Bayesian Neuromorphic Compiler for Non-Stationary Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems for translating Bayesian network models into neuromorphic computing models require manual specification of neuronal network structures and are limited to specific datasets, failing to effectively handle non-stationary relationships between random variables.

Innovation Solution

A Bayesian Neuromorphic Compiler that automatically translates Bayesian network models into neuromorphic computing models using a network composition module with probabilistic computation units, enabling the computation of conditional probabilities and control of devices based on these calculations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If manual specification of neuronal network structure is used, then the system can be implemented on conventional hardware, but the system cannot handle non-stationary relationships and requires careful manual specification for each specific dataset

Engineering Contradiction:

Improveability to handle non-stationary relationshipsVSAvoidmanual specification of neuronal network structure

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system automatically translates Bayesian network models into neuromorphic computing models without requiring manual specification of neuronal network structures. The Bayesian Neuromorphic Compiler performs self-service by generating the appropriate spiking neural network topology and configuration from the input Bayesian network model, eliminating the need for manual intervention and enabling handling of non-stationary relationships

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces manual mechanical specification processes with an automated computational translation system. The Bayesian Neuromorphic Compiler substitutes the manual process of specifying neuronal network structures with an automated algorithmic translation from Bayesian network models to spiking neural networks, reducing complexity and improving adaptability

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If Bayesian network models are translated to neuromorphic computing models for efficient computation, then inference speed improves, but the system requires automatic translation mechanisms that increase implementation complexity

Engineering Contradiction:

Improveinference speedVSAvoidautomatic translation mechanism

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The Bayesian Neuromorphic Compiler acts as an intermediary that translates between Bayesian network models and spiking neural network models. This intermediary system enables efficient neuromorphic computation by automatically converting the input Bayesian network into the appropriate neuromorphic representation, achieving fast inference while managing translation complexity through a dedicated compiler component

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The Bayesian Neuromorphic Compiler provides a universal translation mechanism that can handle various Bayesian network models and transform them into spiking neural networks. This multi-functional compiler serves multiple purposes: translating models, learning conditional probabilities, and enabling deployment on different neuromorphic hardware platforms, thereby justifying the implementation complexity through broad applicability

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Adaptability or versatility

If population coded static distributions are used as in Pecevski and Maass, then the system works for specific datasets, but it fails for non-stationary relationships between random variables

Engineering Contradiction:

Improvehandling of non-stationary relationshipsVSAvoidperformance on specific datasets

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent transitions from static population-coded distributions to dynamic spiking neural network representations that can adapt to non-stationary relationships. The spiking neural networks learned by the Bayesian Neuromorphic Compiler can dynamically adjust to changing statistical properties of the input data, maintaining reliability across both stationary and non-stationary datasets through temporal dynamics

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS11521053B2Network composition module for a bayesian neuromorphic compiler

Publication Date: 2022.12.06 HRL LAB

AI summary

Described is a system for specifying control of a device based on a Bayesian network model. The system includes a Bayesian neuromorphic compiler having a network composition module having probabilistic computation units (PCUs) arranged in a hierarchical composition containing multi-level dependencies. The Bayesian neuromorphic compiler receives a Bayesian network model as input and produces a spiking neural network topology and configuration that implements the Bayesian network model. The network composition module learns conditional probabilities of the Bayesian network model. The system computes a conditional probability and controls a device based on the computed conditional probability.