Statistical Memory Network for Noisy Stochastic Signal Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory networks are unable to effectively process stochastic values such as speech and video signals contaminated with noise due to the requirement for deterministic values, limiting their ability to handle uncertainty information.
Innovation Solution
A statistical memory network is developed, incorporating a stochastic memory, uncertainty estimator, writing and reading controllers, and statistic updaters to estimate and manage uncertainty information, allowing for probabilistic memory positions and storage of stochastic values with average and variance calculations using Kalman filters and neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing memory networks use deterministic values for storage, then memory operations are simple and efficient, but they cannot process stochastic signals such as speech and video signals contaminated with noise
Solution Approach 1:
The patent changes the fundamental parameter of memory storage from deterministic single values to stochastic representations using mean and covariance matrices. This allows the memory network to handle noisy speech and video signals by storing statistical properties rather than exact values, enabling probabilistic reasoning about uncertain inputs
Solution Approach 2:
The memory value is segmented into multiple components: mean vector and covariance matrix. This segmentation allows the system to separately track the central tendency and uncertainty of stochastic signals, providing a more comprehensive representation that handles noise effectively while maintaining manageable computational structure
2Loss of information
If existing memory networks store deterministic values, then reading and writing operations are straightforward, but they lack the capability to estimate and manage uncertainty information of input signals
Solution Approach 1:
The controller uses feedback mechanisms to estimate uncertainty information from input signals and adjusts the memory operations accordingly. The uncertainty estimator continuously monitors input signal characteristics and feeds this information back to the reading and writing controllers, enabling adaptive handling of stochastic data while maintaining structured processing
Solution Approach 2:
An uncertainty estimator is introduced as an intermediary component between the input signals and the memory operations. This mediator estimates uncertainty information and provides it to the controller, which then uses this information to adjust reading and writing operations, effectively managing stochastic data without requiring complete redesign of the memory architecture
3Speed
If existing memory networks use fixed memory positions, then access speed is high, but they cannot handle probabilistic memory positions required for stochastic signal processing
Solution Approach 1:
The memory position is transformed from a fixed deterministic value to a dynamic probabilistic distribution. The reading controller uses the covariance matrix to determine not just a single memory position but a probability distribution over multiple positions, allowing the system to access relevant information across a range of locations while maintaining efficient search through the probabilistic structure
Data Source
AI summary
Provided are an apparatus and method for a statistical memory network. The apparatus includes a stochastic memory, an uncertainty estimator configured to estimate uncertainty information of external input signals from the input signals and provide the uncertainty information of the input signals, a writing controller configured to generate parameters for writing in the stochastic memory using the external input signals and the uncertainty information and generate additional statistics by converting statistics of the external input signals, a writing probability calculator configured to calculate a probability of a writing position of the stochastic memory using the parameters for writing, and a statistic updater configured to update stochastic values composed of an average and a variance of signals in the stochastic memory using the probability of a writing position, the parameters for writing, and the additional statistics.