According to Mitchell (1997) Machine Learning is... Is the area of science concerned with the study of the management of methods, techniques and processes for the purpose of storing, processing and transmitting information and data in digital form. A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P , improves with experience E. The theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
In the Classification task... The computer program is asked to predict a numerical value given some input. The computer program is asked to specify which of k categories some input belongs to. The computer program is asked to specify which of k categories some input does not belong to.
When doing the Classification task... The output is a continous variable. Becomes more challenging with missing inputs. Becomes less challenging with missing inputs.
In the Regression task... The computer program is asked to predict a numerical value given some input. The computer program is asked to specify which of k categories some input belongs to. The computer program is asked to predict a discrete value given some input.
Define the type of tasks of the image: Regression and Clustering Regression and Classification Classification and Regression.
What is true about the Clustering task: The computer program is asked to predict a group for some given input. The input data is not labelled. Both are true.
Which examples are not machine learning tasks? Structured output and anomaly detection. Machine traslation, denoising, synthesis and sampling. Transcription, Imputation of missing values and structured input.
What is true about the performace measure P? This doesn't determine how well it will work when deployed in the real world. The Accuracy measure is always reliable. Are specific to the task T being carried out by the system.
In which set are Performance Measures use? In the training set In the test set In the result set.
Mark the correct statement: The choice of performance measure is always straightforward and objective. There are some performance measures that works properly with all the algorithms. Is often difficult to choose a performance measure that corresponds well to the desired behavior.
The elements of machine learning are... E (Experience), T (Task) and P (Performance measures) E (Experience), T (Train) and P (Performance) E (Experience), T (Task), P (Performance) and R (Results).
When each example is associated with a label or target, the experience is... Supervised learning Unsupervised learning Semisupervised learning.
K-means algorithm is: Reinforcement learning Supervised learning Unsupervised learning.
Autoencoders are exaples of: Reinforment learning Unsupervised learning Self-supervised learning.
In machine learning Generalization is... The ability to perform well on training inputs. The ability to perform well on previously observed inputs. The ability to perform well on previously unobserved inputs.
In machine learning the following sets are used: Train set, test set and experienced set. Result set, train set and test set. Validation set, train set and test set.
The generalization error is estimated... In the test set. In the training set. In both sets.
Underfitting occurs when... The gap between the training error and test error is too large (The model is unable to predict well, very limited to the training data). The model is not able to obtain a sufficiently low error value on the training set (The model is unable to learn). It can happend in both cases.
Overfitting occurs when... The model is not able to obtain a sufficiently low error value on the training set (The model is unable to learn). The gap between the training error and test error is too large (The model is unable to predict well, very limited to the training data). Both are incorrect.
This model is: Overfitted Underfitted Good fitted.
This model is: Overfitted Underfitted Good fitted.
Select the incorrect one about the Capacity of a model: Is defined as the ability of a model to fit a wide variety of functions. You can control whether a model is more likely to overfit or underfit by altering it. Models with low capacity will predict very well.
A model with a high capacity: Can overfit by memorizing properties of the training set that do not serve them well on the test set. May struggle to fit the training set Both are not correct.
The no free lunch theorem for machine learning states that: No machine learning algorithm is universally any better than any other. There is a machine learning algorithm that will be the best solution for all the problems. Jon Lasa es un miron.
Hyperparameters are... Settings that we can use to control the algorithm’s behavior. Settings that we can use to control the algorithm’s inputs. Settings that we can use to control the regularization on the data.
Which statement is not true about the "curse of dimensionality": Mechanisms used to achieve generalization in traditional machine learning are insufficient to learn complicated functions in high-dimensional space (Thats why DL) As the number of data dimensions increases the complexity of the algorithm also increases. As the number of data dimensions increases the complexity of the algorithm also decreases. .
Put it in the correct order: 1-Choosing the Training Experience, 2-Choosing the Training Function, 3-Choosing a Representation of the Training Function, 4-Choosing a Function Approximation Algorithm and 5-The Final Design. 1-Choosing the Training Function, 2-Choosing the Training Experience, 3-Choosing a Representation of the Training Function, 4-Choosing a Function Approximation Algorithm and 5-The Final Design. 1-Choosing the Training Experience, 2-Choosing the Training Function, 3-Choosing a Function Approximation Algorithm, 4-Choosing a Representation of the Training Function and 5-The Final Design.
The main diference between direct or indirect training is: With indirect training you receive a feedback from every state, whereas with direct training you only get the feedback from the final outcome (apart from the sequence of movements). With direct training you receive a feedback from every state, whereas with indirect training you only get the feedback from the final outcome (apart from the sequence of movements). That indirect training is generally easier than direct training.
One-shot and zero-shot learning are used when: When the data is unlabelled. When there is too many data to process. If there are not any examples for one class.
Give an posible example of a target function used in a chest problem: WinGame ChooseMove Palo.
When talking about The Final Design... Performance System, Critic, Generalizer and Experiment Generator are essential. Performance System is not relevant. Performance System and Experiment Generator are most essential elements.
In the machine learning process... Is important to preprocess the data before doing all the steps. Is important to preprocess the data after doing all the steps. Is important to do the model deployment the before doing all the steps.
In the data preprocessing is important to mention that low-quality data will lead to low-quality results. Which well-known phrase supports this: First in, first out (FIFO) Last in, first out (LIFO) Garbage in, garbage out (GIGO).
Which one is not a data preprocessing technique: Data cleaning Data integration Data elimination.
Data preprocessing techniques... Are not mutually exclusive Are mutually exclusive Always work together.
Regarding the data quality... Is essential to have accuracy, completeness, timeliness, believability, interpretability and consistency. Is essential to have accuracy, concurrence, timeliness, believability, and interpretability and consistency. Real-world databases always offer good quality data.
Which one is not a case of having innacurate data: Data collection instruments used may be faulty or errors in data trasmission. Users may purposely submit incorrect data values for mandatory fields when
they do not wish to submit personal information (disguised missing data). There may have been human or computer errors occurring at data entry. All are correct.
Which one is not a case of incomplete data: When the data is nos considered important at the time of entry. When some atributes of interest are not available, such as customer information for sales transaction data. When the data that were inconsistent with other recorded data may have been deleted. All are correct.
In data cleaning when dealing with missing values we cannot the method... Ignore the tuple. Fill in the missing value manually. Binning.
The noisy data... Is the sound in a measured variable. May represent outliers when applaying statistical description techniques (e.g., boxplots and scatter plots).
We cannot use smoothing techniques to try to del with noise.
Which technique is more popular when dealing with missing values: Use a global constant to fill in the missing value Use the attribute mean or median for all samples belonging to the same class as the given tuple Use the most probable value to fill in the missing value (f.e. using regression).
Select the incorrect statement about the data cleaning: Binning methods smooth a sorted data value by consulting its neighborhood. Data smoothing can also be done by classification. Outliers may be detected by clustering.
Data integration deals with: Entity identification, Redundancy reduction, Correlation analysis and Data Value Conflict Detection and Resolution. Entity identification, Outlier reduction, Correlation analysis and Data Value Conflict Detection and Resolution. Entity identification, Outlier reduction, Validation analysis and Data Value Conflict Detection and Resolution.
Data numerosity reduction techniques... PCA is an example of this Can be parametric or nonparametric. Parametric methods include histograms, clustering, sampling, and data cube aggregation. Are directly applicable to any machine learning problem.
Which one is not a dimensionality reduction technique: Discrete wavelet transform (DWT) Principal Components Analysis Feature Subset Selection Data cube aggregation All are correct.
Principal Components Analysis... Searches for k n-dimensional octogonal vectors that can best be used to represent the data, where k <= n Is important to normalize the input data. Combines the essence of attributes by creating an alternative, same-size set of variables.
Mark the incorrect one: You can reduce a 1000 dimension problem into a two dimension. Principal components may be used as inputs to multiple regression and cluster analysis In PCA the components are sorted in increasing order of “significance”, which is the strength or influence of the dimension on the data PCA tends to be better tha Wavelet Transform at handling sparse data.
What is a greedy algorithm? An algorithm that makes the locally optimal choice at each stage without any backtrack. An algorithm that makes the locally optimal choice at each stage using backtracking. An algorithm that makes the optimal choice taking into account the previous stage..
Why are greedy algorithms use in Feature Subset Selection? Because they always make what looks to be the best choice in every stage taking into account the previous stages. Because they are specialised in the most optimal solution An exhaustive search for the optimal subset of features can be prohibitively expensive, especially as n and the number of data classes increase.
The data are transformed or discretizated into forms appropriate for analyzing. Which one is not a transformig technique: Smoothing (binning, regression, and clustering) Aggregation Standarization Aproximation.
Normalization, select the false one: To help avoid dependence on the choice of measurement units, the data should be normalized Normalization is specially useful for nearest-neighbor classification and clustering. The range normalization is [-1, 1] or [0.0, 1.0]. All are correct.
To compute the Feature Subset Selection we can use: SequentialFeatureSelector (SFS), RFE-selector and VarianceThreshold selector (this more oriented to clustering). RFE-selector and VarianceThreshold selector (this more oriented to clustering). SequentialFeatureSelector (SFS) and VarianceThreshold selector (this more oriented to regression).
|
