1.
Kompleksnost neuronskih modela:
Correct Answer(s)
C. Raste sa porastom broja sinapsi
D. Raste sa porastom broja neurona
Explanation
The complexity of neural models increases with the increase in the number of synapses and neurons. As the number of synapses increases, the model becomes more intricate and requires more computational resources to process the information. Similarly, as the number of neurons increases, the model becomes more complex and can handle more intricate tasks. Therefore, both the number of synapses and neurons contribute to the overall complexity of neural models.
2.
Pravilo klasifikacije 1 NN glasi:
Correct Answer
C. X pripada klasi kojoj pripada vecina njenih prvih k suseda
Explanation
The correct answer states that X belongs to the class to which the majority of its first k neighbors belong. This means that when classifying X, we consider the classes of its k nearest neighbors and assign X to the class that has the most representatives among those neighbors.
3.
Ako je klasa hipoteza H skup svih prava u 2D, tada je VC dimenzija ove klase:
Correct Answer
D. 3
Explanation
The VC dimension of a hypothesis class is the maximum number of points that can be shattered by the class. In this case, if the class H is the set of all lines in 2D, then the VC dimension is 3. This means that there exist 3 points that can be shattered by the class, but any set of 4 points cannot be shattered.
4.
Validacioni skup sluzi za:
Correct Answer
A. Selekciju modela
Explanation
The correct answer is "Selekciju modela" (Selection of models). Validation set is used to select the best model among a set of candidate models. It helps in evaluating the performance of different models and choosing the one that performs the best on the validation set. By comparing the performance of different models on the validation set, we can determine which model is the most suitable for the given task.
5.
Sta su konacne klase hipoteza?
Correct Answer
C. One klase koje sadrze konacan broj prediktora (hipoteza)
6.
Biometrijska autentifikacija se moze posmatrati kao: ???
Correct Answer
A. Klasifikacioni problem sa vise od dve klase
Explanation
Biometrijska autentifikacija se može posmatrati kao klasifikacioni problem sa više od dve klase. Kada se koristi biometrijska autentifikacija, sistem treba da klasifikuje korisnike na osnovu njihovih biometrijskih karakteristika, kao što su otisak prsta, prepoznavanje lica ili glasa. Budući da postoji više od dve moguće klase (na primer, autentifikovani korisnici i neautentifikovani korisnici), ova situacija se može posmatrati kao klasifikacioni problem sa više od dve klase.
7.
U scenariju pasivnog obucavanja, na raspolaganju su:
Correct Answer
B. Samo uzorci iz obucavajuceg skupa
Explanation
In passive learning scenario, only samples from the training set are available. This means that the system does not generate any additional samples or create a separate subset of samples. The system solely relies on the samples provided in the training set for learning purposes.
8.
VC dimenzija klase svih hiperravni u K - dimenzionom prostoru je:
Correct Answer
B. K+1
Explanation
The VC dimension of a class of hyperplanes in a K-dimensional space is k+1. The VC dimension represents the maximum number of points that can be shattered by the class of hyperplanes. In this case, since the class consists of hyperplanes in a K-dimensional space, the VC dimension is equal to K+1. This means that the class of hyperplanes can shatter any set of K+1 points, but there exists a set of K+2 points that cannot be shattered by the class.
9.
Masinsko ucenje je neophodno u slucaju da:
Correct Answer(s)
C. Ne postoje eksperti za datu oblast razmatranja
D. Ljudi nisu u stanju da objasne svoju ekspertizu
Explanation
Machine learning is necessary when there are no experts available in the given field of consideration and when humans are unable to explain their expertise. This suggests that machine learning can be used as an alternative approach to gain insights and make predictions in situations where there is a lack of human expertise or understanding.
10.
Kriva obucavanja prikazuje:
Correct Answer
B. Zavisnost greske na test i validacionom skupu prilikom postepenog povecanja broja podataka, koristeci prethodne podatke kao prefiks uvecanom skupu podataka
Explanation
The correct answer is "Zavisnost greske na test i validacionom skupu prilikom postepenog povecanja broja podataka, koristeci prethodne podatke kao prefiks uvecanom skupu podataka." This answer suggests that the training error is not considered, but rather the focus is on the error on both the test and validation sets as the number of data points increases. The previous data is used as a prefix to the increased dataset, indicating that the model's performance is evaluated as more data is added.
11.
Ako je krosvalidaciona greska velika, potrebno je:
Correct Answer(s)
B. Obezbediti vise novih podataka
C. Smanjiti broj obelezja
D. Promeniti model
Explanation
If the cross-validation error is large, it suggests that the current model is not performing well on unseen data. To improve the model's performance, it is necessary to provide more new data, as having more data can help the model learn better patterns and make more accurate predictions. Additionally, reducing the number of features can help to simplify the model and prevent overfitting. Finally, changing the model itself can be beneficial, as different models have different strengths and weaknesses, and a different model may be better suited for the given problem.
12.
Sta je prava mera performansi jednog sistema masinskog ucenja?
Correct Answer
B. Tacnost na nevidjenim primerima
Explanation
The correct answer is "Tacnost na nevidjenim primerima" (Accuracy on unseen examples). This is because the true measure of performance for a machine learning system lies in its ability to accurately predict outcomes on new, unseen data. While accuracy on seen examples can give an indication of the system's performance on the training data, it may not necessarily generalize well to new data. Therefore, the accuracy on unseen examples is a more reliable measure of the system's performance.
13.
Ako je velika greska na treningu, moguca poboljsanja su:
Correct Answer(s)
B. Dodati vise obelezja
C. Produziti obucavanje
F. Izabrati kompleksniji model
Explanation
The possible improvements in case of a large error during training are to add more features, extend the training process, and choose a more complex model. These actions can help improve the model's performance and accuracy. By adding more features, the model can capture more relevant information from the data. Extending the training process allows the model to learn more patterns and adjust its parameters accordingly. Choosing a more complex model can provide the model with more flexibility and capacity to capture complex relationships in the data.
14.
U sistemima masinskog ucenja sa induktivnim transferom:
Correct Answer
B. Sistem obucen za jedan problem se koristi prilikom obucavanja sistema za neki drugi problem
Explanation
In systems with inductive transfer learning, a trained system for one problem is used to train a system for another problem. This means that the knowledge and experience gained from solving one problem can be transferred and applied to solve a different problem. This approach allows for the efficient use of previously trained models and can lead to improved performance on new tasks.
15.
Kod transduktivnog ucenja cilj je sinteza sistema masinskog ucenja koji se dobro ponasa:
Correct Answer
B. Samo na test primerima, koji su bez oznaka prisutni u fazi obucavanja
Explanation
The correct answer is "Samo na test primerima, koji su bez oznaka prisutni u fazi obucavanja" which means "Only on test examples that are unlabeled during the training phase". This suggests that the goal of transductive learning is to synthesize a machine learning system that performs well specifically on test examples that were not labeled during the training phase.
16.
U metodu histograma procena funkcije gustine verovatnoce na skup uzoraka duzine N ima:
Correct Answer
C. P_N(X)=(1/N) ((broj uzoraka u binu u kome je X)/(velicina bina koji sadrzi X))
Explanation
The correct answer is P_N(X)=(1/N) ((broj uzoraka u binu u kome je X)/(velicina bina koji sadrzi X)). This equation represents the method of histogram estimation of the probability density function on a sample set of length N. It calculates the probability density at a specific value X by dividing the number of samples in the bin containing X by the size of the bin, and then scaling it by 1/N to normalize the result.
17.
Kod induktivnog ucenja cilj je da se sintetisani sistem masinskog ucenja dobro ponasa na:
Correct Answer
B. Svim primerima dobijenim iz raspodele koja je ista kao i za obucavajuci skup
Explanation
The goal of inductive learning is for the synthesized machine learning system to perform well on examples obtained from the same distribution as the training set. This means that the system should be able to generalize its learning from the training set to unseen examples that come from the same distribution. Therefore, the correct answer states that the system should perform well on examples obtained from the distribution that is identical to the distribution of the training set.
18.
Ukoliko fiksiramo broj uzoraka u obucavajucem skupu, a povecamo dimenzionalnost (vektor obelezja je sve veci): ???
Correct Answer
A. Binovi ce postajati sve prazniji
Explanation
As the dimensionality of the feature vector increases while keeping the number of samples in the training set fixed, the number of possible combinations of feature values also increases. This means that each bin, which represents a specific combination of feature values, will have fewer samples assigned to it. Therefore, the bins will become increasingly empty.
19.
U sistemima masinskog ucenja, ulaz su:
Correct Answer
A. Program + izlaz
Explanation
In machine learning systems, the input consists of both the program and the output. This means that the program provides instructions or algorithms for the system to follow, while the output is the desired result or prediction that the system aims to achieve. By combining the program and the output, the machine learning system can learn and improve its performance over time.
20.
Sta je ERM (Empirical Risc Minimization) princip sa induktivnim Bijasom?
Correct Answer
D. Minimizacija greske na trening skupu uz restrikciju prostora mogucih prediktora(hipo...
21.
Greska aproksimacije (bajas) se definise kao:
Correct Answer
C. Minimalna greska ostvariva u datoj klasi hipoteza H
Explanation
The correct answer is "Minimalna greska ostvariva u datoj klasi hipoteza H." This means that the approximation error (bajas) is defined as the minimum error achievable within a given class of hypotheses H. This suggests that the goal is to minimize the error in order to achieve the best possible approximation within the class of hypotheses.
22.
Kakva je razlika izmedju klasifikacije i regresije?
Correct Answer(s)
C. Izlaz sistema masinskog ucenja je kontinualan u slucaju regresije
D. Izlaz sistema masinskog ucenja je diskretan u slucaju klasifikacije
G. Ulaz sistema masinskog ucenja je kontinualan u slucaju regresije
H. Ulaz sistema masinskog ucenja je diskretan u slucaju klasifikacije
Explanation
In machine learning, classification is used to predict discrete or categorical outcomes, while regression is used to predict continuous outcomes. Therefore, the correct answer states that the output of machine learning is continuous in the case of regression, and discrete in the case of classification. Additionally, it mentions that the input is continuous in the case of regression and discrete in the case of classification.
23.
Pomeraj ili bajas u masinskom ucenju ima znacenje:
Correct Answer
B. Uvodjenje restrikcija na prostor hipoteza u kome se trazi finalno resenje
Explanation
This answer suggests that "Pomeraj ili bajas" in machine learning refers to introducing restrictions on the hypothesis space in which the final solution is sought.
24.
Greska estimacija (varijansa) se moze smanjiti:
Correct Answer(s)
B. Povecanjem obucavajuceg skupa
C. Redukovanjem klase hipoteza H
Explanation
By increasing the training set, we are providing the model with more data to learn from, which can help reduce the estimation error. Increasing the training set allows the model to capture a wider range of patterns and variations in the data, leading to a more accurate estimation. Similarly, by reducing the hypothesis class H, we are simplifying the model and making it less complex. This can also help reduce the estimation error as a simpler model is less likely to overfit the data.
25.
Kod aktivnog ucenja, sistem masinskog ucenja ima mogucnost da do izvesne mere:
Correct Answer
A. Zahteva od ucitelja da dodaje primere i da ih oznaci
Explanation
The correct answer is "Zahteva od ucitelja da dodaje primere i da ih oznaci" (Requires the teacher to add examples and label them). This means that in active learning, the machine learning system requires the teacher to provide additional examples and label them in order to improve its learning. This is because active learning involves the system actively selecting which examples it wants to learn from, and it needs the teacher's input to identify and label these examples.
26.
Apsolutni bajas u masinskom ucenju ima znacenje:
Correct Answer
A. Uvodjenja restrikcija u prostor hipoteza
Explanation
The correct answer is "Uvodjenja restrikcija u prostor hipoteza" which means "Introducing restrictions into the hypothesis space." This means that the absolute bias in machine learning refers to the introduction of constraints or limitations on the possible hypotheses that can be considered. By imposing restrictions, the hypothesis space is narrowed down, leading to a more focused and accurate model.
27.
Greska estimacije (varijansa) nastaje usled:
Correct Answer
B. Cinjenice da je klasa hipoteza H prebogata
Explanation
The correct answer suggests that the error of estimation (variance) occurs because the hypothesis class H is too rich or complex. This means that the hypothesis class has too many possible hypotheses, making it difficult to accurately estimate the true underlying relationship between the input variables and the output variable. A rich hypothesis class can lead to overfitting, where the model fits the training data too closely and fails to generalize well to new, unseen data.
28.
Klasterovanje je:
Correct Answer
B. Grupisanje primera po slicnosti (najslicniji pripadaju istoj grupi - klasteru)
Explanation
Klasterovanje je proces grupisanja primera na osnovu njihove slicnosti. Prilikom klasterovanja, slični primeri se grupišu zajedno u isti klaster, dok se različiti primeri grupišu u različite klaster. Ova tehnika se koristi u oblastima kao što su analiza podataka, mašinsko učenje i prepoznavanje obrazaca. Klasterovanje omogućava otkrivanje prirodnih grupa i struktura u podacima, što može biti korisno za razumevanje i analizu podataka.
29.
Opsti oblik neparametarske procene funkcije gustine verovatnoce (k-broj uzoraka unutar volumena V, N je ukupan broj uzoraka, V je volumen koji okruzuje uzorak X) je:
Correct Answer
B. P(X)=k/(N * V)
Explanation
The given correct answer is P(X) = k/(N * V). This equation represents the general form of a nonparametric estimation of the probability density function. In this equation, k represents the number of samples within the volume V, N is the total number of samples, and V is the surrounding volume of the sample X. This equation shows that the probability density function is estimated by dividing the number of samples within the volume by the product of the total number of samples and the volume.
30.
Greska aproksimacije (bajas) se:
Correct Answer
C. Smanjuje sa prosirivanjem klase hipoteza H
Explanation
The error of approximation (bias) decreases with the expansion of the hypothesis class H. This means that as the hypothesis class becomes larger and more flexible, it can better fit the underlying data distribution, resulting in a lower approximation error.
31.
Vrste masinskog ucenja su:
Correct Answer(s)
C. Induktivno (nadgledano)
D. Samoobucavanje (nenadgledano)
E. Obucavanje sa pojacavanjem
G. Semi-induktivno
Explanation
The given answer includes different types of machine learning. "Induktivno (nadgledano)" refers to supervised learning, where the model is trained using labeled data. "Samoobucavanje (nenadgledano)" refers to unsupervised learning, where the model learns patterns and structures from unlabeled data. "Obucavanje sa pojacavanjem" refers to reinforcement learning, where the model learns through trial and error and receives feedback in the form of rewards or penalties. "Semi-induktivno" refers to a combination of supervised and unsupervised learning, where the model is trained with both labeled and unlabeled data.
32.
U tradicionalnom programiranju ulaz su:
Correct Answer
A. Podaci + program
Explanation
In traditional programming, the input consists of both data and the program that operates on that data. This means that the program takes in the data and performs operations or calculations on it to produce an output. This is the most common approach in programming, where the program is designed to manipulate the given data according to a set of instructions or algorithms.
33.
Uslov konvergencije (N tezi beskonacno) neparametarske procene funkcije gustine verovatnoce oblika P(X)=k/(N*V) je:
Correct Answer(s)
A. K raste zavisno od N
D. V se skuplja sa povecanjem N
Explanation
The correct answer is "k raste zavisno od N, V se skuplja sa povecanjem N." This means that as the sample size N increases, the parameter k also increases, and the variability V decreases. This is because as the sample size increases, there is more information available to estimate the true probability density function, leading to a more accurate estimate (higher k). Additionally, as the sample size increases, the spread of the data decreases, resulting in a smaller variability (smaller V).
34.
Koja teorema masinskog ucenja daje stav o postojanju univerzalnog sistema obucavanja:
Correct Answer
A. Teorema "Nema besplatnog rucka"
35.
Pravilo klasifikacije na osnovu K najblizih suseda glasi: ???
Correct Answer
C. X pripada onoj klasi kojoj pripada vecina njenih najblizih k suseda
Explanation
The correct answer is "X pripada onoj klasi kojoj pripada vecina njenih najblizih k suseda". This is because the classification rule based on K nearest neighbors states that an instance belongs to the class that is the majority among its K nearest neighbors. In other words, the class of an instance is determined by the most frequent class among its K closest neighbors.
36.
Da li postoji univerzalan sistem obucavanja?
Correct Answer
C. Ne
37.
Sta su slobodni parametri metoda histograma? ???
Correct Answer
D. Dimenzionalnost podataka
Explanation
The free parameters of a histogram method refer to the variables that can be adjusted or chosen by the user. In the context of this question, the dimensionality of the data is considered as a free parameter. This means that the user can decide how many dimensions or features are included in the data when constructing the histogram. The dimensionality of the data can have a significant impact on the results and interpretation of the histogram.
38.
VC dimenzija klase hipoteza H je: ???
Correct Answer
D. Minimalan broj tacaka koji moze biti 'razbijen' svakom hipotezom iz H
Explanation
The VC dimension of a hypothesis class H is the minimum number of points that can be shattered by any hypothesis in H. This means that there exists at least one set of points of that size that can be labeled in all possible ways by a hypothesis in H. Therefore, the correct answer is "Minimalan broj tacaka koji moze biti 'razbijen' svakom hipotezom iz H" which translates to "The minimum number of points that can be shattered by any hypothesis in H."
39.
Preferencijalni bajas u masinskom ucenju ima znacenje:
Correct Answer
A. Uvodjenje uredjenja u prostor hipoteza po nekom kriterijumu i izbor one hipoteze u procesu obucavanja koja ekstremizuje ovaj kriterijum
Explanation
Preference bias in machine learning refers to the introduction of ordering in the hypothesis space based on some criterion and selecting the hypothesis during training that maximizes this criterion. This means that the preference bias aims to prioritize certain hypotheses over others based on a specific criteria, ultimately selecting the hypothesis that best satisfies this criterion during the training process.
40.
Sta je u PAC teoriji empirijska greska (rizik):
Correct Answer
B. Greska na trening skupu
Explanation
In PAC theory, the empirical error (risk) refers to the error that occurs on the training set. This means that the model is not able to accurately predict the correct output for the examples in the training set. The empirical error is used to estimate the generalization error, which is the error that occurs on unseen data. By minimizing the empirical error, we aim to minimize the generalization error and improve the performance of the model on new data.
41.
Medicinska dijagnostika (od simptoma ka bolestima) je primer:
Correct Answer
C. Klasifikacije
Explanation
The correct answer is "Klasifikacije." Medicinska dijagnostika involves classifying symptoms into specific diseases or conditions. This process helps healthcare professionals in accurately identifying and treating illnesses based on the symptoms presented by the patient. Therefore, klasifikacije (classification) is an appropriate example for this scenario.
42.
U modelu stabla odlucivanja, kompleksnost modela:
Correct Answer(s)
A. Raste sa porastom dubine stabla
B. Raste sa porastom brojem listova
Explanation
In a decision tree model, the complexity of the model increases with the increase in the depth of the tree because more decision nodes and branches are added. Similarly, the complexity also increases with the increase in the number of leaves because more rules need to be generated to classify the instances accurately. Therefore, both options stating that the complexity increases with the increase in the depth of the tree and the number of leaves are correct.
43.
Ako je VC dimenzija klase hipoteza H, u oznaci VC(H), jednaka beskonacnosti, tada vazi:
Correct Answer
A. VC dimenzija nije relevantna za pvp pitanje
Explanation
When the VC dimension of hypothesis class H is infinite, it means that the class H can shatter any finite set of points. In this case, the VC dimension is not relevant for the question of whether the hypothesis class H is learnable or not. Hence, the correct answer is "VC dimenzija nije relevantna za pvp pitanje" which translates to "VC dimension is not relevant for the pvp question".
44.
U neparametsrkom pristupu, relevantne funkcije gustine verovatnoce se procenjuj ????
Correct Answer
C. Tako sto se prvo pretpostavi da pripadaju nekoj parametarskoj familiji funkcija
Explanation
In nonparametric approach, the relevant probability density functions are estimated by first assuming that they belong to some parametric family of functions.
45.
Trening skup sluzi za:
Correct Answer
B. Obucavanje modela
Explanation
The correct answer is "Obucavanje modela." This is because "trening skup" refers to the training set, which is used to train or teach the model. The training set is used to optimize and adjust the model's parameters so that it can accurately predict or classify new data. Therefore, the purpose of the training set is to train the model, making this the correct answer.
46.
Obucavanje sa uciteljem (ili nadgledano - supervised) moze biti:
Correct Answer(s)
B. Induktivno
C. Transduktivno
Explanation
Induktivno obucavanje sa uciteljem koristi primere podataka kako bi se izvukli opsti zakljucci i pravila. U ovom slucaju, ucitelj pruza primere i model se uci na osnovu tih primera. S druge strane, transduktivno obucavanje sa uciteljem koristi primere podataka kako bi se naucilo kako klasifikovati nove, nepoznate podatke. U ovom slucaju, ucitelj pruza primere i model se uci kako klasifikovati slicne primere u buducnosti. Deduktivno i agnosticko obucavanje sa uciteljem nisu navedeni kao moguci odgovori.
47.
Kada se duzina obucavajuceg skupa neograniceno povecava, greska klasifikacije po metodu najblizeg suseda P_KNN1 se nalazi u sledecem odsnosu prema gresci P_Bajes optimalnog Bajesovog odlucivanja:
Correct Answer
D. P_KNN1 < 2P_Bajes
Explanation
As the length of the training set increases without limit, the classification error of the nearest neighbor method P_KNN1 will be less than twice the classification error of the optimal Bayesian decision P_Bajes.
48.
Kompleksnost polinomijalnih regresionih modela:
Correct Answer
A. Raste sa porastom reda polinoma
Explanation
The complexity of polynomial regression models increases with the increase in the degree of the polynomial. This means that as the order of the polynomial increases, the model becomes more complex and requires more computational resources to fit the data accurately. Higher order polynomials have more parameters to estimate, which can lead to overfitting and decreased generalization performance. Therefore, the complexity of polynomial regression models is directly proportional to the degree of the polynomial.
49.
Ukoliko klasa hipoteza H ima konacnu VC dimenziju, tada su tacni sledeci stavovi:
Correct Answer
B. H je PAC obuciva kasa
Explanation
The given correct answer states that "H je PAC obuciva kasa" which translates to "H is PAC learnable class". This means that the class of hypotheses H can be learned using the Probably Approximately Correct (PAC) learning framework. PAC learning refers to the ability to learn a concept with high probability and a small error, given a sufficient amount of training data. Therefore, the statement implies that the class H can be effectively learned using algorithms that minimize empirical risk.
50.
Test skup sluzi za:
Correct Answer
B. Procenu tacnosti izabranog i obucenog modela
Explanation
The correct answer is "Procenu tacnosti izabranog i obucenog modela" which translates to "Evaluating the accuracy of the selected and trained model." In machine learning, it is crucial to assess the performance and effectiveness of a model. The evaluation process helps determine how well the model predicts outcomes or classifies data. By using various metrics and techniques, such as cross-validation or confusion matrices, the accuracy of the model can be estimated. This evaluation is essential for understanding the model's strengths and weaknesses and making improvements if necessary.