# US 2021/0279571 A1 · Neural network optimization mechanism

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0279571 A1
- Document ID: US20210279571A1
- Status: Published application
- Kind: A1
- Application number: 17/177,632
- Filed: 2021-02-17
- Published: 2021-09-09
- Earliest claimed: 2017-04-24
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: Narayan Srinivasa; Joydeep Ray; Nicolas C. Galoppo Von Borries; Ben J. Ashbaugh; Prasoonkumar Surti; Feng Chen; Barath Lakshmanan; Elmoustapha Ould-Ahmed-Vall; Liwei Ma; Linda L. Hurd; Abhishek R. Appu; John C. Weast; Sara S. Baghsorkhi; Justin E. Gottschlich; Chandrasekaran Sakthivel; Farshad Akhbari; Dukhwan Kim; Altug Koker; Nadathur Rajagopalan Satish
- Web page: https://patentgrep.com/patent/US20210279571A1

## CPC

- G06N 3/045: Combinations of networks
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/09: Supervised learning
- G06N 3/098: Distributed learning, e.g. federated learning
- G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn
- G06N 3/063: using electronic means
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

An apparatus to facilitate optimization of a neural network (NN) is disclosed. The apparatus includes optimization logic to define a NN topology having one or more macro layers, adjust the one or more macro layers to adapt to input and output components of the NN and train the NN based on the one or more macro layers.

## Claims

### Claim 1

An apparatus comprising:
a graphics processing unit (GPU) to:
define a neural network (NN) topology as having one or more macro layers;
adjust the one or more macro layers to adapt to input and output components of the NN;
train the one or more macro layers of the NN in parallel based on the one or more macro layers; and
identify, for a training data set, at least one macro layer of the one or more macro layers in accordance with a topology-guided optimization process.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the one or more macro layers, wraps the one or more macro layers each comprise a sub-topology including a plurality of NN layers.

### Claim 3 (depends on 2)

The apparatus of claim 2, wherein the GPU is to replace a first topology of the NN with the one or more macro layers, wraps the one or more macro layers within a stub, and provide an input features node to record output from the first topology.

### Claim 4 (depends on 3)

The apparatus of claim 3, wherein the one or more macro layers, the input features node and a second topology are trained in parallel to search for optimum macro layers for a training data set.

### Claim 5 (depends on 2)

The apparatus of claim 2, wherein the one or more macro layers comprise a standard set of components to facilitate training.

### Claim 6 (depends on 1)

The apparatus of claim 1, wherein the GPU is to optimize the NN by automatically tuning one or more layers in the NN.

### Claim 7 (depends on 6)

The apparatus of claim 6, wherein automatically tuning the one or more layers comprises automatically constructing the NN based on received performance and accuracy constraints.

### Claim 8 (depends on 6)

The apparatus of claim 6, wherein the GPU is to provide the auto-tuning based on one or more statistical algorithms.

### Claim 9 (depends on 1)

The apparatus of claim 1, wherein the GPU is to optimize the NN based on the topology of the NN.

### Claim 10 (depends on 1)

The apparatus of claim 1, wherein the GPU is to perform clustering of processing units to process information relating to modalities.

### Claim 11 (depends on 10)

The apparatus of claim 10, further comprising:
a first cluster of two or more processing units;
a second cluster of two or more processing units; and
one or more routers coupled between the first cluster and the second cluster.

### Claim 12

A method comprising:
defining a NN topology as having one or more macro layers;
adjusting the one or more macro layers to adapt to input and output components of the NN;
training the one or more macro layers of the NN in parallel based on the one or more macro layers; and
identifying, for a training data set, at least one macro layer of the one or more macro layers in accordance with a topoloy-guided optimization process.

### Claim 13 (depends on 12)

The method of claim 12, wherein the one or more macro layers each comprise a sub-topology including a plurality of NN layers.

### Claim 14 (depends on 13)

The method of claim 13, further comprising:
replacing a first topology of the NN with the one or more macro layers;
wrapping the one or more macro layers within a stub; and
providing an input features node to record output from the first topology.

### Claim 15 (depends on 14)

The method of claim 14, wherein the one or more macro layers, the input features node and a second topology are trained in parallel to search for optimum macro layers for a training data set.

### Claim 16 (depends on 11)

The method of claim 11, further comprising optimizing the NN by automatically tuning one or more layers in the NN.

### Claim 17

At least one computer readable medium having instructions, which when executed by one or more processors, cause the processors to:
define a NN topology as having one or more macro layers;
adjust the one or more macro layers to adapt to input and output components of the NN;
train the one or more macro layers of the NN in parallel based on the one or more macro layers; and
identify, for a training data set, at least one macro layer of the one or more macro layers in accordance with a topology-guided optimization process

### Claim 18 (depends on 17)

The computer readable medium of claim 17, wherein the one or more macro layers each comprise a sub-topology comprising a plurality of NN layers.

### Claim 19 (depends on 18)

The computer readable medium of claim 18, which when executed by one or more processors, further cause the processors to:
replace a first topology of the NN with the one or more macro layers;
wrap the one or more macro layers within a stub; and
provide an input features node to record output from the first topology.

### Claim 20 (depends on 19)

The computer readable medium of claim 19, wherein the one or more macro layers, the input features node and a second topology are trained in parallel to search for optimum macro layers for a training data set.
