Model architecture
The structural organization of a computational system’s components, relationships, and operations that determines how it processes information or behaves.
Grokking Machine Learning
The structural specification of a neural network, including its layers, nodes, and activation functions.
- How it is used
- Uses connected classifiers to separate data that a single linear classifier cannot divide.
- How deep it goes
- Defines layer roles and connectivity, then explains how activation functions and successive classifiers shape decision boundaries.
- Examples
- Uses alien word counts and connected career, family, and happiness classifiers to illustrate network structure.
- What you can do after
- Understand how connected perceptrons and layers combine to classify data with more complex boundaries.
Read how Grokking Machine Learning explains model architecture →
Delivering Machine Learning Projects: From design to deployment
A model architecture is the structural arrangement of components and connections that defines how a machine learning model processes inputs and produces outputs.
- How it is used
- Positions model design within pre-project delivery planning, where production IT constraints determine whether an ML system is viable.
- How deep it goes
- Develops a high-level system-component feasibility check, but defers detailed interactions and does not specify an internal model topology.
- Examples
- Uses a smart-building system with sensors, a database, model execution, actuators, and user-facing information to identify required components.
- What you can do after
- Practice/perform a high-level feasibility check of model execution, data flow, and available production components.
Math and Architectures of Deep Learning
The chosen arrangement of neurons and their connections in a neural network.
- How it is used
- Frames model design as choosing a parameterized function before training estimates its weights.
- How deep it goes
- Defines linear architectures and relates neural-network neurons and connections to expressive power, while deferring popular architectures.
- Examples
- Uses a cat threat-score function with hardness, sharpness, two weights, and a bias.
- What you can do after
- Evaluate/understand how architecture choice differs from weight estimation and affects a network’s representational capacity.
Read how Math and Architectures of Deep Learning explains model architecture →
Lead Developer Career Guide
The high-level structural organization of a software system, including its components, interfaces, technologies, and interactions.
- How it is used
- Guides a lead developer’s technical approach and helps align project requirements, team estimates, budget, and system stability.
- How deep it goes
- Explains responsibilities, diagrams, and learning tactics, but does not specify component patterns or implementation details.
- Examples
- Uses a simple “Hello, World” application, a reusable prototype, and a lead developer–lead architect skills table.
- What you can do after
- Practice/perform architecture planning through a simple application and reusable prototype while seeking feedback from experienced developers.
Read how Lead Developer Career Guide explains model architecture →
Architecture Modernization: Socio-technical alignment of software, strategy, and structure
Software architecture is the fundamental structure of a software system, including its major components, relationships, and guiding design decisions.
- How it is used
- Connects software structure to business outcomes, team autonomy, and decisions about where to modernize.
- How deep it goes
- Works through domain decomposition and team alignment, but stops short of implementation-level design.
- Examples
- Uses an airline’s monolith and domain map to show how business goals can guide architectural boundaries.
- What you can do after
- Evaluate/understand it by comparing business domains, software boundaries, and team ownership to find change coupling.
LLMs in Production
A model architecture is the structural design of a machine-learning model, including its computational components and their connections.
- How it is used
- Guides model definition before training, especially when building language models from scratch under task, data, and compute constraints.
- How deep it goes
- Defines layers and connections, demonstrates a small implementation, and stops short of deeply explaining modern architecture changes.
- Examples
- Uses a two-layer feed-forward network, a GPT-style module, and a small text-generation training run as concrete architecture examples.
- What you can do after
- Build it from scratch by selecting components, connecting them, defining loss, and integrating the result into a training loop.
Machine Learning Platform Engineering
A model architecture is the structural design of a machine learning model, including its computational components and their arrangement.
- How it is used
- Guides production model selection, training configuration, evaluation, and lifecycle tracking.
- How deep it goes
- Explains operational trade-offs and visualizes layers and connections, but stops short of deriving architecture internals.
- Examples
- Uses YOLOv8 object detection, pretrained weights, TensorBoard graphs, training inferences, and configurable training artifacts.
- What you can do after
- Evaluate and configure architectures through reproducible training runs, metric visualization, and artifact tracking.
Read how Machine Learning Platform Engineering explains model architecture →
Deep Learning with Python, Third Edition
The structural design of a deep learning model, here contrasted with approaches that adapt at test time rather than only scaling fixed architectures.
- How it is used
- Defines the functions gradient descent can search and encodes assumptions that make a learning problem easier or harder.
- How deep it goes
- Defines layers, settings, and connectivity, then works through Functional graphs and ConvNet patterns; treats choosing an effective architecture as practice-dependent.
- Examples
- Uses a support-ticket model with three inputs and two outputs, a two-layer Functional model, and repeated ConvNet blocks.
- What you can do after
- Apply it via Keras’s Functional API to build layer graphs, and evaluate architecture choices against a problem’s assumptions.
Read how Deep Learning with Python, Third Edition explains model architecture →
Grokking AI Algorithms, Second Edition
A neural network architecture is the arrangement of computational layers, connections, inputs, and outputs that defines a neural network's structure.
- How it is used
- Guides the design and selection of neural networks for prediction, language modeling, and image generation.
- How deep it goes
- Explains configurable layers and outputs, then descends into Transformer dimensions and architecture tradeoffs without presenting implementation code.
- Examples
- Uses multilayer ANNs, multiple outputs, toy and real-world Transformer dimensions, and image-generation architecture requirements.
- What you can do after
- Evaluate architecture choices against a problem’s learning requirements, resource limits, latency, or visual-processing demands.
Read how Grokking AI Algorithms, Second Edition explains model architecture →
Fast Python: High performance techniques for large datasets
- How it is used
- Guides data-processing performance work toward improving architecture and implementation instead of relying on adding servers.
- How deep it goes
- Surveys CPU, memory, storage, and networking implications for Python performance, while narrowing its focus to directly relevant problems.
- Examples
- Contrasts scaling from one computer to multiple machines with moving an in-memory workload to disk.
- What you can do after
- Evaluate performance choices by considering architecture and implementation before increasing infrastructure.
Read how Fast Python: High performance techniques for large datasets explains model architecture →
Learn Generative AI with PyTorch
The internal design and structural organization of a neural model, here referring to the overall layout of an autoencoder.
- How it is used
- Frames generative models as inspectable systems whose design can guide better practical use.
- How deep it goes
- Explains architecture through construction goals and a GPT training-generation workflow, without detailing every internal component.
- Examples
- Uses from-scratch GAN and Transformer construction plus GPT prompt-based text generation as concrete architectural examples.
- What you can do after
- Build it from scratch by relating GPT architecture to token preparation, next-token training, and iterative text generation.
Read how Learn Generative AI with PyTorch explains model architecture →
The Complete Obsolete Guide to Generative AI
The structural design of a machine-learning model, including its components and their organization.
- How it is used
- Frames architecture as a design lever for tailoring language models to tasks, domains, privacy needs, and hardware constraints.
- How deep it goes
- Explains why architecture matters and how it relates to optimization, but stops short of describing specific components or construction steps.
- Examples
- Contrasts general-purpose GPT models with custom LLMs optimized for specialized tasks, domains, efficiency, or privacy.
- What you can do after
- Evaluate/understand why custom architecture choices can improve task fit, efficiency, privacy, and organizational control.
Read how The Complete Obsolete Guide to Generative AI explains model architecture →
Financial AI in Practice
The structural design of a predictive model, including its components and how they are organized.
- How it is used
- Uses architecture choices to turn prepared financial data into predictions or insights, including short-term trading patterns.
- How deep it goes
- Mentions advanced neural architectures and re-architecting as adaptation, without detailing component layouts or implementation.
- Examples
- Contrasts creditworthiness scoring, financial-report interpretation, and neural architectures for short-term price patterns.
- What you can do after
- Evaluate architecture choices for financial tasks and recognize when changing conditions may require re-architecting.
Read how Financial AI in Practice explains model architecture →
Related concepts
- transformer architectureincludes · A transformer architecture is a neural-network architecture that processes sequences or other structured inputs using attention mechanisms and related transformations.
- Apache Kafkacontrasts with · Apache Kafka is a distributed event-streaming platform for publishing, storing, processing, and consuming streams of records.
- hexagonal architectureincludes · Hexagonal architecture is a software architecture pattern that isolates an application’s core logic from external systems through ports and adapters.
- autoencoderincludes · An autoencoder is a neural network that encodes input data into an internal representation and decodes it to reconstruct the input or an approximation.
- linear layerincludes · A linear layer is a neural-network layer that maps input features to outputs by a learned affine transformation, typically connecting each output to every input feature.
- architectural subsystemhas part · An architectural subsystem is a distinct, cohesive part of a larger system that provides related functionality through interacting components and defined boundaries.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.