Neural network comprising matrix multiplication
Methods and data processing systems are disclosed for implementing one or more fully connected layers of a neural network in hardware.According to embodiments, coefficient data for at least one fully connected layer is loaded into an input buffer of a hardware accelerator, and input data for the at...
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Patent |
| Sprache: | eng |
| Schlagworte: | |
| Online-Zugang: | Volltext bestellen |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Zusammenfassung: | Methods and data processing systems are disclosed for implementing one or more fully connected layers of a neural network in hardware.According to embodiments, coefficient data for at least one fully connected layer is loaded into an input buffer of a hardware accelerator, and input data for the at least one fully connected layer is loaded into a coefficient buffer of the hardware accelerator.To be accompanied, when published, by Figure 9A of the accompanying drawings. |
|---|