doktorska disertacija
Abstract
Vrednotenje metod strojnega učenja se tradicionalno izvaja z oceno delovanja na ročno označeni testni množici. Ta približek uporabljamo preprosto zato, ker nimamo na voljo boljše metode. Moramo se zavedati, da se je metoda strojnega učenja učila iz zelo sorodnih (podobnih) podatkov. Torej so posledično vsa predvidevanja o zmogljivosti na realnih podatkih, ki so ocenjena na podlagi testne množice, optimistična. Dejanska vrednost metode strojnega učenja temelji na njeni sposobnosti tvorjenja dobrih hipotez.
Ovrednotenje metod strojnega učenja z obdelavo naravnega jezika vpelje popolnoma nov vir: izsledke znanstvenih raziskav in študij, zapisanih v znanstvenih objavah. Ta vir ponuja objektivno metodo ocenitve rezultatov na podlagi podatkov iz raziskav. V veliki meri zmanjša delo domenskih strokovnjakov, ki je potrebno za ovrednotenje rezultatov strojnega učenja. Hkrati je tak pristop zmožen tvoriti enciklopedično zbirko formaliziranega znanja, ki je splošno uporabna.
Keywords
verifikacija;strojno učenje;podatkovno rudarjenje;validacija;obdelava naravnega jezika;
Data
Language: |
Slovenian |
Year of publishing: |
2013 |
Typology: |
2.08 - Doctoral Dissertation |
Organization: |
UM FERI - Faculty of Electrical Engineering and Computer Science |
Publisher: |
[S. Pohorec] |
UDC: |
004.655.3:004.855(043.3) |
COBISS: |
266736128
|
Views: |
2474 |
Downloads: |
331 |
Average score: |
0 (0 votes) |
Metadata: |
|
Other data
Secondary language: |
English |
Secondary title: |
Evaluation of machine learning methods with natural language processing |
Secondary abstract: |
Validation of machine learning methods has traditionally been performed with evaluation on hand annotated test sets. This procedure represents an approximation and is used for lack of a better approach. We should consider that the machine learning method has learned from very similar data, consequently all predictions on real data performance, based on this test, are optimistic. The real value of a machine learning method lies in its ability to form good hypothesis.
Natural language processing as a method of evaluation of machine knowledge introduces a new source of validation: research results from scientific studies and papers published in respected conferences and journals. This new source offers a method of objective evaluation of machine learning results. It can greatly diminish the manual effort of domain experts who perform machine learning evaluation. At the same time this approach is capable of forming an encyclopedic database of formal knowledge |
Secondary keywords: |
verification;machine learning;data mining;validation;natural language processing;Strojno učenje;Disertacije;Podatkovno rudarjenje; |
URN: |
URN:SI:UM: |
Type (COBISS): |
Dissertation |
Thesis comment: |
Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko |
Pages: |
XX, 155 str. |
ID: |
8726095 |