**Machine Learning**

Deep Learning

by Li Dai

Based On

Anshul Joshi Machine learning

Zhen Zuo Deep Learning

Deep Learning

by Li Dai

Based On

Anshul Joshi Machine learning

Zhen Zuo Deep Learning

How humankind know world

Machine Learning

Deep Learning

Trends

Opportunities

Do you like it

Why&How

**Outline**

Machine Learning

Work flow

Supervised Learning

Unsupervised Learning

Reinforcement Learning

Types of Machine Learning

What is Machine Learning?

What is machine learning.

Supervised and Unsupervised learning.

Some algorithms of supervised and unsupervised learning.

Reinforcement Learning

Clustering.

Recommendation systems. (Use case, example.)

"A computer program is said to learn from experience (E) with some class of tasks (T) and a performance measure (P) if its performance at tasks in T as measured by P improves with E"

Examples:

Automatic speech recognition

Automatic Voice/Face/Finger print recognition

Natural Language Processing

Automatic Medical diagnostics

Email spam detection

Advertisements

Content (image, video, text) categorization

Suspicious activity detection from CCTVs, fraudulent activities in banks. (ex. credit card fraud).

Frequent pattern mining

Predicting tastes in music (Pandora), in movies/shows (Netflix), predicting interests (Facebook, Grindr, Tinder), shopping list (Flipkart, Amazon) etc.

Supervised Learning

Unsupervised Learning

Reinforcement Learning

Supervised Learning

The correct labels of the training data are known.

Classification

: predict class from observations.

discrete/categorical variable

group the output into a class.

example: Email spam detection, Content (image, video, text) categorization, Google News.

Regression:

statistical process for estimating the relationships among variables.

means to predict the output value using training data.

real number/continuous.

example: Housing price prediction, sales prediction, web traffic prediction

Unsupervised Learning

The correct classes of the training data are not known

Clustering:

Task of grouping a set of objects in such a way that objects in the same group (called a cluster) are more similar (in some sense or another) to each other than to those in other groups (clusters).

Blind signal separation:

separation of a set of source signals from a set of mixed signals.

Use Case

Recommendation Engine

Decision Trees

A flow chart like tree structure

Internal node represents a test on the attribute

Branch represents the test of the result

Leaf nodes represents the class labels or class distribution.

Used to classify an unknown example.

k-Nearest Neighbour algorithm

All instances corresponds to n-D space.

The nearest neighbor are defined in terms of Euclidean distance.

The target function could be discrete or real valued.

For discrete valued, the kNN returns the most common value among the training examples.

Linear Regression (Prediction)

Predicting the value of one variable on the basis of other variable.

If two variables are involved, one is dependent variable that is to be found and one is independent variable which is the basis of finding the independent variable.

k Means clustering

Organizing data into classes such that there is:

high intra-class similarity

low inter-class similarity

Recommender systems have changed the way people find products, information, and even other people.

They study patterns of behavior to know what someone will prefer from among a collection of things he has never experienced.

content-based filtering, user-user collaborative filtering, item-item collaborative filtering, dimensionality reduction, and interactive critique-based recommenders.

Deep Learning

Fundamental Neural Network

Mainstream deep learning approaches

Conclusions

Function decomposition:

Convolutional Neural Network (CNN)

Motivated by the deep architecture of human brain and cognitive process, Deep Learning is a new area of Machine Learning research, which has been introduced with the objective of moving Machine Learning closer to one of its original goals: Artificial Intelligence.

Deep Learning is about learning multiple levels of representation and abstraction that help to make sense of data such as images, sound, and text.

Key principles of deep architecture:

Unsupervised learning of representations is used to pretrain each layer.

Unsupervised training of one layer at a time, on top of the previously trained ones. The representation learned at each level is the input for the next layer.

Use supervised training to fine-tune all the layers

Motivation

Fundamental Neural Network

Mainstream deep learning approaches

Recent applications of deep learning

Conclusion

Motivation

In order to learn the kind of complicated functions that can represent high-level abstractions, one may need deep architectures.

Current Recognition Approach

Traditional Neural Network

Deep Learning

Deep architectures are composed of multiple levels of non-linear operations, such as in neural nets with many hidden layers or in complicated propositional formulae re-using many sub-formulae.

image/video pixels

object class

Hand-designed Feature Extraction

Trainable Classifier

Features are not learned (e.g. SIFT, HOG)

Trainable classifier is often generic (e.g. SVM)

Need supervised training labels

problem

Combining multiple hand-designed features or do multi-kernel learning will lead to improvement of performance, but only few gain, and extracting multiple features is time consuming.

Can we learn better features? Unsupervised/Semisupervised?

Inspired by the architectural depth of the brain, neural network researchers had wanted for decades to train deep multi-layer neural networks, but no successful attempts were reported before 2006: researchers reported positive experimental results with typically two or three levels (i.e., one or two hidden layers), but training deeper networks consistently yielded poorer results.

Perceptron Neural Network

Feed-forward Neural Network

Recurrent Neural Network

Hierarchies in Vision

Mid-level features

Combination of attributes

Spatial pyramid

New methods for unsupervised pre-training have been developed (Restricted Boltzmann Machines = RBMs, autoencoders, contrastive estimation,etc.)

More efficient parameter estimation methods

Better understanding of model regularization

Improvements

Recent applications of deep learning

Forward-propagation/Backward-propagation:

A multilayer neural net can be thought of as a stack of logistic regression classifiers. Each input is the output of the previous layer.

Logistic neurons:

Loss:

1) Compute loss on small mini-batch (F-prop)

2) Compute gradient w.r.t. parameters (B-prop)

3) Use gradient to update parameters

Deep Belief Networks (DBN)

Deep Boltzmann Machine (DBM)

Conceptual example of convolutional neural network

The convolution and subsampling process

Graphical model of DBN: The structure is similar to a sigmoid belief network, except for the top two layers formed a Restricted Boltzmann Machine (RBM).

Comparison of different types of deep neural networks

What's

RBM

?

Sturcture of DBM: Each unit dependencies between hidden variables. All connections are undirected.

Flowchart of Deep Autoencoder

Restricted Boltzmann Machine (RBM)

Undirected graphical model of RBM

A pseudo-code of k-step Contrastive Divergence (CD-k)

Face expression recognition

Multi-modal learning

Parsing

Pattern completion

Share knowledge across categories

Data dimension reduction

2000-500-250-125-2

2000-2

Object Recognition

**Abstract**

**Simple**

**Chaos**

**How humankind know world**

Humankind

Emotion

Reasoning

Creation

Imagination

Learning

Computer

Hardworking

Fast

Machine Learning

Deep Learning

[10000hours Expert]

**Trends**

**Opportunities**

**Do you like it**

From Deductive To Inductive

From Explanation To Modeling

From Accuracy To Approximation

Trends of way knowing world

Artificial Intelligence Startups

https://angel.co/artificial-intelligence

http://en.wikipedia.org/wiki/List_of_mergers_and_acquisitions_by_Google

http://en.wikipedia.org/wiki/List_of_mergers_and_acquisitions_by_Facebook

http://en.wikipedia.org/wiki/List_of_mergers_and_acquisitions_by_Apple

http://en.wikipedia.org/wiki/List_of_mergers_and_acquisitions_by_Yahoo!

Artificial Intelligence

Business

Interdisciplinary

What humankind can think

will all become reality

Why&How

Panacea

The 10 Best Tech Careers In 2015

No. 1: Software Engineer, $98,074

No. 2: Database Administrator, $97,835

No. 3: Product Manager, $113,363

No. 4: Data Scientist, $104,476

No. 5: Solutions Architect, $121,657

No: 6: QA Engineer: $77,499

No. 7: Network Engineer: $87,518

No. 8: IT Project Manager, $103,710

No. 9: Mobile Developer: $79,810

No. 10: Sales Engineer: $91,318

We are still pursuing the truth of the world.

But the output of machine learning or deep learning from data is just the world from data, it works well but is not the world itself.

So machine learning and deep learning just help us know how but not know why.

We still need genius and more powerful methods to know the world.

Questions & Thanks

Programming

– R Python

Math Essential

statistics

– Probability

– Statistical inference

– Validation

– Estimates of error, confidence intervals

Linear algebra

– Hugely useful for compact representation of linear

– transformations on data

– Dimensionality reduction techniques

Optimization theory

Business Feeling

In Detail

Top Masters in this field

Peter L. Bartlett

Professor of Computer Science and Statistics

Department of Statistics,

Division of Computer Science/EECS

University of California, Berkeley

Research Expertise and Interest

statistics, machine learning, statistical learning theory, adaptive control

Michael I. Jordan

Pehong Chen Distinguished Professor

Department of EECS

Department of Statistics

University of California, Berkeley

Research Expertise and Interest

Mixtures of experts, spectral clustering, Graphical model, nonparametric, Bayesian.

Yoshua Bengio

Full Professor

Department of Computer Science

and Operations Research

Canada Research Chair in

Statistical Learning Algorithms

Université de Montréal

Research Expertise and Interest

Deep Learning

Geoffrey E. Hinton

Inventor of the backpropagation and

contrastive divergence training algorithms

Part time for Google researcher

Department of Computer Science

University of Toronto

Research expertise and interest

deep learning

Yann LeCun,

Director of AI Research, Facebook

Founding Director of the NYU Center for Data Science

Silver Professor of Computer Science, Neural Science,

and Electrical and Computer Engineering,

The Courant Institute of Mathematical Sciences,

Center for Neural Science, and

Electrical and Computer Engineering Department, NYU School of Engineering

New York University.

ANDREW NG

Andrew Ng is Associate Professor of Computer

Science at Stanford;

Chief Scientist of Baidu;

and Chairman and Co-founder of Coursera.

Research expertise and interest

machine learning

Daphne Koller

Professor

Computer Science Department at Stanford University

MacArthur Foundation Fellowship

Research expertise and interest

Probabilistic models and machine learning

John Lafferty

Louis Block Professor

Department of Statistics

Department of Computer Science

Physical Sciences Division

University of Chicago

Before 2011 Professor of Computer Science,

Machine Learning, and Statistics

at Carnegie Mellon University.

Research expertise and interest

statistics and machine learning

Yaser S. Abu-Mostafa

Professor

Electrical Engineering and Computer Science

California Institute of Technology

Research expertise and interest

machine learning and computational finance

David M. Blei

Professor

Statistics and Computer Science

Columbia University

Research expertise and interest

Probabilistic graphical models

Approximate posterior inference

Topic models

Bayesian nonparametric statistics

Journal of Machine Learning Research

Based On

Anshul Joshi Machine learning

Zhen Zuo Deep Learning

Basic setup of the learning problem

Death will be precious

Robots will come to protect mind and intelligence

Bodies Gone Thoughts Left

Love will be commodity

Global Brain dominates world

Machine Learning

Deep Learning

Artificial Intelligence