Overview

Machine Learning

Artificial Neural Networks - ANN’s

Deep Learning

Computer Vision - CV

Natural Language Processing - NLP

Why Big Data?

Machine Learning

Explicitly programmed:

power <- function(a,b) {
  c <- a^b # known function
  return(c)
}
power(1,3)
## 1
power(2,3)
## 8
power(3,4)
## 81

Machine Learning

What if we don’t know what function to apply, but we have a lot of examples?

func(2)
## 2.895133
func(1)
## 1.666248
func(6)
## 6.399109

Machine Learning

Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed.

  • Supervised machine learning algorithms
  • Unsupervised machine learning algorithms
  • Reinforcement machine learning algorithms

Supervised machine learning

Unsupervised machine learning

Reinforcement machine learning

Artificial Neural Network - ANN

Single-layer Perceptron

  • Input Nodes (input layer)
  • Connections and Weights
  • Activation Function
  • Output Nodes (output layer)
## 
## Attaching package: 'matrixStats'
## The following object is masked from 'package:plyr':
## 
##     count

Artificial Neural Network - ANN

Single-layer Perceptron

Let’s try and predict rating:

calories protein fat sodium fiber rating
70 4 1 130 10.0 68
120 3 5 15 2.0 34
70 4 1 260 9.0 59
50 4 0 140 14.0 94
110 2 2 180 1.5 30

Artificial Neural Network - ANN

Single-layer Perceptron

Artificial Neural Network - ANN

Single-layer Perceptron

calories protein fat sodium fiber rating pred_rating
120 3 5 15 2.0 34 31
70 4 1 260 9.0 59 70
110 2 2 180 1.5 30 34
90 2 1 200 4.0 49 47
120 1 2 220 0.0 18 22

Artificial Neural Network - ANN

Deep Learning

Artificial Neural Network - ANN

Deep Learning

calories protein fat sodium fiber rating pred_rating
120 3 5 15 2.0 34 34
70 4 1 260 9.0 59 65
110 2 2 180 1.5 30 34
90 2 1 200 4.0 49 48
120 1 2 220 0.0 18 20

Deep Learning

Components:

Network has 1 Layer Input with 5 nodes and 1 Layer Output with 1 node.

Deep Learning

Components:

Network has 3 hidden layers.

Deep Learning

Components:

1st Layer: 20nodes

Deep Learning

Components:

2nd Layer: 15nodes

Deep Learning

Components:

3d Layer: 5nodes

Deep Learning

Components:

4th Layer: 1node

Deep Learning

Components:

Train it with Backpropagation (change weights)

Computer Vision - CV

Computer Vision - CV

Computer Vision - CV

Natural Language Processing - NLP

Natural Language Processing - NLP

Word Embeddings

Cats are mammals.

Natural Language Processing - NLP

Word Embeddings

Cats are mammals.

w1 w2 w3

cats - are

are - cats

are - mammals

mammals - cats

cats - mammals

Natural Language Processing - NLP

Word Embeddings

cats - are 1 0 0 - 0 1 0

Natural Language Processing - NLP

Word Embeddings

are - cats 0 1 0 - 1 0 0

Natural Language Processing - NLP

Word Embeddings

mammals - cats 0 0 1 - 1 0 0

Natural Language Processing - NLP

Word Embeddings

Keep values of middle network as embeddings for input word

cat = [0.8874 0.4184 0.9876 0.2662 0.0479 0.5825 0.3993 0.0556 0.1302 0.8568 0.4025 0.5344 0.9492 0.0258 0.7850 0.1779 0.4387 0.9506 0.2799 0.9741 0.4159 0.9542 0.9834 0.1766 0.1999 0.2395 0.8560 0.1893 0.5676 0.2971 0.4173 0.6742 0.6488 0.4010 0.5583 0.4708 0.1981 0.4244 0.7771 0.9306 0.0068 0.3926 0.1718 0.5491 0.2884 0.3237 0.8570 0.2799 0.4225 0.2305]

Natural Language Processing - NLP

Word Embeddings

Keep values of middle network as embeddings for input word

cat = [0.8874 0.4184 0.9876 0.2662 0.0479 0.5825 0.3993 0.0556 0.1302 0.8568]

Natural Language Processing - NLP

Word Embeddings

Keep values of middle network as embeddings for input word

cat = [0.8874 0.4184 0.9876 0.2662 0.0479 0.5825 0.3993 0.0556 0.1302 0.8568]

The Marvelous Mathematics of Computational Linguistics:

King - Man + Woman = Queen

Natural Language Processing - NLP

Word Embeddings

The Marvelous Mathematics of Computational Linguistics:

King - Man + Woman = Queen

Natural Language Processing - NLP

Word Embeddings

Natural Language Processing - NLP

Word Embeddings

Why Big Data?

Why Big Data?

How much is Big Data?

Why Big Data?

Why Big Data?

Talon Supports Machine Learning

Talon Supports Machine Learning

Will add:

Talon Supports Machine Learning

Fast prototyping.

Machine Learning code templates.

Model Interpretation.

Big Data pipelines.

Workshops and tutorials.

Future Workshops & Tutorials?

Computer Vision.

Natural Language Processing.

Multi GPU training.

General Deep Learning.

Handle Big Data.

Quick Prototyping.

Other?

Thank you!