basics of machine learning€¦ · basics of machine learning joel reijonen lead data scientist at...

22
Ericsson Internal | 2018-02-21 Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson [email protected]

Upload: others

Post on 18-Oct-2020

13 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Basics of Machine Learning

Joel ReijonenLead Data Scientist at [email protected]

Page 2: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Machine Learning –Motivation

— We are living in a connected world where the connected devices and users produce more data than we have ever seen

— How do we cope with such high volumes of data?

— How do we ensure that services today will also function tomorrow?

— Here machine learning may become handy..

Figure taken from Ericsson Mobility Report: Internet of Things forecast

Page 3: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Machine Learning –What is it?

— Machine learning refers to the machine’s ability to improve its own performance independently

— Machine learning strives to extract additional, hidden, information from the data

— Machine learning performs well with multi-dimensional data whereas human gets easily lost in the details

Intelligence Amplification

Artificial Intelligence

Machine Learning

Machine Intelligence

Big Data Models

Algorithms

Page 4: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Rules

Machine Learning –New way of thinking

Algorithm Programming

Machine Learning

Rules

DataAnswers

Data

Answers

Page 5: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Functionality of Machine Learning

— In machine learning, the main idea is to construct a model

— The constructed model consists of parameters or weights

— Learning happens by updating weights

x1

x2

w11

w12

w13

w14

w21

w22

w23

y

Input layer

Hidden layer

Output layer

wij = weighted connection

Page 6: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Machine Learning Techniques

Unsupervised Learning

Supervised Learning

Machine Learning

Other Learning Techniques

Classification Regression Clustering Reinforcement Learning

Deep Learning

Page 7: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Supervised Learning

— Supervised learning is a technique where learning is based on training from examples or experiences in the past

— Training set is composed of labeled data which consists real observed input and output values.

— Goal: Predict class or value

Raw data

Supervisor

Supervised Learning

Training Outputs

Training inputs

Output

Page 8: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Supervised Learning –Classification

— Classification is a supervised learning method that determines for which output or class do the sampling of inputs belong to

— Classification can be used in applications such as image recognition, pattern detection and natural language processing

— Potential industrial use-cases: surveillance, quality control, system health monitoring

Class 2

Class 1

Classification rule

Page 9: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Supervised Learning –Regression

— Regression is a supervised learning method where the inputs of the system are mapped on outputs that are numeric values

— Regression is used in applications that take advantage of estimations, predictions and forecasting

— Potential industrial use-cases: battery power consumption, power plant engine load optimization, stock-market prediction

Regression

Page 10: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Unsupervised Learning

— Unsupervised learning is a technique that does not utilize the observed output values rather takes advantage of input values

— Unsupervised learning pursues to extract main features and structures of the data

— Goal: Determine groupings and pattern

Raw data

Unsupervised Learning

Output

Self-guided learning

Page 11: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Unsupervised Learning –Clustering

— Clustering is an unsupervised learning method where inputs with similar features and attributes are allocated in the same cluster

— Clustering can be utilized in applications such as image analysis, data mining and pattern recognition

— Potential industrial use-cases: customer clustering, fault detection in power plant engines

Cluster 1

Cluster 3

Cluster 2

Page 12: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Other Learning Techniques

— Other learning methods are techniques which may have similarities with supervised or unsupervised techniques but yet they have significant differences in their functionalities to be categorized differently

— Most of the other learning techniques take advantage of feature extraction and rewarding policies

— Goal: Learn features and achieve rewards from the environment

Raw data

Other Learning techniques

Output

Feature learning

Environment

Page 13: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Reward

Other Learning Techniques –Reinforcement Learning

— In reinforcement learning, the machine is rewarded if the actions or decisions that the machine has made have increased overall performance in a certain environment

— Reinforcement learning can be applied in applications that take advantage of multi-agent systems, controlling systems and swarm intelligence

— Potential industrial use-cases: traffic light management, robotics, video games

Action

Environment

Machine

State

Page 14: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Other Learning Techniques –Deep Learning

— Deep learning takes a new approach on learning multiple layers of meaningful representations

— Each learned layer introduces new extracted features

— Applications that take advantage of deep learning include, e.g., near-human-level image classification, speech recognition and natural language processing

— Potential industrial use-cases: robotics, advanced system management, digital assistants

Page 15: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Other Learning Techniques –Deep Learning

Fault detection

FailureNo Failure

Traditional Machine Learning approach

Input Feature extraction

Output

Fault detection

FailureNo Failure

Deep Learning approach

Input Feature extraction

Output

Page 16: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Neural Networks

— Neural networks are a common way to implement previously described machine learning techniques

— Neural networks are inspired by biological neural networks where high number of interconnected neurons process information in parallel

— In neural networks, the connections are called synapsesand processing units as neurons

— Goal: Neural networks strive to achieve human-like performance in wide field of applications

x1

x2

w11

w12

w13

w14

w21

w22

w23

y

Input layer

Hidden layer

Output layer

Page 17: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Output

Neural Networks –Training

Input

Neural Network

Target

Loss function

Optimizer

Modify weights

— Neural network training is iterative process where the learned performance of the neural network is evaluated by using a loss function

— Loss function (cost function) refers to an objective function that evaluates the performance of the neural network

— By utilizing results from loss function, the optimizer handles the weight modification by updating the weights in such a way that it strives to improve the performance of the neural network

Page 18: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Beyond Machine Learning –Artificial Intelligence

— In artificial intelligence, machine learning is an essential concept in order to enable continuous learning from examples or from experiences in the past

— Artificial intelligence take advantage of multiple functionalities but the two most obvious are learning and decisiong making

— In learning, artificial intelligence utilizes machine learning techniques to train models that describe the system behaviour and performance based on the training data

— In decision making, artificial intelligence utilizes trained machine learning models in inferences

— Based on these inferences, artificial intelligence can perform decisions by taking advantage of deduction logic

Machine Learning

Artificial Intelligence

Page 19: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Beyond Machine Learning –Data preparation

— In order to achieve reliable results in machine learning, raw data needs to be pre-processed

— Raw data may include noise, redundancy, missing values and domination that may cause counterproductive effects

— Proper data preparation is a preliminary requirement for efficient machine learning

Raw Data

Prepared Data

Structured Data

Cleaned Data

Page 20: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Beyond Machine Learning –Skills to Master

— In order to become a superstar in machine learning, I would recommend to learn following skills:

— Linear algebra

— Calculus

— Probability theory and statistics

— Theoretical computer science

— Programming skills e.g. in Python, R or C++

— As interesting machine learning frameworks, I would recommend following:

— TensorFlow

— PyTorch

— Scikit-learn

— Keras

— MXNet

Page 21: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Ericsson Internal | 2018-02-21

Q & A

Page 22: Basics of Machine Learning€¦ · Basics of Machine Learning Joel Reijonen Lead Data Scientist at Ericsson joel.reijonen@ericsson.com. Ericsson Internal | 2018-02-21 Machine Learning

Thank you for participating!Contact: [email protected]