Exercise 1 - Basic Concepts in Machine Learning Flashcards
Learning
Learning is the acquisition of new information or knowledge or the process to acquire knowledge or skill by systematic study or by trial and error.
Google Translate ML?
Yes
calls with Google to restaurant ML based?
Yes. Only work in small domain
WildCat (The World’s Fastest Quadruped RobotI) ML based?
No
ML definition
- field of study that gives computers the ability to learn without being explicitly programmed
Four components of a ML system
- Dataset
- Model
- Objective Function
- Algorithm
Is ML a prerequisite for the implementation of cognitive functions in artificial cognitive systems?
- yes
four cognitive functions in artificial cognitive systems
- Learning and development
- Memory, knowledge, and internal simulation
- Perception
- Autonomy
Learning and Development in the context of artificial cognitive systems
- modelling and implementation of biological learning mechanisms (operant conditioning, implicit learning, explicit learning, perception)
implicit learning
learning without conscious operations
explicit learning
conscious operation where the individual works and tests hypothesis in a search for structure
How is perception implemented in ML?
- e.g. unsupervised learning of visual features
What is autonomy?
dynamic adaption to changes in the environment
For which environments is ML useful?
For artificial cognitive systems that are situated in complex dynamic environments.
Why is it not possible to program everything in advance in complex dynamic environments?
- environment is changing continuously
- dynamics too complex to be modeled explicitly (faces)
- system itself is subject to change (through growth, aging etc. )
feature engineering
- selection of the right features. Features must contain the information required for predicitions
inductive learning
- specific examples -> general rule
Definition of ML task
Train model M in a hypothesis space H using a learning algorithm A so that M minimizes loss L for dataset S. This type of learning is inductive learning.
-> The choice of H and L depends heavily on the properties of S
Mixed data
labeled and unlabeled data
Dynamic data
Zeitreihen
Do natural datasets have specific structural featues?
Yes.
Semi-supervised learning
- Mixed data: labeled and unlabeled training samples
- a priori assumptions on input data required
reinforcement learning
-Dynamic environment: interaction with the environment
- reward signal encodes feedback for the policy
(trail and error)
Types of ML
Unsupervised learning
Semi-supervised learning
Reinforcement learning
Supervised learning
What is the choice of a learning paradigm motivated by?
The type of data available
