Scribd
Upload
10 views25 pages

W1.2 Introduction

The document provides an overview of different machine learning techniques including supervised learning, unsupervised learning, and reinforcement learning. It discusses concepts like classification, regression, clustering, and evaluation of unsupervised models. It also covers common datasets and preprocessing steps for different data types.

Uploaded by

hiphoplistener
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
10 views25 pages

W1.2 Introduction

The document provides an overview of different machine learning techniques including supervised learning, unsupervised learning, and reinforcement learning. It discusses concepts like classification, regression, clustering, and evaluation of unsupervised models. It also covers common datasets and preprocessing steps for different data types.

Uploaded by

hiphoplistener
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 25

Chapter 1 - Introduction

Prepared by: Shier Nee, SAW


Based on: Probabilistic Machine Learning by Kevin Murphy
What is Machine Learning?

By Shakir Mohamed, research at Deep Mind.


2
AI Techniques

Unsupervised Reinforcement
Supervised Learning
Learning Learning

Interacts with
Labelled data with No labelled without environment, decide
guidance guidance action, learns by trial
and error method

3
Supervised Learning - Classification

4
Supervised Learning - Classification

5
Supervised Learning - Classification

Empirical Risk Minimization

6
Supervised Learning - Classification

Uncertainty  using conditional probability distribution

)))

Constraints: Softmax function


0≤ 𝑓 𝑐≤1
𝐶

∑ 𝑓 𝑐 =1
𝑐=1 7
Supervised Learning - Classification

Maximum Likelihood Estimation


Minimizing Negative Log Likelihood

8
Supervised Learning - Regression

Output = Real-value  Quadratic loss


Supervised Learning - Regression
Uncertainty  Assume output distribution = Gaussian

10
Supervised Learning - Regression

Linear Polynomial Deep Neural


Regression Regression Network
1 Feature Features
Engineering
Feature
Extraction
Automatically
Generalization
Unsupervised Learning

Unsupervised Reinforcement
Supervised Learning
Learning Learning

Interacts with
Labelled data with No labelled without environment, decide
guidance guidance action, learns by trial
and error method
Unsupervised Learning - Clustering

Goal:
Partition the input into regions that contain “similar” points.
Unsupervised Learning - Clustering

Linear Model
Non-linear Model
Unsupervised Learning –
Self-supervised Learning

From https://medium.com/analytics-vidhya/self-supervised-representation-learning-in-computer-vision-part-2-8254aaee937c
Unsupervised Learning – Evaluation

• It is difficult to evaluate since no ground truth.

• Use learned unsupervised representation as features to a


downstream supervised learning method

• If unsupervised model learn useful features, the prediction


in supervised model will increase.

• Evaluate unsupervised model by reducing the number of


labelled sample to get good performance.
Reinforcement Learning

Unsupervised Reinforcement
Supervised Learning
Learning Learning

Interacts with
Labelled data with No labelled without environment, decide
guidance guidance action, learns by trial
and error method
Reinforcement Learning
A system or agent has to learn how to interact with its environment.
This can be encoded by means of a policy a = (x), which specifies which action to take in response to
each possible input x (derived from the environment state).
Common Small Image Datasets
MNIST Fashion - MNIST

EMNIST CIFAR
Common Large Image Datasets
ImageNet

• This dataset spans 1000 object classes


• 1,281,167 training images,
• 50,000 validation images and
• 100,000 test images

https://www.image-net.org/download.php
Natural Language Processing

IMDB movie review


Natural Language
Processing (NLP)
- Classification
Natural Language Processing
Natural Language Processing (NLP)
- Translation
 Canadian parliament (English-French pairs)
 the European Union (Europarl).

- Document summarization, Question answering


Discrete Input Data
One hot encoding
3 colors (say red, green and blue)

one-hot(red) = [1; 0; 0],


one-hot(green) = [0; 1; 0],
one-hot(blue) = [0; 0; 1].

Feature Interaction
Text Data
Bag of words
• dropping punctuation,
• converting all words to lower case;
• dropping common but uninformative words – “and”, “the” (stop word removal_
• Replace word with their base form – “running”, “runs”  “run” (word stemming)

TF-IDF: Term Frequency — Inverse Document Frequency

DFi is the number of documents with term i

https://towardsdatascience.com/tf-idf-for-document-ranking-from-scratch-in-python-on-real-world-dataset-796d339a4089

You might also like