High Level Product Requirement Specification for a Real Time AI - DSP- ASSP
Table of Contents
A. Background
Overview of Processor Objective
Target Use Case: Geometric Deep Learning (GDL)
Deep Learning Architectures Considered
B. Hypothetical Algorithm
Graph Neural Network (GNN) with Convolution
Convolutional Recursive Neural Network (CRNN)
Forward Leapfrog Integration
Non-Euclidean Techniques
Non-Predefined Topology
Voice Recognition Output
Training Process
C. List of Potentially Required Mathematical Operations
Graph Convolution
Matrix Multiplication
Element-wise Operations
Leapfrog Integration
Attention Mechanisms
Non-predefined Topology Operations
Voice Recognition Tasks
Activation Functions
Loss Function Computations
Optimization Algorithms
Quantization and Compression
Memory Management
Parallelization
Appendices
Appendix A: Graph Convolution Mathematical Operations
Appendix B: Matrix Multiplication Mathematical Operations
Appendix C: Element-wise Operations (C-1 and C-2)
Appendix D: Leapfrog Integration Operations
Appendix E: Attention Mechanism Operations
Appendix F: Non-Predefined Topology Operations
Appendix G: Voice Recognition Operations
Appendix H: Activation Functions
Appendix I: Loss Function Computations
Appendix J: Optimization Algorithms
Appendix K: Quantization and Compression
Appendix L: Memory Management
Appendix M: Parallelization