| Popis: |
Cilj diplomskega dela je bil razviti algoritem za učenje najboljše učne strategije v primeru interaktivnega učenja med robotom in človekom. Predstavili smo definicijo in formalizacijo učne strategije, ki določa obnašanje učenca in učitelja v interaktivnem učnem procesu. Predstavili smo tudi genetski algoritem, s katerim smo reševali naš optimizacijski problem. Vektorje, s katerimi so predstavljene učne strategije, smo vrednotili, križali in mutirali ter tako poskušali priti z vsako iteracijo do vedno boljše učne strategije. Vektorje smo vrednotili z vrednostjo mere prepoznavanja, ki nam pove, kako je bil uspešen algoritem pri prepoznavanju določenega koncepta po učenju z neko učno strategijo, in ceno tutorstva, ki nam pove, koliko je v učnem procesu sodeloval tutor. Rezultati so pokazali, da skozi iteracije genetskega algoritma narašča kakovost učnih strategij. To se odraža na povečani uspešnosti vrednosti mere prepoznavanja in zmanjšanju cene tutorstva. The main goal of this thesis was to develop an algoritem for learning the best strategy in the case of interactive learning between a human and a robot. We presented the definition and formalization of a learning strategy. A learning strategy specifies the behaviour of a student and a teacher in a interactive learning process. We also presented a genetic algorithm to resolve our optimisation problem. We tryed to inpruve vectors which are used to present learning strategies. The vectors were evaluated, crossed and mutated each iteration. For evaluation of the vectors we used recognition score which measures how successful was the algorithm at recognizing a specific concept after learning it by using a specific strategy. We also used tutoring cost to measure the tutor involvment in the learning process. The results show that the quality of the strategies increases each iteration. We notice the inprovement as an increase in the recognition score and decrease of tutoring cost. |
