Concepts starting with E

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

Earlier NLP Models and Task-Specific Design — LLMs are AI models that can understand, generate, and interpret human language, where understanding means producing coherent…
Echo and Delta: Two Halves of One Craft — Founder-phase skills are the same skills the FDE role requires, and the labor market values them more highly inside the FDE…
Echo Mode: Listening Before Building — Echo mode listens and learns how the work actually happens; Delta mode ships a working system based on that learning.
Economic Viability of Go-to-Market Motions — The viability of the FDE motion depends on a company's average contract size, not on the motion's general appeal.
Effective Contract Negotiation for AI Projects — A contract's structure, especially how payments are scheduled and released, determines whether the engagement is aligned with…
Effective Demo Techniques for Customer Engagement — In customer control, the customer holds the keyboard and types the inputs they would use in their actual work.
Effective Internal AI Deployment Teams — A team's FDE title and placement describe its intended role but do not guarantee how it operates.
Effective Size — A class with small effective size enjoys the uniform convergence property.
Effective Thread Programming — A shared resource is an object that more than one thread uses, such as the image object in the source's example.
Efficient Action-Value Estimation — An average can be updated incrementally rather than recomputed from all previous rewards.
Efficient Computation of Masked Attention Weights — Information leakage means future tokens influence the current token, and causal attention must prevent it.
Efficient Implementation of Multi-Head Attention — Multi-head attention extends causal attention across several heads, each operating independently with its own weights.
Efficient Implementation of NN Rule — Approximate nearest neighbor search improves performance by allowing a bounded approximation instead of requiring an exact result.
Efficient learnability in the realizable case — A Boolean conjunction maps X = {0, 1}^n to Y = {0, 1}.
Eigenvalue Decomposition — Non-invertibility can arise when training instances do not span the entire space of R^d.
Eigenvalues — A symmetric matrix is positive definite when every eigenvalue is positive.
Eigenvectors and Eigenvalues — Matrix A uses X transposed X and has dimensions determined by the original features.
Element-wise Product — The tensor dot operation combines entries in its input tensors.
Eligibility Traces in Reinforcement Learning — Miller's 1981 Law-of-Effect-like rule included synaptically local contingent eligibility traces.
else-if (elif) Statements — The if statement is used to check a condition: if the condition is true, we run a block of statements (called the if-block )…
Embedded Input Vectors — A context vector is the weighted sum of all input vectors, where each input is multiplied by its attention weight for one chosen…
Embedding Input Tokens for Attention — Self-attention creates three views of each input token by multiplying it with three trainable weight matrices: W_q for queries…
Embedding Layers and Weight Matrices — A token embedding is retrieved by looking up the row of the embedding layer's weight matrix that matches the token ID, with rows…
Empirical Loss — The PAC-Bayes bound can be converted into a learning rule.
Empirical Risk — Approximation error measures the gap between the best hypothesis in a class and the true hypothesis.
Empirical Risk Minimization — This lesson orients you to the workflow of this course.
Empirical Risk Minimization for Linear Classifiers — The Perceptron learns by correcting one currently misclassified sample at a time.
Encapsulation and Access Control — A class is a template or blueprint; an object is a concrete instance created from that class.
Encapsulation and Data Hiding — Procedure-oriented programming centers on functions that manipulate data, keeping data and logic separate; object-oriented…
Encapsulation: Bundling Data and Behavior — Abstraction is the ability to hide complexity so you can focus on what matters to you and ignore the rest.
Encapsulation: Hiding Implementation Details — Large programs cannot be held entirely in a programmer's mind at once, making it impossible to manage them as a single…
Encoder and Decoder in the Transformer — The self-attention mechanism lets a transformer weigh the importance of different words or tokens in a sequence relative to each…
Encoder-Decoder Models for Sequence Generation — The Bahdanau attention mechanism lets the decoder selectively access different parts of the input sequence at each decoding step.
Encoder-Decoder Models for Translation — RNNs can handle short sentences, but they struggle as inputs get longer.
Encoding and Decoding Text with a Tokenizer — SimpleTokenizerV2 handles words outside the vocabulary by replacing each one with the special <|unk|> token during encoding.
Encoding Input Text and Decoding Output: Tokenization and Token IDs — The tokenizer's .decode method converts a sequence of output token IDs back into text, and the decoded string is what gets…
Encoding Text into Token IDs — The output tensor from the model has three dimensions: batch, one row per input token, and embedding values per token.
Engagement Agreements and Governance — The section ENGAGEMENT AGREEMENT - OUTLINE is an outline of the structure or content of an engagement agreement.
Engagement Phases — The third phase of an engagement is the one that proves you can do the work without the shadow, meaning without the accompanied…
Engagement Scope and Schedule Documents — The engagement scope is defined by a signed scoping document attached as Schedule A, so the scope is read through that document…
Engagement Stages — The frame is also a diagnostic: first name the stage an engagement is in, then check that stage's condition.
Engineering Discipline for Agentic Systems — The first version of an AI feature slice is a literal sequence of API calls in a single file, with no framework, abstraction, or…
Enhancing B2B SaaS Sales with AI-Powered Pre-Call Research — AEs skip pre-call research mainly because the time between calls is too short to absorb the fifteen-minute research process…
Enterprise AI Adoption and Deployment — Forward-deployed engineering means embedded engineers working on-site with customers to implement and deploy software solutions.
Entity-Relationship Diagrams and Data Modeling — A one-to-many relationship connects one entity instance to multiple instances of another entity.
Entity-Relationship Diagrams: Visualizing Database Structure — A many-to-many relationship exists when records on both sides can connect to multiple records on the opposite side, unlike…
Entity-Relationship Models — Crow's Foot Diagrams use boxes to represent tables and lines with notation symbols to represent relationships between tables.
Environment Dynamics — The central question is what model of the environment is available to the agent.
Environment Models in Reinforcement Learning — Cognitive maps and environment models are representations of environments, described respectively in psychological and…
Environment Variables in Windows — The PATH variable is a list of directories that Windows searches to find executable programs; adding Python's directory to PATH…
Episode Returns and Step Rewards — The racetrack is a discrete environment in which both position and velocity are represented using grid-based values.
Episodes and Time Steps in Reinforcement Learning — MDP notation distinguishes individual objects from sets of possible objects.
Episodes and Transitions in Reinforcement Learning — Models let reinforcement-learning systems simulate experience rather than relying only on directly observed experience.
Episodes, Returns, and State-Value Estimates — Backup diagrams show what information feeds an update to a root state node.
Episodic and Continuing Tasks — Episodic tasks have interactions that naturally divide into separate episodes.
Episodic Semi-gradient Control Methods — A learning curve is a way to represent performance over time.
Episodic Semi-gradient Sarsa for Control — A learning curve is a way to represent performance over time.
Episodic Tasks — Continuing tasks describe agent-environment interactions that do not naturally divide into identifiable episodes.
Episodic Tasks in Reinforcement Learning — The on-policy distribution describes how state probability is distributed under a policy.
Epsilon-Greedy Action Selection — On the 10-armed testbed, UCB generally performs better than epsilon-greedy action selection.
Equality and Object Identity — Sorting requires a defined comparison, because there is no natural ascending order for arbitrary objects.
equals() Method and Object Comparison — hashCode() is defined in the Object class, returns an int, and every Java object has one.
Equation (16.4) — Soft-SVM with kernels is the focus of the section.
Equity Grants and Notional Value in Early-Stage Companies — The Founder-FDE arbitrage is the gap that appears when the same skills are compensated differently in a founder role than in an…
ERM Hypothesis — An ERM hypothesis is any hypothesis consistent with all past examples.
Error Decomposition — Selecting features determines the dimensionality of the feature space.
Error Handling and API Response Codes — An API is a published contract between applications that specifies the exact rules for requesting and receiving services.
Error Handling and Exception Prevention — Python evaluates logical expressions left to right and stops early when the result is determined (short-circuit evaluation)
Error Handling and Exceptions — Files are opened using open(filename, mode), where mode controls read/write/append permissions and what happens to existing data.
Error Handling and File Operations — Closing files is not optional when writing data; it is the mechanism that flushes buffered data to disk.
Error Handling and Input Validation — math.log10() computes base 10 logarithms, essential for decibels, pH scales, and other common logarithmic scales
Error Handling and Retry Logic in API Calls — Rate limits are a deliberate constraint on API usage designed to protect server resources and ensure fair access across users.
Error Handling in Network Applications — Binary files must be accumulated in a bytes buffer and cannot be printed as they arrive like text files.
Error Handling in Network Code — recv() returns the amount of data currently available in the network buffer, which may be less than the requested amount and…
Error Handling in Python — ET.fromstring converts an XML string into a tree structure, transforming flat text into a navigable, queryable object.
Error Handling in Scripts — Verbosity levels (-v and -q flags) let users control how much information a script outputs, making it suitable for different…
Error Handling in Stack Operations — A stack is a sequence of items piled one on top of the other, and only the top item is accessible at any time.
Error Handling in Web Scraping — BeautifulSoup transforms raw HTML strings into searchable tree-structured objects that mirror the document's logical hierarchy.
Error Handling Strategies — The Guardian Pattern strategically places a guard condition before a risky operation in an 'and' expression to prevent runtime…
Error Handling When Files Do Not Exist — Opening a file asks the operating system to locate the file by name and verify it exists and is readable.
Error Handling with try and except — The computegrade function takes a score as a parameter and returns a letter grade as a string, using if and elif conditions to…
Error Handling with Try-Except — Defensive programming means asking 'What could go wrong?' before your program reaches users.
Error Handling with Try-Except Blocks — Short-circuit evaluation stops evaluating a logical expression once the result is already determined.
Error Messages and Debugging — exit() terminates the entire program immediately and never returns to any caller.
Error Messages and How to Read Them — Testing and debugging are essential phases where you verify that your program works and fix errors that prevent it from working…
Error of a Classifier — A prediction rule maps a domain point to a label.
Escalating Requests to Sponsors — Pushback from IT, security, and data teams can be incorrect, and the errors tend to follow predictable patterns.
Escape Sequences — Escape sequences are two-character combinations (backslash plus another character) that tell Python to interpret the second…
Escape Sequences in Python Strings — The newline character (\n) is a single invisible character that represents the end of a line, despite appearing as two symbols…
Escape Sequences in Regular Expressions — Square brackets [...] define a set of acceptable characters for a single position in a regex pattern.
Escaping Special Characters in Regular Expressions — The period (.) in a regular expression is a wildcard that matches any single character—letters, digits, spaces, punctuation, or…
Estimated Action Values — Exploration chooses a nongreedy action to improve knowledge about its value.
Estimating Action Values with Monte Carlo Methods — A deterministic policy repeatedly selects the same action from a given state.
Estimation Error — A hypothesis class is the set of possible classifiers from which the learning algorithm chooses.
Euclidean Distance — Random projections transform a vector using a random matrix.
Eval Discipline: Matching Checks to Case Difficulty — In AI engineering, the eval set is the specification: a collection of input-output pairs, sometimes with rubrics, that defines…
Eval-Driven Validation in AI Projects — Eval-driven validation judges a model with a labeled eval set before the model is deployed to production.
Evals as the foundation of agent evaluation — Agent readiness is a measure of how many deterministic validation loops are present inside a codebase.
Evaluating a Classifier During Training — train_classifier_simple nests a batch loop inside an epoch loop, and each batch resets gradients, computes loss, backpropagates…
Evaluating a Convnet — The original Dogs vs. Cats dataset contains 25,000 color JPEG images.
Evaluating a Fine-Tuned LLM — Instruction fine-tuning is a three-stage process: preparing the dataset, fine-tuning the LLM, and evaluating the LLM.
Evaluating a First Version Before Adding Structure — The first version of an AI feature slice is a literal sequence of API calls in a single file, with no framework, abstraction, or…
Evaluating a Loaded GPT Model — BASE_CONFIG holds the shared GPT-2 settings, and BASE_CONFIG.update copies the chosen size's emb_dim, n_layers, and n_heads into…
Evaluating AI Outputs with LLM-as-Judge — LLM-as-judge works with caveats: GPT-4 class judges reach roughly 80 percent agreement with human raters, which is approximately…
Evaluating AI Products with Eval Suites — The per-trace loop starts by reading individual sampled traces with their actual input and actual output, rather than the…
Evaluating AI Systems Across Usage Contexts — An aggregate metric drop cannot by itself tell you whether a regression is systemic or confined to one slice of input.
Evaluating AI Systems in Production — The delivery phase deploys a developed system to a live environment, and it goes well when scoping and validation have been done…
Evaluating AI Systems: A Shift in Engineering Mindset — With a specification, the engineer's job is to produce a deterministic system that exactly matches the spec.
Evaluating and Implementing the Pre-Call Briefing Agent — Adoption, perceived usefulness, and business outcome are different signals, and a pilot needs all three to make a credible case.
Evaluating Candidates for Forward Deployed Roles — Hiring managers evaluate FDE candidates against three criteria: a worthy product ambition, total alignment with the customer's…
Evaluating Career Risk in Early-Stage Companies — A high-variance bet is a career choice with substantial potential upside and substantial potential downside, such as working at…
Evaluating Classification Models — Evaluating a classification model involves calculating the classification accuracy.
Evaluating Contract Review Output with Evals — The project is organized into a README, a prompts folder, a src folder, an evals folder, and a traces folder, and each part has…
Evaluating Contract-Review Systems — A representative evaluation set is built from real-world examples that reflect the mix of work a system will actually face.
Evaluating Customer-Facing AI Systems — Start every scope discussion by restating the agreed baseline and its delivery date, so the existing commitment stays protected.
Evaluating Fine-Tuned Classifiers — Instruction fine-tuning aims at flexibility and interaction quality across many complex user instructions, and it demands larger…
Evaluating Fine-Tuned Language Model Responses — Larger models such as gpt2-medium (355M) are more computationally intensive than gpt2-small (124M), because they have more…
Evaluating Fine-Tuned Large Language Models — Installation depends on your operating system: on macOS and Windows you open the Ollama application and select Yes for…
Evaluating Generative Text Models — Pretraining is stage 2 of coding an LLM and includes the training loop, model evaluation, and saving and loading model weights.
Evaluating Large Language Model Responses — Fine-tuning adjusts a pre-trained LLM to a specific task, here converting active sentences to passive voice, using a single…
Evaluating Large Language Models — The chapter's plan has three parts: stage 1 covers data preprocessing and the attention mechanism, stage 2 covers coding and…
Evaluating LLMs — The attention mechanism is described in the source as a fundamental component at the heart of every LLM, so it anchors the rest…
Evaluating Machine Learning Models — A machine learning problem is defined by its inputs, predicted outputs, and problem type.
Evaluating Model Output Against a Labeled Evaluation Set — A high completeness score only shows that required fields are present; it does not show that the values are correct.
Evaluating Model Performance for Spam Classification — Classification accuracy is the performance measure a practitioner reports, but it is not differentiable, so a gradient-based…
Evaluating Model Performance on Realistic Test Sets — Lead with the bad news directly, because the customer needs the headline before any context.
Evaluating Multi-Agent Systems — Ask questions about the customer's needs before you propose or reject a solution, because the answers decide which design fits.
Evaluating Postfix Expressions with a Stack — A stack-based evaluation algorithm can detect errors in a postfix expression without needing parentheses or a human reader.
Evaluating Response Quality of an Instruction-Tuned Model — The training and validation loss curves reveal how learning progressed across the run.
Evaluating Trained Models — Supervised learning trains a model on a labeled dataset, pairing each example with its correct label.
Evaluating Vendor Deployments — A vendor's use of the FDE label or a role name does not show that it is doing FDE work.
Evaluating Vendors for Forward Deployed AI — Every working FDE function has failed engagements, so a vendor who cannot describe one is either inexperienced or dishonest.
Evaluation Criteria for Proposed Systems — The validation phase consists of building evaluation criteria, instrumenting them, running the proposed system against them on a…
Evaluation Functions in Game-Playing Programs — Samuel's evaluation function judged checkers positions and guided the program's decisions.
Evaluation Set Design for AI Models — An evaluation set is a living artifact that decays over time if it is not maintained, a failure the source calls eval-set rot.
Evaluations as a Starting Point for Agent Work — Agent readiness is a measure of how many deterministic validation loops, such as linters, type checkers, tests, and security…
Event Handling in Graphical User Interfaces — A state machine is something that has a state and can change state in response to events or inputs, and its response to an event…
Event Handling in GUI Programs — MosaicDraw.java demonstrates menus and a color chooser dialog in a program where the user colors mosaic squares by clicking and…
Event Listeners and ActionEvents — The javax.swing Timer class generates a sequence of ActionEvents separated by a specified interval.
Event Listeners and Registration — Mouse motion events are handled through the MouseMotionListener interface, which declares mouseDragged and mouseMoved.
Event-Driven Programming in GUI Applications — Python bindings are translation layers that allow Python code to use GUI libraries written in C/C++, combining Python's…
Event-Driven Programming in GUIs — A GUI program is responsive when it responds to events very soon after they are generated.
Every-Visit MC — First-Visit MC keeps only the return after a state's first occurrence in each episode.
Evolution of Reinforcement Learning — Reinforcement learning grew from two independently pursued historical directions.
Evolution of User Interfaces — A command-line interface is defined by typed commands and typed computer responses.
Evolutionary Methods — A policy gradient method searches over parameter-defined policies rather than treating a value estimate as the primary search…
Exact Reconstruction in Compressed Sensing — RIP is defined through a restriction involving nonzero vectors with ‖x‖0 ≤ s.
Examples of VC-Dimension — VC-dimension is a measure of the capacity of a class of functions.
Exception Class and Inheritance — Custom exceptions inherit from Exception and let you encode domain-specific error information as instance attributes.
Exception Classes and the Throwable Hierarchy — A throw statement lets a program deliberately throw an exception when it finds an error it cannot handle where it was discovered.
Exception Handling Best Practices — Custom exceptions are user-defined classes that inherit from Exception, enabling you to represent errors specific to your…
Exception Handling for File Operations — A JFileChooser object is only a description of a dialog until showOpenDialog or showSaveDialog is called, and those calls block…
Exception Handling Fundamentals — Overtime pay requires applying a 1.5x multiplier only to hours worked above 40; the calculation path diverges at this threshold.
Exception Handling in Python — The logging module allows you to store debugging, information, warning, and critical messages to a file for post-execution…
Exception Handling with try and except — The assert statement verifies that a condition is true; if false, it raises an AssertionError
Exception Handling with try-catch — The actionPerformed method is the entry point for handling a button press, and it receives an ActionEvent describing that press.
Exception Handling with Try-Except — EOFError is raised when input() encounters an end-of-file signal, typically triggered by pressing Ctrl-D (or Ctrl-Z on Windows).
Exception types and how to catch them — The finally block executes unconditionally after try and except blocks, making it ideal for resource cleanup.
Exception Types in Java — An exception is an event that disrupts the normal flow of control in a program.
Exception types in Python — The else block in exception handling runs only when the try block succeeds without raising an exception.
Exceptions and try..catch in Java — A try..catch block lets a program continue after a runtime error. When an exception occurs, the rest of the try block is skipped…
Exceptions in Java Constructors — A canonical constructor takes all required parameters and carries the validation logic, such as the firstName null check in the…
Exceptions in Java Methods — The Hand class specification is read through its signatures and doc comments, which state what each method promises and what it…
Exchanging Data Across the Web: XML, JSON, and APIs — An API is a published contract between applications that specifies the exact rules for requesting and receiving services.
Executing External Commands from Python — A backup program automates file archiving by using Python to construct and execute a zip command with a timestamped filename.
Executive Presence in Engineering Roles — Status is the rank a person plays through speech and bodily presence in each exchange, which can differ from the rank they hold.
Expectation of a Random Variable — Markov's inequality applies to nonnegative random variables.
Expected Loss — Generalized loss functions provide a broad way to assign nonnegative values to model-domain pairs.
Expected Next State Value — Action value computation evaluates a candidate bet by its expected resulting value.
Expected Reward and Decision Making — A k-Armed Bandit Problem is a repeated decision problem with k possible actions.
Expected Reward in Multi-Armed Bandits — The gradient bandit algorithm can be understood as a stochastic approximation to gradient ascent.
Expected Sarsa — Expected Sarsa replaces Sarsa's single sampled next action with an expected next-action value.
Expected Values — D is the distribution; z ∼ D describes sampling from it.
Experiments Leading to the Reward Prediction Error Hypothesis — The hypothesis is about an error between old and new estimates of expected future reward.
Exploration and Off-Policy Learning — Off-policy prediction separates the policy generating experience from the policy whose value function is learned.
Exploration in Action Selection — Action-value estimates are uncertain, so the action that currently looks best may not actually be best.
Exploration vs Exploitation — Greedy actions are those with the greatest current estimated value.
Exploration-Exploitation Trade-off — Dyna-Q+ is Dyna-Q with an exploration bonus.
Exploratory Data Analysis — Clustering organizes objects into groups based on resemblance.
Exploring Goal-Oriented Behavior — A mobile robot can make a choice between searching for more trash and returning to its recharging station.
Exploring the Diverse Applications of Large Language Models — LLMs can power sophisticated chatbots and virtual assistants such as ChatGPT and Gemini.
Exponential Functions with math.exp() and math.pow() — math.log10() computes base 10 logarithms, essential for decibels, pH scales, and other common logarithmic scales
Exporting and Visualizing Cached Geocoding Results — SQLite serves as a local cache to store geocoding results, preventing redundant API calls to the same locations.
Expression Evaluation with a Parser — A syntax error is an error in the syntax of a programming language, and a parser must respond to it deliberately rather than…
Expressions — An assignment statement tells the computer to store a value in a variable.
Expressions and Boolean Values in Java — Java's relational operators are ==, !=, <, >, <=, and >=.
Expressions and Operations — A value is one of the basic things a program works with, like a letter or a number.
Expressions and Statements — Most statements (logical lines) that you write will contain expressions . A simple example of an expression is 2 + 3 . An…
Expressions in Java — ++ increments a variable by 1, and -- decrements a variable by 1.
Expressive Power of Neural Networks — Neural networks with sign activation can be understood through geometric illustrations.
Extended Unpacking with the Asterisk Operator — Unpacking is a syntactic feature that assigns multiple variables from a sequence in one statement, equivalent to index-based…
Extensibility with C/C++ — The ability to integrate C or C++ code into a Python program to handle specific requirements like performance or code protection.
Extracting and Evaluating Model Responses — Instruction-fine-tuned LLMs such as chatbots cannot be judged by a simple label-accuracy calculation, so they are evaluated…
Extracting Attributes and Text from XML Elements — findall retrieves a Python list of all XML nodes matching a specified path, enabling batch processing of similar structures.
Extracting Data from Web Pages — Tag inspection in BeautifulSoup means accessing the individual components of a Tag object: its attributes (via get() or attrs)…
Extracting Data with BeautifulSoup Selectors — BeautifulSoup is a Python library designed to parse real-world HTML, which is often malformed in ways that cause strict XML…
Extracting Data with Groups and Parentheses — The * quantifier matches zero or more characters, while + matches one or more characters.
Extracting Data with re.findall — The dir() function lists all methods and attributes in the re module, helping you discover available regex tools.
Extracting Model Responses — The generate function completes the response instruction step by producing token IDs from a formatted test input.
Extracting Model Responses from Generated Text — A model response is evaluated by placing it next to the correct answer for the same instruction and checking what the…
Extracting Text and Attributes from XML — findall() returns a Python list of all Element objects matching a given path in the XML tree.