Concepts starting with B

166 concepts, each with a plain-language explanation, examples and diagrams. Pick a letter or open a concept.

Background Computation with Threads — A shared resource is an object that more than one thread uses, such as the image object in the source's example.
Backpropagation — Tensors provide the multi-dimensional representations used by neural networks.
Backpropagation and the computational graph — torch.no_grad() is a context manager that tells PyTorch not to track gradients, which saves memory and computation when a model…
Backpropagation and Weight Updates — Identical starting weights make nodes in a layer perform the same operations and receive the same updates, so they never learn…
Backpropagation in Neural Network Layers — The embedding layer is a lookup operation: a token ID selects a row of the weight matrix by position.
Backward Elimination — Greedy selection evaluates feature subsets through the learning algorithm rather than separating selection from learning.
Backward-view temporal-difference learning — The on-line λ-return algorithm motivates this topic because it combines strong performance with a complex forward-view…
Balancing and Preprocessing the Dataset — Undersampling balances a dataset by reducing the majority class until it matches the minority class, using random selection.
Balancing Exploration and Exploitation with Upper-Confidence-Bound Action Selection — UCB combines an estimated action value with an uncertainty bonus.
Base and Derived Classes in C++ — Default copy and move operations are problematic in a class hierarchy because the base class does not know about the members of…
Base Cases in Recursive Algorithms — A recursive binary search has two base cases: an empty range, which returns -1, and a value found at the middle position, which…
Base Classes and Their Role — A class hierarchy is a set of classes ordered in a lattice created by derivation, and it is used to represent concepts that have…
Base Hypothesis Class — The base class B provides the individual hypotheses used as building blocks.
Basic for loop syntax — Loops scan through data by following a three-part structure: initialize a tracking variable, compute on each item, and inspect…
Basic if Statements — An if-else statement creates two mutually exclusive branches: one executes if the condition is true, the other if it is false.
Basic Mathematical Functions — Python's trigonometric functions (sin, cos, tan) take arguments in radians, not degrees
Basic Operators in Python — Operator precedence is a set of rules that determines which operations Python performs first when multiple operators appear in…
Basic Output and Formatted Output — System.out.print(x) outputs x after converting it to a string when necessary.
Basic Recursion and the Call Stack — A non-empty linked list splits into a head node and a tail, and the tail is itself a shorter linked list.
Basic Types in C++ — Arithmetic operators can be used with appropriate combinations of types, not with every combination by default.
Basic while loop syntax — Loops scan through data by following a three-part structure: initialize a tracking variable, compute on each item, and inspect…
Batch Monte Carlo Methods — The maximum-likelihood model is built from observed transition frequencies and average observed rewards.
Batch Processing in LLM Training — A PyTorch training run repeats a core update cycle: reset gradients, calculate loss, perform a backward pass, and update weights.
Batch Processing in Neural Network Training — Hyperparameters are settings chosen before training, and a validation dataset is the held-out data used to find the best of them.
Batch Size and Gradient Accumulation — There is no universal optimal learning rate, so you find a workable one by experimenting within the typical range and reading…
Batch Size and Its Effect on Training — Stride is the number of positions by which the input window shifts when the next batch is created from an input dataset.
Batch Size and Training Stability — Batch size is a tradeoff: small batches give noisy updates, and it is a hyperparameter to experiment with when training LLMs.
Batch Size Trade-offs in Training — The source suggests testing the data loader with max_length=2 and stride=2, and with max_length=8 and stride=2, to build…
Batch Updating — Batch updating keeps all episodes seen so far and repeatedly presents that growing batch until convergence.
Batching and Training Loops in PyTorch — The DataLoader class from torch.utils.data produces batches from a dataset for training, validation, and testing.
Batching in Inference — Online APIs optimize for latency and process requests as they arrive, while batch APIs optimize for cost and can accept longer…
Batching Input Data for Language Models — CausalAttention extends self-attention with a causal mask and a dropout layer, and it accepts batched inputs of shape (batch…
Batching Instruction Examples with a Custom Collate Function — Instruction masking replaces the instruction and input tokens with the -100 mask, and Exercise 7.2 asks you to evaluate whether…
Batching Text Data for Training — Padding extends each input shorter than the target length until it reaches the maximum number of tokens the model supports, so…
Batching Text for Language Model Training — A round trip of encode followed by decode reveals which words the vocabulary can represent.
Bayes Optimal Classification — 1-NN is analyzed as a sample-dependent hypothesis h_S for binary classification.
Behavioral Training — Shaping uses reinforcement to guide behavior through successive approximations.
Bellman Equations — v*(s) is the maximum expected return from a particular state under an optimal policy.
Bellman Optimality Equation for q* — v* assigns each state its maximum expected return.
Benchmark Correlation — A public leaderboard ranks models on a small subset of benchmarks, so its rank does not comprehensively evaluate a model or…
Benchmark Selection — A public leaderboard ranks models on a small subset of benchmarks, so its rank does not comprehensively evaluate a model or…
Benchmarking Foundation Models — Evaluation mitigates risks and uncovers opportunities across model selection, benchmarking, deployment readiness, and production…
Benchmarks and Saturation in Model Evaluation — Open-ended tasks can have many correct responses, so a complete answer key is impossible to build and exact-match comparison is…
Benefits and Applications of the Three-Phase Frame — The three-phase frame is useful both for the day-to-day work and as an organizational tool.
Benefits and Trade-offs of Heuristic Search — Heuristic search focuses on the current state and its likely successors.
Benefits of Class Hierarchies — Implementation inheritance means a base class provides functions or data that simplifies the implementation of derived classes.
Bernoulli Variables — A Chernoff-bound problem begins with independent Bernoulli variables and their sum.
Best Practices for Numerical Computations — Numerical problems are often subtle, so you should be certain about their mathematical aspects before relying on a result.
Best Practices for Strings and Regular Expressions — Returning a string by value means the function sends the string itself back to its caller as the result.
Best Practices for Template Programming — Templates are used to express algorithms that apply to many argument types.
Best Practices for Time Measurement — std::chrono is the subnamespace of the C++ standard library that holds the standard facilities for dealing with time-related…
Best Practices for Using C++ Libraries — The guidance 'don't reinvent the wheel; use libraries' recommends using pre-existing libraries instead of recreating…
Best Practices for Using STL Algorithms — STL algorithms do not directly add or subtract elements from their argument sequences, so the sequence size is unchanged by the…
Bias Mitigation in Language Models — Finetuning improves a model's quality on a specific task and can teach it to follow particular output structures.
Bias-Complexity Tradeoff — Model selection is the task of choosing an algorithm for a particular problem and setting its parameters.
Biases in User Feedback — Implicit feedback is inferred from actions users already take, such as choosing an image option or accepting a code suggestion…
Bibliography and References in Machine Learning — Three important references are associated with neural network learning, pattern recognition and machine learning, and support…
Big O Notation — Computational complexity measures the number of operations an algorithm performs.
Binary Classification — X names the collection of objects being studied, while Y names the collection of possible labels.
Binary Search — A recursive binary search has two base cases: an empty range, which returns -1, and a value found at the middle position, which…
Binary Search Algorithm — A search compares the item with the root, then discards either the left or right subtree, and repeats on the remaining subtree.
Binary Search on Sorted Arrays — A binary sort tree is a binary tree in which every node is at least as large as everything in its left subtree and no larger…
Binary Tree Structure and Terminology — A subtree is a node together with all of its descendants, and it is itself a binary tree.
Binding Function Arguments with std::bind — A random number generator in the C++ standard library is created by combining an engine, which is the source of randomness, with…
Binomial Distribution — Slud's inequality converts a specified binomial probability into a normal-tail lower bound.
Bipartite Ranking — A linear predictor for ranking is represented through the output y′ = h_w(x̄).
Bitset Operations and Access — A bitset can be initialized with an integer or a string, using the brace form bitset<N> bs {value};
Bitwise Operators on Integers — The complement operator ~ flips every bit of a bitset, and the other bitwise operators such as & combine bitsets one position at…
Blackjack Policy Evaluation — Monte Carlo ES turns simulated Blackjack episodes into a search for a strong policy.
Blocking Operations and the GUI Thread — ConnectionHandler is a subclass of Thread that opens the network connection and reads incoming messages, keeping blocking…
Blocking Queues — wait() and notify() are instance methods of class Object, so any object can coordinate two threads
Blocks and Compound Statements in Java — The if statement takes the form if ( boolean-expression ) statement-1 else statement-2.
BNF Grammars and Syntax Rules — A recursive descent parser models every BNF grammar rule as a subroutine, so the structure of the program follows the structure…
Boolean conjunctions — A Boolean conjunction maps X = {0, 1}^n to Y = {0, 1}.
Boolean Expressions and Comparison Operators — Most statements (logical lines) that you write will contain expressions . A simple example of an expression is 2 + 3 . An…
Boolean Expressions in Java — Boolean operators form complicated conditions from boolean values.
Boolean Logic and Conditions — Sequential execution performs statements one after another in the order they appear in the script.
Boolean Logic and Short-Circuit Evaluation — Defensive programming means asking 'What could go wrong?' before your program reaches users.
Boolean Operators in Java — Java's relational operators are ==, !=, <, >, <=, and >=.
Boolean Values and Comparison Operators — The and operator returns True only when both conditions are true; use it when you need multiple requirements to be satisfied…
Boolean Values and Expressions — A boolean expression is any expression that evaluates to True or False; a condition is a boolean expression inside a conditional…
Boolean Values and Logic — An if statement evaluates a condition and executes a block of code only if that condition is true.
Boolean Values and Operators — The in operator is a boolean operator that returns True if a substring is found in a larger string, and False otherwise.
Boosting Weak Learners — The task is binary face recognition: classify an image as a human face or not.
Boosting: Improving Weak Learners — Weak learnability allows a learning algorithm to be useful even when its accuracy is only slightly better than random guessing.
Bootstrapping in Temporal-Difference Learning — TD(0) and Monte Carlo can disagree because they use different evidence from the same episodes.
Bootstrapping Methods — SGD variations are used to find a good weight vector for function approximation.
Bottleneck Analysis in Operational Processes — A claims processing workflow is a chain of steps: opening the claim, gathering documentation, writing a narrative summary…
Branches of Machine Learning — Supervised learning connects inputs with known targets through a collection of examples.
Branching Statements — A while loop is a control structure for condition-based repetition.
Breaking Down Complex Problems — Problem analysis identifies gaps and ambiguities in a problem statement before you begin coding.
Breaking Down Problems into Steps — Programming is a creative partnership where humans provide ideas, creativity, and judgment while computers provide speed…
Breaking Out of Loops with break — The else block is executed when the while loop condition becomes False - this may even be the first time that the condition is…
Breaking Out of Loops: Using break and continue — An infinite loop never terminates because its condition never becomes false.
Building a Basic Tokenizer — The tokenizer splits text with re.split at punctuation, the double dash, and whitespace, keeping matched punctuation through a…
Building a Catalog of Reusable Work — A slice library is a moat for solo operators and small shops and a highway for larger teams, and it is valuable in both settings…
Building a Content Database from Email Archives — Raw email data contains inconsistencies (mixed-case addresses, subdomain variations, gateway artifacts) that must be…
Building a Convolutional Neural Network — Max pooling replaces each local window with its maximum value.
Building a Custom Model Leaderboard — Benchmark results help you identify promising models, but there are too many benchmarks to examine them all, so selection is…
Building a FeedForward Module — The FeedForward module maintains the shape of the input tensor, so the output has the same dimensions as the input.
Building a GPT Model Architecture — Initializing the 124M GPT model and printing its output tensor gives a shape of [2, 4, 50257], where 50257 is the tokenizer's…
Building a GPT Model from a Configuration Dictionary — BASE_CONFIG holds the settings shared by every GPT-2 size, including vocabulary size 50257, context length 1024, dropout rate…
Building a GPT-Style Model from Transformer Blocks — The section's overall goal is to code a GPT-like large language model that can be trained to generate human-like text.
Building a Growth Mindset for Programming — Programming mastery, like language fluency, requires time and patience to absorb and internalize concepts.
Building a Language Model from Scratch — A large language model is a deep neural network designed to understand, generate, and respond to humanlike text.
Building a Large Language Model — Emergent behavior is the ability of a model to perform tasks it was not explicitly trained to perform.
Building a Minimal BPE Tokenizer from Scratch — Byte Pair Encoding is a tokenization method used in natural language processing, including in GPT-2.
Building a Platform for Forward Deployed Engineers — An FDE is a customer-facing software engineer who deploys a technical product for a customer, and the role is justified only in…
Building a Portfolio with Anthropic Plugins — Anthropic plugin templates are enterprise templates introduced at Code with Claude in 2026, spanning seven named industry areas.
Building a PyTorch module for layer normalization — Layer normalization adjusts the activations of a layer so they have a mean of 0 and a variance of 1, which the source says…
Building a Real Transformer Block — DummyGPTModel is a simplified GPT-style nn.Module. It builds token and positional embeddings, dropout, a stack of transformer…
Building a Reliable Evaluation Pipeline for Open-Ended Applications — Foundation models are open-ended and powerful, which makes them harder to evaluate than traditional ML models, and investment in…
Building a Reusable Slice Library — A slice library is a collection of reusable code slices accumulated from past engagements and meant to be adapted for new…
Building a Simple Tokenizer with encode and decode — The encode method converts a text string into a list of integer token IDs, and the allowed_special parameter tells it which…
Building a Simple Web Client — When you call encode() on a string, Python looks up each character and converts it to one or more bytes according to the UTF-8…
Building a Vocabulary from Text — A tokenizer class implements encode and decode, and it holds a vocabulary that maps text tokens to integer IDs along with an…
Building a Vocabulary from Training Text — A tokenizer's vocabulary can be extended with special context tokens that carry meanings ordinary words cannot express.
Building a Vocabulary from Unique Tokens — Special tokens such as '<|endoftext|>' and '<|unk|>' are added to the token list after the regular unique words, using extend().
Building a Web Scraper with Error Handling — Regular expressions operate on text as a linear sequence, while HTML is hierarchical. This fundamental mismatch causes regex…
Building a Web Server with Sockets — A socket connection to port 80 establishes a reliable communication channel with a web server through a three-way handshake.
Building AI Systems Incrementally — A minimal end-to-end slice is the smallest unit of work that forms a complete, end-to-end demonstration, and it is the first…
Building and Consuming APIs — A web service is a set of services in an application's API that are made available over the web via HTTP.
Building and Maintaining Competitive AI Products — Milestone planning turns a measurable goal into a sequence of intermediate targets, and the right sequence depends on where the…
Building and Using Large Language Models — Custom-built LLMs are tailored to specific tasks or domains and can outperform general-purpose LLMs within them.
Building Applications with Foundation Models — A Foundation Model is one component inside an application, and the application surrounds it with the logic that defines the task.
Building Bespoke Layers for Competitive Advantage — AI has lowered the time and effort needed to build bespoke workflows, so many previously marginal workflows are now worth…
Building Complex Models with the Keras Functional API — The Functional API describes a model as a connected route of tensor transformations.
Building Complex Patterns with Alternation and Grouping — Raw strings (r'...') are mandatory for regex patterns because they prevent Python from interpreting backslashes as escape…
Building Custom Neural Networks in PyTorch — PyTorch's deep learning utilities provide building blocks for creating custom deep neural networks.
Building Deployment Templates — A demo succeeds when it creates desire that leads to a contract, which means showing one valuable outcome clearly rather than…
Building Dictionaries from Text — Convert a dictionary to a list of (value, key) tuples so that sorting operates on the values (frequencies) rather than the keys…
Building Foundation Model Applications — A low entry barrier makes AI applications easy to build, and therefore easy for competitors to copy
Building Histograms from Data — Character counting uses a dictionary to store each unique character and its count.
Building Histograms with Dictionaries — The get() method safely retrieves dictionary values, returning a default if the key is missing.
Building Interactive Maps with OpenStreetMap and Leaflet — geodump.py converts database records into executable JavaScript code by extracting location names, latitudes, and longitudes and…
Building LLMs from Scratch — The source material for this article does not describe the cover artwork, so the article teaches the historical context given in…
Building Lookup Tables — Dictionaries outperform lists and variables for character counting because they only allocate space for characters that actually…
Building Mental Models of Code — Programming is a language that requires gradual absorption over time, not linear mastery of isolated concepts.
Building Network Applications — Web scraping is a program-based technique for retrieving web pages and extracting specific data from their HTML source code…
Building Network Applications with Python — curl and wget are command-line tools for retrieving files from the web on Unix-like systems using HTTP or FTP protocols.
Building Network Applications with Sockets — HTTP responses follow a three-part structure: status line, headers, blank line, and body.
Building Network Applications: From HTTP to Web Scraping — A socket is a two-way connection between two programs that allows both reading and writing of data, similar to a file but…
Building Normalized Email Databases with SQLite — Three Python scripts (gbasic.py, gword.py, gline.py) each answer a different question about email participation: who sent the…
Building Panels and Containers in Swing — A grid layout can only have one component in each position, so a row that needs two pieces cannot place both directly into one…
Building Programs with Variables, Expressions, and Statements — input() pauses program execution and waits for keyboard input from the user.
Building REST APIs with JSON — JSON has become the industry standard for data exchange because its structure maps directly onto native data structures in…
Building Swing Components — A layout manager is a java.awt class passed to setLayout() to decide how a container arranges its components.
Building Your First API Request — An API is an application-to-application contract that publishes rules for accessing services provided by one program for use by…
Building Your First C++ Program — Comments in C++ are written for the human reader, and the compiler ignores them.
Building Your First GUI Application — Python bindings are translation layers that allow Python code to use GUI libraries written in C/C++, combining Python's…
Building Your Own Evaluation Benchmarks — A top rank on a general public leaderboard does not guarantee the best model for your application, so start by matching…
Building Your Own Objects — Web scraping demonstrates object orchestration: urllib fetches HTML, BeautifulSoup parses it into objects, and your code…
Built-in arrays and pointer conversion — An array is a fixed-size sequence of elements of a single given type, with the number of elements specified at compile time.
Built-in Arrays and Pointer Decay — An array is less flexible than a vector but simpler, and it can give a performance advantage when its elements are placed…
Built-in Classes for Dialog Messages — Event handling is the process of handling events in a GUI application, and it can be implemented in different ways.
Built-in Exception Types — Custom exceptions are user-defined classes that inherit from Exception, enabling you to represent errors specific to your…
Built-in Functions and Their Types — Defining a function creates a variable that holds a function object with type 'function'.
Built-in Functions for Inspection — The dir() function reveals all capabilities (methods and attributes) of any Python object by returning a list of strings.
Built-in Functions: A Reference Guide — Built-in function names like max, len, sum, and print should be treated as reserved words and never used as variable names.
Built-in Functions: max() and min() — Collect all numbers from the user into a list using a while loop that terminates when the user enters a sentinel value like…
Built-in Subroutines and Functions — A subroutine is a named set of program instructions designed to perform a task.
Built-In Types and Standard Strings in I/O — The iostream library handles input and output for built-in types, standard strings, and programmer-defined types.
Built-in Types in C++ — A concrete class is a class designed so that its objects behave just like built-in types.
Byte Pair Encoding (BPE) Tokenizer — Tokenization breaks raw text into tokens, which can be words or characters, and then converts those tokens into integer token IDs.
Byte Streams and Binary Data in Java — InputStreamReader and OutputStreamWriter wrap an existing byte stream so that a program can read and write characters instead of…
Byte Streams with OutputStream and InputStream — DataOutputStream writes typed values such as int and double into a stream in internal, binary-number format, and it is used when…