Ex 01. ML Part 1

field of study

that gives computers the ability to learn

without being explicitly programmed

dataset S

Model M

Objective Function L

Algorithm A

set of samples generated by some sysetm or process

samples can be single datapoints or pairs of inpiut and output values

adjustable and compact representaiton

of a certain class of input/output relationships

that is hypothesized to be capable of

modeling the system or process which generates dataset S

function that encodes the current performance of M (e.g. loss or reward)

learnign algorithm

that adjusts M

based on S and L

learning and development

memory, knowledge and internal simulation

perception

autonomy

modeling and implementation of

biologica learning mechanisms

• operant conditioning

• implicit learning

• explicit learning

• perception

• …

modelin and implementation of the

• encoding

• storage

• and retrieval

of

• facts

• experiences

• actions

-> e.g. associative memory

learning basic features

to detect and categorize

perceptual stimuli

(e.g. unsupervised learning of visual features)

dynamic adaption to changes

in the environment

(e.g. continuous online learning from live data stream)

important for cognitive systems situated in complex dynamic environments

=> not possible to program everything in advance

environment changing continuously

dynamics and objects are too complex to be modeled explicitly

system itself is subject to change

• -> e.g. growth, aging, injury, tool use,…

image classification

speech recognition

autonomous driving

recommendation systems

threat protection

control systems

training part

• have / gather training data

• extract, engineer, select features

• choose model and learning algorithm and train

execution

• get feedback

  • data labels, reward signals

• do test/validation

feature engineering = selection of right features

critical for the performance of machine learning model

=> as features must contain information required for predicitons

there usually are a vast multitude of features to choose from

to describe the data at hand

-> e.g. shapes, colors, histograms, filters,…

grouped into a feature vector

train model M

in hypothesis space H

using a learning algorithm A

so that M minimizes Loss L for dataset S

=> reffered to as inductive learning

on the properties of dataset S

set of all possible hypothesis

=> set from which ML algo determines the best possible one (through learning…)

unlabaled data

• S = {x1, x2, x3, …, xn}

labeled data

• S = {(x1,y1), (x2,y2),…,(xn,yn)}

mixed data

• mix from labeled and unlabeled data

• S = {(x1,y1), (x2,y2),…,(xn,yn)} U {x1, x2, x3, …, xn}

• x1 != (x1, y1)…

dynamic data

• S = {(xt,st) | st+1 = D(xt,st) AND 0 <= t <= T}

• -> dynamical system where st denotes system state at time t

• => where next state is funciton of current state and input…

unsupervised learning

• solely unlabaled data

• discovery of structural features in the data set

reinforcement learning

• interaction with the environment

• reward signal encodes feedback for the policy

semi-supervised learning

• labeled and unlabeled training samples

• a priori assumptions on input data required

supervised learning

• all training samples are labeled

• desired output is specified exactly

in most cases by the type of data available

• unlabeled -> unsupervised

• labeled -> supervised

• mixed -> semi-supervised

• dynamic environment -> reinforcement

manual labeling of data

as preprocessing tool for initial data analysis