Jargon is the set of terms and expressions that have defined meaning within a field but lack that meaning (or perhaps any meaning) outside that field. Computing’s jargon is primarily based on English words with computing-specific meanings which may or may not be related to the English meanings of those words.

On this page are terms I have defined in class and which I expect you to understand.

Abstract Data Type

Some kinds of information can be stored in multiple ways, with sometimes nuanced difference in how the different data structures that store it perform. An abstract data type provides an abstraction barrier, allowing algorithms to be designed ignoring the details of the data structure chosen.

Practically, an abstract data type consist of a set of named algorithms (i.e. functions) that perform specific tasks on the data. For example, an abstract data type representing a queue of things waiting their turn might have functions for add to back of line and remove from front of line. And data structure for which those functions can be created is said to implement the abstract data type.

Abstraction

An essential tool in organizing large programs. Consists of generalization of many related actions (often with parameters) and detail removal to guide focus and simplify action (often with functions).

Abstraction barrier

Abstraction strong enough that we can use the abstraction without thinking about the details it hides.

No abstraction barrier is perfect: there are always limitations of nuances which do matter in some cases. For example, Ariane flight V88 exploded and a Pariot Missle hit the army that fired it both because the abstraction barrier between bits and numbers failed.

Algorithm

Unambiguously-defined steps to accomplish some task.

Algorithms are generally described in terms of the problem they solve (or approximately solve in the case of heuristics), the data used to encode that problems inputs, and the steps used to compute the answer.

The word algorithm is a Latinization of the name of Muḥammad ibn Mūsā al-Khwārizmī who wrote books that Europeans adopt the ideas of algorithms with place-value numbers. Algorithms themselves long predate his work.

Analog

Often used to mean simply not digital, or the part of the world that is not computers.

Sometimes used to mean a signal with a meaning based on being analogous to (directly or naturally representing, not using symbols but more intrinsic properties like quantity or frequency) the thing is describing.

Arguments

Concrete values provided to a function or program in place of its parameters.

The square root function has one parameter, the number to find the root of. In $\sqrt{2}$, the number $2$ is the argument to that function.

Artificial Intelligence (AI)

A field within Computer Science that works on algorithms that humans find to be intelligent. Exactly what that means changes over time. Currently, it is seen to consist of

• Things that can fool a human into thinking the computer is human, like chat.
• Planning, picking actions based on an evolving and uncertain future, like playing games.
• Summarizing, turning high-dimension inputs into low-dimension outputs, like naming what is in an image.
• Machine learning.

Assembly

A textual representation of machine instructions. Also sometimes called assembly code or assembler code.

Backwards compatibility

A new system is backwards compatible with an old system if data created by/for the old system works without modification in the new system.

Base

Place-value number systems are defined by a single integer called their base. The base tells us

• How many digits the system uses (including 0)
• The value multiple for moving a digit one place to the left

The most popular bases are:

• Base 2, also called binary with digits called bits, used to build computer processors.
• Base 10, also called decimal, used by humans in most modern cultures.
• Base 16, also called hexadecimal or hex, used by computing professionals to communicate binary values more easily.
• Base 256, with digits called bytes, used to build computer memory, storage, and communication devices and protocols.

Binary

Multiple meanings:

• The base-2 number system.
• A program that is ready to be run directly on a processor without further translation.
• Any computer file containing non-textual information.

When it is necessary to distinguish that a number is written in binary, it is traditional to precede it with 0b; thus 0b1101 means a binary number with digits 1101 and represents the value thirteen.

Bit

A single digit in base 2: either 0 or 1.

The word bit is a portmanteau of binary digit.

Byte

A group of 8 bits. Equivalently, a group of 2 hex digits. Equivalently, a number between 0 and 255 (inclusive).

When it is necessary to write out a byte for human viewing, it is traditional to write it as two hex digits; thus 0D means a byte with the value thirteen and F3 means a byte with the value two hundred forty-three.

Call

A colloquialism for invoking a function.

Categorical

A family of AI tasks where the output is one of a discrete set of options (labels, group membership, words, etc.), not continuous real numbers.

Classical AI

Artificial intelligence where humans designed the algorithms based on their understanding of the problem and how to solve it. Often contrasted with machine learning, though many successful AI systems use a mix of both.

Classification

A family of AI tasks where the output is categorical and the set of possible outputs is known during training. This is generally supervised learning.

Examples might be pick the next word or tell which person is shown in this image.

Clustering

A family of AI tasks where the output is categorical and the set of possible outputs is not specified but instead learned by the AI system. This is a form of unsupervised learning.

Examples might be organize these images into groups of similar images – the number of groups and what kinds of images belong in each is learned, not provided in the training data.

Compiler

A program that converts source code written in a programming language into a sequence of machine instructions that can be executed by a specific type of computer processor.

Typically, programming languages that use a compiler provide fewer abstractions and thus require more effort to program in than those that use an interpreter; but they also tend to create programs that run more quickly than those created for an interpreter.

Continuous

A family of AI tasks where the output is one or more real numbers (a score, a grid of pixel intensities, etc.), not categorical data..

Data

A defined, unambiguous representation of information.

Ways of representing simple information like numbers are commonly called data types or primitive data types. Ways of representing multipart information like lists are commonly called data structures.

Data Structure

A way of storing data that consists of multiple related but distinct pieces of information.

Most data structures are designed to implement some abstract data type. For example, the data structure array is implemented by putting one value after another in memory and is one implementation of the abstract data type list. List could also be implemented with other data structures, such as by having each element in the list store both a value and the location in memory to find the element that comes after it.

Detail Removal

One of two common forms of abstraction.

The most common form of detail removal is to refer to a complicated set of actions by name instead of by the actions themselves. Functions are the primary tool for that form of detail removal.

Another important form of detail removal is an abstract data type, which helps use build algorithms that do not depend too closely on the specifics of how we represet data.

Digital

Represented or operating using a set of discrete symbols called digits, rather than using continuous or analog signals.

Because of the prevalence of programmable general-purpose digital computers, digital is sometimes used as a synonym for done with computers.

Execute

Formally: load the machine instructions of a compiled program into memory and have the processor begin following those instructions.

Informarlly, also used for running an interpreter on a program’s source code or, less often, for invoking a function.

Exploit

A concrete use of a vulnerability, in most cases to enact crime.

The possibility that an attacker could take control of my machine using a flaw in one of its systems is a vulnerability. The specific tools they use to do so is an exploit.

Float

Short for floating-point number, a float is the binary version of scientific notation stored as three parts:

• • or ‒
• An exponent
• The rest of the number (the significand or mantissa)

For example, 56⅔, which in binary is 111000.1010101010101010…, stored as a float with 8-bit significand would be +1.1100010 × 2¹⁰¹ (or in decimal as 1.765625 × 2⁵).

Function

A tool in programming to name a set of actions and use them later by name.

A function definition defines the actions and the name used to refer to them. Many function definitions include parameters.

A function invocation requests that the actions in a previously-defined function be performed. If the function definition included parameters, the function invocation includes arguments to use for those parameters this time.

The phrase call a function or function call is synonymous with invoke a function or function invocation, but with a stronger verbal abstraction barrier, referring to the function as if it were some external entity you call up instead of a set of actions you take.

Gate

An abstraction of a group of transistors that implement some simple operation on a few bits.

The most commonly discussed gates are and, or, not, and xor. For some specific hardware cases, nand and nor are also used.

General-purpose

One of the key qualifiers of a computer, indicating that the computer hardware is capable of doing any information-processing task that any other computer can do.

The Church-Turing thesis posits the existence of general-purpose computers, and of related Turing-complete systems and forms of expression that are capable of describing the operation of general-purpose computers. While not formally proven, this thesis is widely accepted as true.

One consequence of the fact that almost all computers are general-purpose is that new computer technologies (like quantum computers) might differ from others in efficiency and speed, but not in the set of tasks they could eventually do, given unlimited time and storage.

Generalization

One of two common forms of abstraction.

The most common form of generalization is by adding parameters to a function. For example, $\sqrt{x}$ is a one-parameter function that finds a square root; $\sqrt[y]{x}$ is a two-parameter function that generalizes $\sqrt{x}$ to also compute other roots.

Another common form of generalization is to add parameters to a data structure; for example by defining a generic list of $x$ structure instead of a specific list of integers structure. Parameterized data structures have more variation in how they are implemented between programming languages than do parameterized functions, leading to them having many names: generics, templates, type parameters, mixins, etc.

Generative AI

A family of AI tasks that aim to complete a pattern, filling in the missing parts. Because this is generally trained on complete patterns, it is a form of supervised learning. It can be either categorical (such as when picking the missing word in a piece of text) or continuous (such as when picking the missing pixel color in an image).

GHz

The SI abbreviation for giga meaning one billion (G) followed by the abbreviation for Hertz (Hz) which means per second, together meaning one billion per second. Commonly used to measure the clock speed of processors, roughly meaning how many billion arithmetic problems the processor can complete per second.

Heuristic

An algorithm that (usually) only approximates the solution to a problem rather than finding the full solution.

Heuristics are more often applied to problems that seek the best of many options, and return not the best option possible, but a pretty good option.

Most heuristics return the optimal output for some inputs but not for most and may return quite poor outputs for a few particularly bad inputs.

Hex

A common abbreviation for hexadecimal.

Hexadecimal

A group of 4 bits. Equivalently, a number between 0 and 15 (inclusive).

Hexadecimal digits are typically represented as 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 (with the same meanings they have in decimal) and A, B, C, D, E, F (meaning ten through fifteen in the order listed). Capitalization of the letters is not important:

When it is necessary to distinguish that a number is written in hexadecimal, it is traditional to precede it with 0x; thus 0x10E means a hexadecimal number with digits 10E and represents the value two hundred seventy.

Hill climbing

A technique used to train a machine learning model by

1. Picking a random function
2. Analyzing in what direction each parameter might need to change for that function to better match the training data
3. Picking a new function with parameters moved in that direction
4. If the new function is not good enough, return to step 2

Information

Usually, information is used in a loose, human way to refer to the human-understood meaning carred by data.

The information that I am 20 years old could be conveyed by various data: the text I’m 20, the bits 00010100 in a location where age values are expected, the difference between the current date and a birth date, etc.

Less often, information is used as a technical term for the bits of some digital communication that could not be anticipated, an important concept in designing efficient networking, compression, and storage technologies.

Interpreter

A program that reads the source code of a programming language and takes the actions described by each line of that code.

Typically, programming languages that use an interpreter run much more slowly than those that use a compiler; but they also tend to provide more abstractions and thus require less effort to program in than those created for a compiler.

Machine instructions

A set of simple actions a processor can take, encoded in a compact binary format. Common machine instruction types include:

• Arithmetic operations, like $x + y → z$.
• Operations that copy data between locations.
• Jumps that continue execution at a new location in the set of machine instructions, often only if certain conditions are met.

Machine instructions can be directly converted to and from assembly. Compilers generate machine instructions from source code, but that operation generally cannot be reversed.

Machine learning (ML)

A form of programming that works by

1. Defining a general family of functions or algorithms, where a specific function from the family can be selected by setting one or more parameters.
2. Searching though the function family for one that comes closest to matching some training data.
3. Using that function to solve problems.

Metadata

Data about the (main) data.

For example, in an image file the color of each pixel is the main data; information about the camera that took the image is metadata.

Many file formats store metadata along with the data, and many applications ignore most of the metadata.

Offline learning

A type of machine learning where the training happens separately from and before the function selected during training is used.

Offline learning is easier to design than online learning and thus is the default for many machine learning applications.

Online learning

A type of machine learning where new training data is collected and used to change the function being used after some use of the function.

Online learning is harder to design than offline learning and thus is generally limited to cases where the applicaiton needs to start acting before it has all the training data, such as when it needs to learn from the outcomes of earlier decisions it has made.

Overfitting

A common way that machine learning might fail, caused by the function family having so many parameters that the function selected matches the details of the specific training data used rather than the underlying patterns the training data was supposed to exemplify.

The most common safeguard against overfitting is to ensure there are many fewer parameters than there are training data. Other techniques pick function families that are not flexible enough to conform to the training data’s details or add mathematical criteria to the training process that detects and avoids parameters that appear to be overfitting.

Overflow

Most numbers in computers are represented in a fixed number of bits. If math would produce a number that requires more bits than that, the number overflows. What happens after overflow depends on the kind of number:

• For integers, the most common is to simply discard the digits that won’t fit; so 10011 stored in 4 digits would become 0011.

• For floats, when the exponent overflows the number changes into a special infinity value. It loses its ability to engage in most math, but is still clearly a big number.

Parameter

A placeholder, often implemented as a variable, to achieve generalization.

An algorithm to compute $\sqrt{2}$ has no parameters: each time it is used it creates the same result. An algorithm to compute $\sqrt{x}$ has one parameter, $x$: to use the algorithm we have to provide an argument, a value to use in place of $x$ for this run of the algorithm; and for each $x$ we provide the algorithm will create a different result.

Pretraining and Fine-tuning

A method of training an AI system in several steps.

The pretraining step typically uses as much training data as possible to learn broad patterns without overfitting.

The fine-tuning step moves the function selected during pretraining closer to a smaller, more desireable subset of the training data.

Fine-tuning can be repeated to further adjust the learned function.

Problem

Often used as it is in English, to refer to anything that’s not going well or needs to be addressed.

Also used as a desired the relationship between inputs and outputs defining the goal of an algorithm. Find if a number is odd is a problem; write it in binary and see if its low-order bit is 1 is one of many algorithms for solving that problem.

An algorithm solving a problem needs to pick what data to use to represent the input and output information. An implementation of the algorithm as a function will have the input data as its parameters.

Programmable

One of the key qualifiers of a computer, indicating that the specific computations it performs is not fixed when it is created but rather can be specified as inputs to the computer.

Usually, programmable is also assumed to mean that the program it follows is stored inside the computer’s memory and can be an entire algorithm, not just one of a machine-defined set of operations selected by simple switches and buttons.

Regression

A family of AI tasks that predict one or more values based on other values. This is continuous (the categorical version is called classification) and supervised (because the training data has the value that should be predicted.

Regression is an older topic than AI and is widely studied in statistics.

Repetition

A component of programs that indicates that some set of actions should be performed repeatedly. Also called iteration or looping.

The most common programming language vocabulary for repetition are for and while.

Run

A colloquialism meaning to execute a program or invoke a function.

Selection

There are two types of selection.

One, also called conditionals, jumps, or branches, is a component of a program that indicates that some set of actions should only be performed under certain conditions and skipped if those conditions do not hold. The most common programming language vocabulary for this type of selection are if and else.

The other, also called indexing, is a component of a program that indicates which one out of several candidate pieces of data should be used. The most common programming language vocabulary for this type of selection are the name of the set of candidates, followed by which candidate in the set to chose in brackets, like some_list[3] to pick item 3 out of a list.

Supervised learning

A category of machine learning where the training data includes the desired output for each input.

TB

The SI abbreviation for tera meaning one trillion (T) followed by the abbreviation for Byte (B). Commonly used to measure the disk space of computers, indicating they can store a trillion bytes of data.

Some of that space is typically used to store an index of sorts to help find the rest of the data, so the usable space is generally lower than the advertised, theoretically-provided space.

Train

The step in machine learning where a function is selected that matches the training data. In some cases training can be done exactly using some mathematical properties of the function family, but more often some type of guess-and-check or hill climbing is employed.

Training data

Example data that an algorithm solving a problem should be able to process.

For supervised learning, training data pairs each input with an expected output and the quality of the function is determined by how close the function’s output is to that input.

For unsupervised learning, training data is just input and some other property of output (typically how the outputs on different inputs compare) determined the quality of the function.

Type

How the bits inside some piece of data are converted into meaning.

For example, the bits 01100110 could have many meanings. If they are interpreted with the data type integer they mean 102. If they are interpreted with the data type UTF-8 character they mean lower-case F (f). The bits (the information we actually store in the computer) have not changed, only how we interpret it.

Unsupervised learning

A category of machine learning where the training data does not include desired outputs, only a set of example inputs.

Variable

The name of a location where data can be stored.

Unlike in mathematics, variables in computing are expected to change their meaning as a program progresses. It is perfectly OK for a program to say x = 3 on one line and x = 4 on another.

Most programming languages use = not to mean is the same as but rather gets the value of, in a directional way. x = 3 + 4 means the memory named by variable x should have the number 7 stored inside it. Conversely, 3 + 4 = x is a nonsensical an erroneous request that the number 7 have the contents of the memory named by variable x put inside it.

Vulnerability

Some aspect of a computer system – usually software, but sometimes hardware – that has the potential to be used in ways that the designers did not intend and the legitimate users do not desire.

Vulnerabilities can arise from many sources: programmer error, software designer oversight, unexpected interplay between individually-correct components, failure to remove development aids like bypass codes, malicious insertion by software developers with secret agendas, and so on.

It is generally assumed that nearly all computer systems contain many vulnerabilities, and new vulnerabilities are discovered every day. It is essential that software be regularly updated to close vulnerabilities as they are discovered to prevent them being targeted by exploits.