Close
September 16, 2022

What does mean „to learn” in machine learning

What does mean „to learn” in machine learning

Machine learning is a subgroup of the wider field of artificial intelligence. The field of machine learning is devoted to understanding and building a system that can “learn”. It means that system improves performance on some set of tasks based on some data.

 

In traditional machine learning, learning is divided into 3 main groups due to the nature of the signal or feedback, these are [1]:

  • supervised learning;
  • unsupervised learning:
  • reinforcement learning.

In supervised learning an algorithm is trying to find optimal function g:X→Y that maps input data to output data.

 

Let top image be supervised dataset of n samples such that xi is input (feature vector of i-th sample) and yi is output (class).

 

During learning process input data is feed to Learning system and the Learning system generates output top image. The optimal function g is selected on the basis of minimalizing summary loss function for all elements of dataset, so that R(g) is minimal, where R(g) is defined by equation (1) [2, 3].

 

top image

 

where:
λ – parameter that controls the bias-variance tradeoff.
C(g) - regularization penalty function of g.

 

In unsupervised learning input data x is not labeled, that means. z=x ϵ Rd, so the algorithm learns patterns within the input data. Method of unsupervised learning can be divided into 3 groups according to task [4]:

  • clustering;
  • association Rules;
  • dimensionality reduction.

Reinforcement learning algorithms learns agents so they can find the way to make a set of actions in an environment that maximize the notion of cumulative reward. Actions for single agents are called policy map – equation 2 and 3 [5].

 

top image

 

top image

 

where:
a - action;
s - state.

 

Function a,s retruns probability of taking action a at state s. The algorithm has to find a policy that maximized expected return – state-value function V(s) which is defined by equation (4) [5].

 

top image

 

where:
t – timestep;
rt – reward;
γ ϵ [0,1) – discount-rate.

 


 

Bibliography:

 

[1]
H. D. Wehle, „Machine Learning, Deep Learning, and AI: What’s the Difference?”, w Data Scientist Innovation Day, 2017.
[2]
Q. Liu I Y. Wu, „Supervised Learning,” w Encyclopedia of the Sciences of Learning, Boston, MA, Springer, 2012, s. 3243-3245.
[3]
V. Vapnik, „Principles of Risk Minimization for Learning Theory”, w Advances in Neural Information Processing Systems, The MIT Press, 1991, s. 831-838.
[4]
IBM, „What is Unsupervised Learning?”, IBM Cloud Education, 21 09 2020. [Online]. Available: https://www.ibm.com/cloud/learn/unsupervised-learning. [Date of access: 14 07 2022].
[5]
R. S. Sutton i A. G. Barto, Reinforcement Learning: An Introduction, MIT Press, 2018.

Wystąpił błąd. Sprawdź pola formularza i spróbuj ponownie.
Ten adres jest już w naszej bazie.
Subskrypcja została dodana. Dziękujemy!
CosmoEye LLC based in Lublin announces that it has received on 11.05.2022 through Arkley Brinc limited liability company ASI S.K.A. under the Program PFR Starter Closed Investment Fund public aid from the European Funds in the amount of 2.000. 000.00 (two million PLN) for the implementation of the project on the development and commercialization of a streaming B2B system for warehouse management and enterprise resource planning, using integrated cameras (hardware) and artificial intelligence tool for real-time image analysis according to the management plan.

Strona korzysta z plików cookie w celu realizacji usług zgodnie z Polityką Prywatności. Możesz określić warunki przechowywania lub dostępu mechanizmu cookie w Twojej przeglądarce.

Akceptuję
Pobierz prezentację Download