IP Library › Granted Patent US 11,989,933
Granted Patent B2
US 11,989,933 · App. 16/976,412 · Granted May 21, 2024

Polynomial convolutional neural network with late fan-out

Inventors: Felix Juefei Xu (Pittsburgh, PA); Marios Savvides (Pittsburgh, PA)
Assignee: Carnegie Mellon University
G06V10/82G06F18/21G06F18/21355G06F18/2414G06F18/253G06N3/045G06N3/048G06N3/08G06N3/084G06V10/454
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 11,989,933
App. No.
16/976,412
Granted
May 21, 2024
Kind
B2
Abstract

The invention proposes a method of training a convolutional neural network in which, at each convolutional layer, weights for one seed convolutional filter per layer are updated during each training iteration. All other convolutional filters are polynomial transformations of the seed filter, or, alternatively, all response maps are polynomial transformations of the response map generated by the seed filter.

Claims (18)

1. A method for training a neural network comprising, for each convolution layer in the neural network:

receiving a gradient function;

adjusting a convolutional filter based on the gradient function;

receiving an input;

convolving the convolutional filter with the input to generate a seed response map;

generating a plurality of augmented response maps based on the seed response map, wherein each element of each augmented response map is generated by applying a polynomial to one or more elements of the seed response map; and

generating a plurality of feature maps by applying an activation function to the seed response map and to each of the plurality of augmented response maps.

2. The method of claim 1 further comprising:

applying a vector of learnable coefficients to the plurality of feature maps to generate a single feature map.

3. The method of claim 2 wherein:

m feature maps are generated from the seed response map and m−1 augmented response maps; and

the vector contains m learnable coefficients.

4. The method of claim 2 further comprising:

adjusting the m learnable coefficients based on the gradient function.

5. The method of claim 4 wherein the single feature map is generated at layer l of the neural network, further comprising:

using the single feature map as the input for layer l+1 of the neural network.

6. The method of claim 1 wherein generating a plurality of augmented response maps comprises raising each element of the seed response map to a different exponent.

7. The method of claim 6 wherein the different exponents are integer exponents or fractional exponents randomly sampled from a distribution.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Feb 9, 2021
From: XU, FELIX JUEFEI; SAVVIDES, MARIOS
To: CARNEGIE MELLON UNIVERSITY
Reel/Frame 055190/0511 →
Continuity (2)
Provisional Application 62762292 · Apr 27, 2018
Related Publication 20210042559A1 · Feb 11, 2021