Deep neural networks have revolutionized the data analytics scene by improving results in several and diverse benchmarks with the same recipe: learning feature representations from data. These achievements have raised the interest across multiple scientific fields, especially in those where large amounts of data and computation are available. This change of paradigm in data analytics has several ethical and economic implications that are driving large investments, political debates and sounding press coverage under the generic label of artificial intelligence (AI). This talk will present the fundamentals of deep learning through the classic example of image classification, and point at how the same principal has been adopted for several tasks. Finally, some of the forthcoming potentials and risks for AI will be pointed.
Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020
1. Deep Learning
Representations for All
(a.k.a. The AI Hype)
Xavier Giro-i-Nieto
@DocXavi
xavier.giro@upc.edu
Associate Professor
Universitat Politècnica de Catalunya
Spring 2020
[Summer School website]
5. Classic Machine Learning classification pipeline
Raw data (ex: images)
Feature
Extraction
Classifier Decisor y = ‘CAT’
X1: weight
X2: height
Probabilities:
CAT: 0.7
DOG: 0.3
5Slide credits: Santiago Pascual (UPC TelecomBCN 2019). Garfield is a character created by Jim Davis.
6. Raw data (ex: images)
Feature
Extraction
Classifier Decisor y = ‘CAT’
X1: weight
X2: height
Probabilities:
CAT: 0.7
DOG: 0.3
Neural
Network
Shall we
extract
features now?
6
Classic Machine Learning classification pipeline
Slide credits: Santiago Pascual (UPC TelecomBCN 2019). Garfield is a character created by Jim Davis.
7. Raw data (ex: images)
Classifier Decisor y = ‘CAT’
Probabilities:
CAT: 0.7
DOG: 0.3
Neural
Network
We CAN inject the
raw data, and
features will be
learned!!
End to End concept
7
Deep Learning classification pipeline
Slide credits: Santiago Pascual (UPC TelecomBCN 2019). Garfield is a character created by Jim Davis.
9. 9
DL basic unit: The Perceptron
The Perceptron is seen as an analogy to a biological neuron, because it fire an
impulse once the sum of all inputs is over a threshold.
Minsky, Marvin, and Seymour A. Papert. Perceptrons: An introduction to computational geometry. 1969
11. 11
DL basic unit: The Perceptron
Weights and bias are the parameters that define the behavior. They must be
estimated during training.
12. 12
DL basic unit: The Perceptron
Multiple options as activation functions f(·):
13. 13
DL basic unit: The Perceptron
A single perceptron can only define linear decision boundaries.
Height
2D feature space
Weight
Slide credits: Santiago Pascual (UPC TelecomBCN 2019). Garfield and Odie are characters created by Jim Davis.
19. 19
Multilayer Perceptron (MLP)
In practice, deep neural networks nets can usually represent more complex
functions with less total neurons (and therefore, less parameters)
21. 21
Deep Neural Networks (DNN)
s1
s2
s3
CAT
DEER
DOG
.
.
.
Keep stacking hidden
layers to build deep nets
.
.
.
.
.
.
.
.
.
Slide credits: Santiago Pascual (UPC TelecomBCN 2019). Garfield is a character created by Jim Davis.
22. 22
Deep Neural Networks (DNN)
Slide credit: Santiago Pascual (UPC TelecomBCN 2019)
s1
s2
s3
CAT
DEER
DOG
.
.
.
Keep stacking hidden
layers to build deep nets
.
.
.
.
.
.
.
.
.
The concept of Deep Learning
arises when we have deep models
(many layers of processing), like in
Deep Neural Networks (DNNs)
23. 23
Deep (Hierarchical) Data Representations
Slide credit: Santiago Pascual (UPC TelecomBCN 2019)
Image Speech
Figure ref
24. 24
How to estimate the parameters ?
Rumelhart, David E., Geoffrey E. Hinton, and Ronald J. Williams. "Learning representations by back-propagating errors."
Cognitive modeling 5, no. 3 (1988).
Training a neural network with the
back-propagation algorithm.
25. 25
How to learn a memory unit ?
#RNN Alex Graves, “Supervised Sequence Labelling with Recurrent Neural Networks”
The hidden layers and the output
depend from previous states of the
hidden layers
Recurrent layer (RNN)
26. 26
How to learn a memory unit ?
#RNN Alex Graves, “Supervised Sequence Labelling with Recurrent Neural Networks”
Recurrent
Weights (U)
Feed-forward
Weights (W)
27. 27
How to reuse neurons ?
Fully Connected layer (FC) Convolutional layer (Conv)
Figures: Ranzatto
28. 28
Convolutional Neural Network (CNN)
#CNN #LeNet-5 LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document
recognition. Proceedings of the IEEE, 86(11), 2278-2324.
29. 29Oriol Vinyals, ”The Deep Learning Toolkt”. MIT Embodied Intelligence Seminar (2020)
30. 30
Many other researchers have also
contributed to the field as, for
example, those pointed out by LSTM
co-author Jürgen Schmidhuber in
“Deep Learning Conspirancy”.
32. 32
Big data for Vision: ImageNet
● 1,000 object classes
(categories).
● Images:
○ 1.2 M train
○ 100k test.
Deng, Jia, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. "Imagenet: A large-scale hierarchical image
database." CVPR 2009.
33. 33
Data Challenge: Social Biases
#Equalizer Burns, Kaylee, Lisa Anne Hendricks, Trevor Darrell, and Anna Rohrbach. "Women also Snowboard: Overcoming
Bias in Captioning Models." ECCV 2018.
34. 34
Data Challenge: Who owns data ?
Personal data
Internet of things - IoT
Neil Lawrence, OpenAI won’t benefit humanity without open data sharing (The Guardian, 2015)
37. 37
Computation ecological cost
Strubell, Emma, Ananya Ganesh, and Andrew McCallum. "Energy and Policy Considerations for Deep Learning in NLP." ACL
2019. [tweet]
49. Deep Reinforcement Learning (DRL)
Deep Reinforcement Learning (DRL) refers agents controlled by deep neural
networks.
50. 50
Mnih, Volodymyr, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, and Martin Riedmiller.
"Playing atari with deep reinforcement learning." NIPS Deep Learning Workshop (2013).
51. 51
Beyond Multimedia
#AlphaGo Silver, David, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George Van Den Driessche, Julian
Schrittwieser et al. "Mastering the game of Go with deep neural networks and tree search." Nature 2016.