Reservoir Calculation Device for Time-Series Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning algorithms like RNN and LSTM for time-series data processing require significant computational resources and operation amounts, leading to long training times and high power consumption, necessitating elaborate parameter tuning for high performance.
Innovation Solution
A reservoir calculation device with an input circuit, a reservoir circuit, and an output circuit, where the reservoir circuit includes intermediate circuits with neuron circuits and time constant circuits to generate intermediate signals that undergo transient changes, allowing for efficient processing and representation of feature information in input signals without extensive training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RNN or LSTM is used for time-series data processing, then memory representation capability is improved, but operation amount and power consumption increase significantly
Solution Approach 1:
The invention segments the neural network training process into two distinct parts: (1) a fixed reservoir network that processes time-series data without training, and (2) only the output layer weights that require training. This segmentation eliminates the need for backpropagation through the entire network, dramatically reducing computational operations and power consumption while maintaining memory representation capabilities.
Solution Approach 2:
The reservoir network is pre-configured with fixed random weights and structure before processing time-series data. This preliminary setup allows the reservoir to immediately process data with temporal memory capabilities without requiring iterative training, reducing the operation amount during actual data processing.
2Measurement precision
If RNN or LSTM is used for time-series data processing, then memory representation capability is improved, but training time increases
Solution Approach 1:
By segmenting the network into a fixed reservoir and a trainable output layer, the invention eliminates the need for time-consuming backpropagation through the reservoir, reducing training time to only the output layer weight optimization while preserving temporal memory capabilities.
Solution Approach 2:
The invention extracts the computationally intensive training requirement from the reservoir network and concentrates it only in the output layer. The reservoir itself requires no training, taking out the time-consuming iterative optimization process from the majority of the network while maintaining its memory functions.
3Productivity
If reservoir computing is used to reduce operation amount, then computational efficiency is improved, but accuracy of weight representation decreases
Solution Approach 1:
The invention introduces an intermediary analog multiply accumulator circuit that performs weight calculations in the analog domain, maintaining high precision while enabling efficient parallel processing. This intermediary component bridges the fixed reservoir and output layer, preserving accuracy through precise analog computation before digital conversion.
Solution Approach 2:
The invention replaces digital computation with analog circuit implementation for the reservoir and weight operations. Analog circuits naturally perform multiplication and accumulation operations with high precision and parallelism, improving computational efficiency while maintaining accuracy through physical signal processing rather than discrete digital operations.
4Speed
If hardware implementation of reservoir unit is used, then processing speed is improved, but data conversion errors increase
Solution Approach 1:
The invention maintains continuous analog signal processing throughout the reservoir and weight operations, avoiding repeated analog-to-digital and digital-to-analog conversions. This continuity preserves signal integrity and minimizes conversion errors while enabling high-speed parallel processing through analog circuit operations.
Solution Approach 2:
The analog multiply accumulator serves as an intermediary that performs precise analog computations and only converts to digital at the final output stage. This single conversion point minimizes data loss compared to multiple conversions, preserving processing speed while reducing conversion errors through careful analog signal management.
Data Source
AI summary
A reservoir calculation device according to an embodiment includes a reservoir circuit and an output circuit. The reservoir circuit receives input data and outputs intermediate signals, each undergoing a transient change when the input data changes. The output circuit outputs an output signal obtained by combining the intermediate signals. The reservoir circuit includes intermediate circuits, each including a neuron circuit and an intermediate output circuit. The neuron circuit generates an intermediate voltage undergoing a transient change corresponding to weight data and the input data when the input data changes. The intermediate output circuit outputs an intermediate signal representing a level of the intermediate voltage from the neuron circuit. The neuron circuit includes a time constant circuit capable of changing a time constant. The time constant circuit is connected between a reference potential and an intermediate terminal outputting the intermediate voltage.