

New text doc image sound transaction Model Expected Label Predictive Modeling Data Flow Feature vector Training text docs images sounds transactions Labels Machine Learning Algorithm Feature vectors

Slide 8

Slide 8 text

Inventory forecasting & trends detection Predictive modeling examples Personalized radios Fraud detection Virality and readers engagement Predictive maintenance Personality matching

Slide 9

Slide 9 text

Artiﬁcial Intelligence Predictive Modeling (Data Analytics)

Slide 10

Slide 10 text

Artiﬁcial Intelligence Predictive Modeling (Data Analytics) Self-driving cars IBM Watson Movie recommendations Predictive Maintenance

Slide 11

Slide 11 text

Artiﬁcial Intelligence Hand-crafted symbolic reasoning systems Predictive Modeling (Data Analytics)

Slide 12

Slide 12 text

Artiﬁcial Intelligence Hand-crafted symbolic reasoning systems Machine Learning Predictive Modeling (Data Analytics)

Slide 13

Slide 13 text

Artiﬁcial Intelligence Hand-crafted symbolic reasoning systems Machine Learning Deep Learning Predictive Modeling (Data Analytics)

Slide 14

Slide 14 text

Artiﬁcial Intelligence Hand-crafted symbolic reasoning systems Machine Learning Deep Learning Predictive Modeling (Data Analytics)

Slide 15

Slide 15 text

Deep Learning • Neural Networks from the 90’s rebranded in 2006+ • « Neuron » is a loose inspiration (not important) • Stacked architecture of modules that compute internal abstract representations from the data • Parameters are tuned from labeled examples

Slide 16

Slide 16 text

Deep Learning for Computer Vision

Slide 17

Slide 17 text

Deep Learning in the 90’s • Yann LeCun invented Convolutional Networks • First NN successfully trained with many layers

Slide 18

Slide 18 text

Early success at OCR

Slide 19

Slide 19 text

Natural image classiﬁcation until 2012 Feature Extractions Classiﬁcation Data independent Supervised Learning dog

Slide 20

Slide 20 text

Natural image classiﬁcation until 2012 Feature Extractions Classiﬁcation Data independent Supervised Learning dog cat

Slide 21

Slide 21 text

Natural image classiﬁcation until 2012 Feature Extractions Classiﬁcation Data independent Supervised Learning cat

Slide 22

Slide 22 text

NN Layer Supervised Learning dog Supervised Learning Supervised Learning NN Layer NN Layer Image classiﬁcation today

Slide 23

Slide 23 text

Image classiﬁcation today NN Layer Supervised Learning Supervised Learning Supervised Learning NN Layer NN Layer dog cat

Slide 24

Slide 24 text

Image classiﬁcation today NN Layer Supervised Learning Supervised Learning Supervised Learning NN Layer NN Layer dog cat

Slide 25

Slide 25 text

Image classiﬁcation today NN Layer Supervised Learning Supervised Learning Supervised Learning NN Layer NN Layer dog cat

Slide 26

Slide 26 text

ImageNet Challenge 2012 • 1.2M images labeled with 1000 object categories • AlexNet from the deep learning team of U. of Toronto wins with 15% error rate vs 26% for the second (traditional CV pipeline)

Slide 27

Slide 27 text

No content

Slide 28

Slide 28 text

ImageNet Challenge 2013 • Clarifai ConvNet model wins at 11% error rate • Many other participants used ConvNets

Slide 29

Slide 29 text

No content

Slide 30

Slide 30 text

ImageNet Challenge 2014 • Monster model: GoogLeNet at 6.7% error rate

Slide 31

Slide 31 text

GoogLeNet vs Andrej • Andrej Karpathy evaluated human performance (himself): ~5% error rate • "It is clear that humans will soon only be able to outperform state of the art image classiﬁcation models by use of signiﬁcant effort, expertise, and time.” source: What I learned from competing against a ConvNet on ImageNet

Slide 32

Slide 32 text

ImageNet Challenge 2015 • Microsoft Research Asia wins with networks with depths ranging from 34 to 152 layers • New record: 3.6% error rate

Slide 33

Slide 33 text

Recurrent Neural Networks

Slide 34

Slide 34 text

source: The Unreasonable Effectiveness of RNNs

Slide 35

Slide 35 text

Applications of RNNs • Natural Language Processing  (e.g. Language Modeling, Sentiment Analysis) • Machine Translation  (e.g. English to French) • Speech recognition: audio to text • Speech synthesis: text to audio • Biological sequence modeling (DNA, Proteins)

Slide 36

Slide 36 text

Language modeling source: The Unreasonable Effectiveness of RNNs

Slide 37

Slide 37 text

Shakespeare source: The Unreasonable Effectiveness of RNNs

Slide 38

Slide 38 text

Linux source code

Slide 39

Slide 39 text

Attentional architectures for Machine Translation

Slide 40

Slide 40 text

Neural MT source: From language modeling to machine translation

Slide 41

Slide 41 text

Attentional Neural MT source: From language modeling to machine translation

Slide 42

Slide 42 text

Attention == Alignment source: Neural MT by Jointly Learning to Align and Translate

Slide 43

Slide 43 text

source: Show, Attend and Tell

Slide 44

Slide 44 text

Learning to answer questions

Slide 45

Slide 45 text

Paraphrases from web news

Slide 46

Slide 46 text

source: Teaching Machines to Read and Comprehend

Slide 47

Slide 47 text

source: Teaching Machines to Read and Comprehend

Slide 48

Slide 48 text

Conclusion • ML and DL progress is fast paced • Many applications already in production (e.g. speech, image indexing, translation, face recognition) • Very promising results for QA and robot control • Machine Learning is now moving from pattern recognition to higher level reasoning

Slide 49

Slide 49 text

No content

Slide 50

Slide 50 text

Thank you! http://twitter.com/ogrisel http://speakerdeck.com/ogrisel TIP: download the PDF version of the slides to click on the source links