Evolutionary deep learning

As explained in Evolutionary Deep Learning: Genetic algorithms and neural networks

Evolutionary deep learning integrates evolutionary computation with deep learning to optimize or automate deep-learning models and their development.

Evolutionary deep learning (EDL) combines evolutionary computation with deep learning to automate and optimize model development. It is a broad family of patterns rather than one algorithm: evolutionary search can operate on data and features, hyperparameters, validation and selection, network architecture, or network weights. section 1.0 section 1.1 section 1.3.1

Use EDL when conventional development leaves difficult search choices or when a complex problem needs further optimization. Evolutionary methods explore candidate solutions through search patterns that can uncover alternative strategies or architectures; they can also replace backpropagation for direct parameter search. The trade-off is substantial computation, so simpler systems may not justify the cost. section 1.1 section 1.2 section 6.0

Common EDL applications include:

• Hyperparameter optimization, with random and grid search serving as non-evolutionary baselines before PSO, evolutionary strategies, or differential evolution. section 5.0
• Neuroevolution for hyperparameters, network parameters or weights, and network optimization; direct weight search can eliminate error backpropagation. section 1.6 section 6.0
• Architecture and model-workflow optimization, including tuning network size and form, feature engineering, validation, and model selection. section 1.3.1
• Specialized tasks such as discontinuous weight search, balancing generative adversarial networks, and replacing deep reinforcement learning. section 1.2

From baseline search to neuroevolution

A practical path starts by manually tuning a network, then establishes random and grid-search baselines. Evolutionary methods can next search its hyperparameters. For direct neuroevolution, construct a simple NumPy multilayer perceptron, evolve network parameters instead of propagating loss backward, then extend the approach to a Keras model and image classification. section 5.0 section 6.0

Figure 1.4 Applying EC (EDL) to the AI/ML model development workflow for DL

Maps evolutionary-computation strategies onto stages of the AI/ML development workflow, clarifying that EDL can automate more than network training. section 1.5

How Evolutionary Deep Learning: Genetic algorithms and neural networks covers it

How deep it goes

Defines a broad technique family, then develops HPO and direct network-parameter optimization through practical implementations.

Examples

Uses random and grid search, PSO, evolutionary strategies, differential evolution, NumPy multilayer perceptrons, Keras models, and image classification.

What it assumes

Accommodates novices while building deep-learning and evolutionary-computation fundamentals before combining them.

What you can do after

Apply it via evolutionary search methods to optimize deep-learning hyperparameters, architectures, or network parameters.

How it connects

Connects evolutionary computation, AutoML, hyperparameter optimization, neuroevolution, backpropagation, and generative adversarial networks.

liveBookmanning.com

Read Evolutionary Deep Learning: Genetic algorithms and neural networks on liveBook Get the book at manning.com

Related concepts

• evolutionary computationuses · Evolutionary computation is a family of search and optimization methods that iteratively improve candidate solutions through variation and selection inspired by biological evolution.
• deep learninguses · Deep learning is a machine-learning approach that uses neural networks with multiple computational layers to learn representations from data.
• hyperparameter tuninguses · Hyperparameter tuning is the process of selecting or adjusting configurable model or procedure settings and evaluating them against a chosen objective.
• deep reinforcement learningalternative to · Deep reinforcement learning is reinforcement learning that uses deep neural networks to represent policies, value functions, or other decision-making components.
• AutoMLkind of · Automated machine learning (AutoML) uses computational procedures to automate selected tasks in developing and applying machine-learning models, such as data preparation, model selection, training, evaluation, and tuning.
• neuroevolutionincludes · Neuroevolution is the application of evolutionary optimization to neural-network architectures, parameters, or hyperparameters.

This explainer was generated by Manning's knowledge graph from the text of Evolutionary Deep Learning: Genetic algorithms and neural networks. The section links point to the parts of the book it draws on.