Invention Grant
- Patent Title: Incremental precision networks using residual inference and fine-grain quantization
-
Application No.: US15869515Application Date: 2018-01-12
-
Publication No.: US11556772B2Publication Date: 2023-01-17
- Inventor: Abhisek Kundu , Naveen Mellempudi , Dheevatsa Mudigere , Dipankar Das
- Applicant: Intel Corporation
- Applicant Address: US CA Santa Clara
- Assignee: Intel Corporation
- Current Assignee: Intel Corporation
- Current Assignee Address: US CA Santa Clara
- Agency: Jaffery Watson Mendonsa & Hamilton LLP
- Priority: IN201741015052 20170428
- Main IPC: G06N3/08
- IPC: G06N3/08 ; G06N5/04 ; G06N3/04 ; G06T15/00 ; G06F9/46 ; G06N3/063 ; G06T17/20 ; G06T15/80 ; G06T17/10 ; G06T15/04 ; G06V10/94

Abstract:
One embodiment provides for a computing device comprising a parallel processor compute unit to perform a set of parallel integer compute operations; a ternarization unit including a weight ternarization circuit and an activation quantization circuit; wherein the weight ternarization circuit is to convert a weight tensor from a floating-point representation to a ternary representation including a ternary weight and a scale factor; wherein the activation quantization circuit is to convert an activation tensor from a floating-point representation to an integer representation; and wherein the parallel processor compute unit includes one or more circuits to perform the set of parallel integer compute operations on the ternary representation of the weight tensor and the integer representation of the activation tensor.
Public/Granted literature
- US20180314940A1 INCREMENTAL PRECISION NETWORKS USING RESIDUAL INFERENCE AND FINE-GRAIN QUANTIZATION Public/Granted day:2018-11-01
Information query