4. Machine Learning Basics Flashcards
What is a machine learning algorithm?
An algorithm that is able to learn from data
This definition emphasizes the ability to adapt based on data inputs.
How does Mitchell (1997) define learning in a computer program?
A computer program learns 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
This definition highlights the relationship between experience, tasks, and performance.
What are the three components involved in the definition of learning by Mitchell?
- Experience E
- Tasks T
- Performance measure P
These components interact to define the learning process in machine learning.
Fill in the blank: A computer program is said to learn from experience ______.
[key learning term] E
What improves with experience E according to Mitchell’s definition?
Performance at tasks in T
This improvement is measured by the performance measure P.
True or False: The book attempts to formally define experiences, tasks, and performance measures.
False
Instead, the book provides intuitive descriptions and examples.
What is the focus of the following sections in the book regarding machine learning?
Intuitive descriptions and examples of different kinds of tasks, performance measures, and experiences
This approach aims to clarify the concepts without formal definitions.
What enables us to tackle tasks that are too difficult to solve with fixed programs?
Machine learning
Machine learning allows for adaptive solutions rather than static programming.
From what perspectives is machine learning considered interesting?
Scientific and philosophical
It helps us understand the principles underlying intelligence.
In the context of machine learning, what is the distinction between ‘task’ and ‘learning’?
‘Task’ refers to the activity to be performed, while ‘learning’ is the means to attain the ability to perform the task
For example, walking is the task, and learning to walk is the process.
What is an example in machine learning?
A collection of features that have been quantitatively measured from some object or event
Examples are crucial for training machine learning models.
How is an example typically represented in machine learning?
As a vector x∈Rn
Each entry xi of the vector corresponds to a feature.
What do the features of an image usually represent in machine learning?
The values of the pixels in the image
Pixel values serve as features for image processing tasks.
True or False: In machine learning, the process of learning is considered the task.
False
The task is the activity to be performed, while learning is a process to achieve that activity.
Fill in the blank: Machine learning tasks are usually described in terms of how the machine learning system should process an _______.
example
This processing is essential for training and making predictions.
What is classification in machine learning?
A task where the computer specifies which of k categories some input belongs to
What function does a learning algorithm produce for classification tasks?
f : Rn →{1,…,k}
In classification, what does y = f(x) represent?
The model assigns an input described by vector x to a category identified by numeric code y
What is a variant of the classification task?
When f outputs a probability distribution over classes
Give an example of a classification task.
Object recognition
What is the input for an object recognition task?
An image described as a set of pixel brightness values
What does the output of an object recognition task represent?
A numeric code identifying the object in the image
What is the capability of the Willow Garage PR2 robot?
To recognize different kinds of drinks and deliver them to people on command
What technology is modern object recognition best accomplished with?
Deep learning
What is a practical application of object recognition technology?
Automatically tagging people in photo collections
True or False: Object recognition technology can enable computers to interact more naturally with their users.
True
What makes classification more challenging in the presence of missing inputs?
The computer program cannot guarantee that every measurement in its input vector will always be provided.
What must the learning algorithm define when inputs are missing?
A set of functions corresponding to classifying with different subsets of inputs missing.
In which field does the issue of missing inputs frequently arise?
Medical diagnosis.
What is one method to define a large set of classification functions efficiently?
Learn a probability distribution over all relevant variables and marginalize out the missing variables.
How many different classification functions can be obtained with n input variables?
2^n different classification functions.
What does the computer program need to learn to handle missing inputs?
A single function describing the joint probability distribution.
True or False: Classification with missing inputs is a unique example of machine learning applications.
False.
Fill in the blank: Classification with missing inputs is just one example of what ______ can do.
machine learning
Who provides an example of a deep probabilistic model applied to classification with missing inputs?
Goodfellow et al. (2013b)
What is regression in machine learning?
A type of task where the program predicts a numerical value given some input.
The learning algorithm outputs a function f : Rn →R.
How does regression differ from classification?
The format of output is different; regression outputs numerical values while classification outputs categories.
Both are types of supervised learning tasks.
Give an example of a regression task.
Prediction of the expected claim amount for insurance premiums or future prices of securities.
These predictions are also used in algorithmic trading.
What is transcription in machine learning?
The task of transcribing unstructured data into discrete textual form.
Examples include optical character recognition and speech recognition.
What is an example of transcription?
Optical character recognition where a photograph of text is converted into ASCII or Unicode format.
Google Street View uses this technology to process address numbers.
What role does deep learning play in speech recognition?
It is a crucial component of modern speech recognition systems.
Major companies like Microsoft, IBM, and Google utilize deep learning for this purpose.
What is machine translation?
The task where a sequence of symbols in one language is converted into a sequence of symbols in another language.
Commonly applied to natural languages, such as translating English to French.
How has deep learning impacted machine translation?
Deep learning has recently begun to have an important impact on this kind of task.
Notable advancements have been made by researchers like Sutskever et al. and Bahdanau et al.
What is the definition of structured output tasks?
Tasks involving output as a vector or data structure with important relationships between elements
What broad category do structured output tasks include?
Transcription and translation tasks, among others
What is an example of structured output tasks in natural language processing?
Parsing a natural language sentence into a tree structure
What does parsing involve?
Mapping a natural language sentence into a tree by tagging nodes as verbs, nouns, etc.
What is an example of structured output tasks in image processing?
Pixel-wise segmentation of images
What is the purpose of pixel-wise segmentation?
Assigning every pixel in an image to a specific category
How can deep learning be applied to aerial photographs?
By annotating the locations of roads
What is image captioning?
A task where a program observes an image and outputs a natural language sentence describing it
What is required for the output of image captioning?
The words must form a valid sentence
Why are these tasks called structured output tasks?
Because the program must output several interrelated values
