Hearing Aid Neural Weight Encoding for Low-Power DNN Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hearing devices implementing deep neural networks (DNNs) face computational inefficiencies and potential performance degradation due to high computational costs and memory requirements.
Innovation Solution
A hearing device with a processor that retrieves weights from a weight data structure using an index parameter, employing 4-bit non-uniform quantization to reduce the bit width of neural network weights, thereby improving computational efficiency and memory usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep neural networks are implemented in hearing devices, then advanced signal processing capabilities are achieved, but computational cost and power consumption increase significantly
Solution Approach 1:
The patent applies quantization to change the precision parameter of neural network weights from high precision (e.g., 32-bit floating point) to low precision (e.g., 4-bit integers). This parameter change reduces the computational complexity and memory bandwidth requirements, thereby lowering power consumption while maintaining adequate signal processing performance for hearing device applications
2Reliability
If high precision weights are used in neural networks, then network performance is maintained, but memory usage and processing time increase
Solution Approach 1:
The patent changes the precision parameter of weight representation from high precision to low precision through quantization. By carefully designing the quantization process and using techniques such as weight sharing and lookup tables, the system achieves acceptable network performance while dramatically improving processing efficiency and reducing memory usage in resource-constrained hearing devices
3Measurement precision
If full precision weight data is stored, then accurate neural network computation is achieved, but memory requirements become prohibitive
Solution Approach 1:
The patent applies quantization to reduce the bit precision of stored weight data from standard floating-point formats to compressed integer representations. This parameter change in precision allows the system to store neural network weights in limited memory capacity while maintaining sufficient accuracy for hearing device signal processing through techniques like lookup tables and weight sharing
Solution Approach 2:
The patent uses lookup tables to store pre-computed quantized weight values, allowing the system to retrieve approximate weight values quickly without storing full precision data. This copying approach enables efficient memory usage by storing only essential weight information in compressed form while maintaining functional equivalence for the application
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A hearing device is disclosed. The hearing device comprises a set of input transducers for provision of transducer input data, the set of input transducers comprising a first input transducer for provision of a first transducer input signal as part of the transducer input data. The hearing device comprises a processor for processing transducer input data and providing an electrical output signal based on the transducer input data. The hearing device comprises a receiver for converting the electrical output signal to an audio output signal. The hearing device comprises a memory having stored thereon a weight representation indicative of a weight of a plurality of weights of a neural network based on the transducer input data.