Computing Device FFT Projections for Resource-Efficient AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML methodologies for controlling fault-intolerant and safety-critical apparatuses, such as vehicles, face computational inefficiencies and resource constraints, leading to high costs and environmental impacts, especially when performing real-time processing of sensor data.
Innovation Solution
A method involving random projections in the frequency domain using Fast Fourier Transformations (FFT) to compute output values from input values, reducing the number of operations and memory needed for AI/ML tasks, particularly suitable for ELM models, allowing increased model capacity and computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional AI/ML methodologies are used for real-time processing of sensor data in fault-intolerant systems, then model capacity and processing capability are improved, but computational efficiency deteriorates and resource consumption increases
Solution Approach 1:
The patent applies parameter changes by transforming data from the spatial domain to the frequency domain using Fast Fourier Transform (FFT). This transformation changes the representation of input data, enabling convolution operations to be performed more efficiently. By working in the frequency domain, the system achieves reduced computational complexity while maintaining model capacity for processing sensor data in real-time.
Solution Approach 2:
The patent substitutes conventional mechanical computation with optimized mathematical operations in the frequency domain. Instead of performing traditional spatial convolution which requires numerous multiplications and additions, the system uses FFT-based approaches that leverage mathematical properties of frequency transformations to achieve the same computational result with fewer operations, thereby improving productivity.
2Productivity
If conventional AI/ML methodologies are used for real-time processing, then processing capability is improved, but resource consumption and operational costs increase
Solution Approach 1:
The patent transforms computational operations from the spatial domain to the frequency domain using FFT, changing the parameters of data representation. This transformation enables the system to achieve the same processing capability with significantly reduced computational resources, directly lowering energy consumption and operational costs while maintaining real-time processing capability.
Solution Approach 2:
The patent performs preliminary actions by pre-computing and storing frequency-domain representations of filter kernels or convolution weights. This preliminary transformation allows the actual processing to be performed through simpler operations during real-time execution, reducing the computational burden and resource consumption during critical processing operations.
3Adaptability or versatility
If conventional AI/ML methodologies are used, then model capacity is improved, but the number of operations and memory requirements increase
Solution Approach 1:
The patent changes the parameter space in which computations are performed by transforming data to the frequency domain. This parameter transformation exploits the convolution theorem to convert complex spatial convolutions into simpler frequency-domain multiplications, significantly reducing the number of operations required while maintaining the model's capacity to handle complex patterns.
Solution Approach 2:
The patent moves computations from the spatial dimension to the frequency dimension using Fourier transformations. This dimensional change allows the system to represent and process data in a different space where operations become more efficient. The frequency-domain representation enables the same model capacity to be achieved with fewer computational operations and reduced memory requirements.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances computational efficiency and model capacity, enabling AI/ML applications in areas previously constrained by resource limitations, particularly benefiting fault-intolerant systems like vehicles and medical devices, with reduced operational costs and environmental footprint.
Implementation Method 1
A method involving random projections in the frequency domain using Fast Fourier Transformations (FFT) to compute output values from input values
Data Source
AI summary
A method for computing an output value from a number of input values, a computer program, apparatus, and a vehicle, an aircraft. The method includes an input number representing a vector size of an input vector containing the input values; defining a projection number of projections as a multiple of the input number based on a multiplying number; precomputing a transformed projection tensor as a transformation of a projection tensor having a projection size of the input number multiplied by the multiplying number and containing weight values in the frequency domain; obtaining a transformed embedded input tensor by computing a transformed input vector as a transformation of the input vector into the frequency domain; embedding the transformed input vector into the transformed embedded input vector; and using the transformed projection tensor and the transformed embedded input vector for computing the output value.