Instinct pool

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

An instinct pool is a shared collection of reusable neural-network layers or functions from which environment-specific models select instincts.

An instinct pool is a reusable set of neural-network layers that multiple environment-specific models can select and recombine. In instinctual deep reinforcement learning, each DQN keeps its own input and output layers while sharing selectable middle layers. section 12.6

A Keras functional model receives an ordered instincts list: an environment-specific input layer, three selected instinct layers, and an environment-specific output layer. The instinct outputs are concatenated before producing actions. section 12.6

A practical implementation separates:

• A shared pool of Dense layers, from which each environment selects three instincts. section 12.6
• Environment-specific input, output, and replay-memory objects. section 12.6
• An evolutionary evaluation loop that assembles each candidate network and scores it across environments. section 12.6

Figure 12.15 IL applied to DL

See how two DQN networks share a pool of middle-layer instincts while retaining environment-specific outer layers. section 12.6

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

How deep it goes

Implements layer slicing, concatenation, candidate evaluation, and evolutionary selection, but does not derive the underlying optimization theory.

Examples

Uses two environments, three shareable instinct slots, Dense layers, concatenated outputs, and an evolutionary evaluation notebook.

What it assumes

Assumes familiarity with DQN or DRL and introduces the Keras functional API for treating layer sections as functions.

What you can do after

Build it from scratch with Keras functional models and evolve environment-specific DQN compositions.

How it connects

Links deep reinforcement learning, DQN, Keras functional API, and evolutionary optimization.

liveBookmanning.com

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

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.