Dynamic Neural Network Creativity via Neuron Activation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural network (ANN) generative models struggle to induce creativity through unsupervised learning, as they often rely on static structures that fail to adapt effectively to changes in input data, making it difficult to generate novel and meaningful data instances.
Innovation Solution
A system comprising an AI platform with an encoding manager, evaluation manager, and activation manager that automatically selects and manipulates neurons from one or more layers of an encoder in an ANN, altering activation patterns to generate novel data instances from the latent space, promoting creativity without requiring re-training of the network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If static structures are used in ANN generative models, then the model structure is simple and easy to implement, but the model fails to adapt effectively to changes in input data, reducing creativity and novelty in generated outputs
Solution Approach 1:
The patent applies dynamics by transforming the static neural network structure into a dynamic one where neuron activation patterns can be manipulated and changed during operation. The system dynamically selects and manipulates subsets of neurons in the encoder layer, allowing the model to adapt to different input data changes without requiring structural modifications or retraining, thus resolving the contradiction between adaptability and structural simplicity.
Solution Approach 2:
The patent changes the activation parameters of selected neurons in the encoder layer to induce creativity. By modifying activation patterns (e.g., setting activation thresholds, selecting active/inactive neurons) rather than changing the overall network structure, the system achieves adaptability while maintaining the original simple model architecture, preventing catastrophic forgetting and avoiding the need for retraining.
2Productivity
If neuron activation patterns are manipulated to induce creativity, then novel and meaningful data instances are generated, but the process requires complex selection and manipulation mechanisms
Solution Approach 1:
The patent extracts and isolates specific neurons from the encoder layer for manipulation, rather than modifying the entire network. By selecting subsets of neurons based on their relevance to the input data and creatively manipulating only those specific neurons, the system generates novel outputs with meaningful changes while avoiding the complexity of managing the entire network, thus resolving the contradiction between productivity and system complexity.
3Productivity
If static activation patterns are used, then the model is computationally efficient and easy to operate, but the model cannot generate creative or novel outputs
Solution Approach 1:
The patent implements self-service by enabling the neural network to automatically select relevant neurons and manipulate their activation patterns based on the input data itself. The system autonomously determines which neurons to activate and how to modify their patterns without requiring complex external control mechanisms, thus achieving creativity while maintaining operational simplicity and ease of use.
Data Source
AI summary
Embodiments relate to a system, program product, and method for inducing creativity in an artificial neural network (ANN) having an encoder and decoder. Neurons are automatically selected and manipulated from one or more layers of the encoder. An encoded vector is sampled for an encoded image. Decoder neurons and a corresponding activation pattern are evaluated with respect to the encoded image. The decoder neurons that correspond to the activation pattern are selected, and an activation setting of the selected decoder neurons is changed. One or more novel data instances are automatically generated from an original latent space of the selectively changed decoder neurons.