# US 11,451,242 B2 · Method and apparatus for variable rate compression with a conditional autoencoder

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

- Number: US 11,451,242 B2
- Document ID: US11451242B2
- Status: Granted
- Kind: B2
- Application number: 17/009,213
- Filed: 2020-09-01
- Published: 2020-12-17
- Granted: 2022-09-20
- Earliest claimed: 2019-03-18
- Assignees: Samsung Electronics Co., Ltd.
- Applicants: Samsung Electronics Co., Ltd.
- Inventors: Yoo Jin Choi; Mostafa El-Khamy; Jungwon Lee
- Web page: https://patentgrep.com/patent/US11451242B2

## CPC

- H03M 7/3059: {Digital compression and data reduction techniques where the original information is represented by a subset or similar information, e.g. lossy compression}
- H03M 7/6011: {Encoder aspects}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/0495: Quantised networks; Sparse networks; Compressed networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- H03M 7/40: Conversion to or from variable length codes, e.g. Shannon-Fano code, Huffman code, Morse code
- H04L 1/0003: {by switching between different modulation schemes}
- H04L 1/0009: {by adapting the channel coding (H04L1/1812 takes precedence)}
- H04L 1/0016: {involving special memory structures, e.g. look-up tables}
- H04N 19/13: Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC]
- H04N 19/147: according to rate distortion criteria (rate-distortion as a criterion for motion estimation H04N19/567)
- H04N 19/19: using optimisation based on Lagrange multipliers
- H04N 19/91: Entropy coding, e.g. variable length coding [VLC] or arithmetic coding
- G06N 3/045: Combinations of networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/063: using electronic means
- H04N 19/103: Selection of coding mode or of prediction mode
- H04N 19/124: Quantisation
- H04N 19/70: characterised by syntax aspects related to video coding, e.g. related to compression standards
- H03M: CODING; DECODING; CODE CONVERSION IN GENERAL (using fluidic means F15C4/00; optical analogue/digital converters G02F7/00; coding, decoding or code conversion, specially adapted for particular applications, see the relevant subclasses, e.g. G01D, G01R, G06F, G06T, G09G, G10L, G11B, G11C, H04B, H04L, H04M, H04N; ciphering or deciphering for cryptography or other purposes involving the need for secrecy G09C)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION

## Abstract

A method and apparatus for variable rate compression with a conditional autoencoder is herein provided. According to one embodiment, a method includes training a conditional autoencoder using a Lagrange multiplier and training a neural network that includes the conditional autoencoder with mixed quantization bin sizes.

## Claims

### Claim 1

A method for compression, comprising:
determining a modulation coding scheme (MCS) from among a plurality of MCSs included in a lookup table;
selecting, from the lookup table, a Lagrange multiplier that corresponds to the determined MCS in the lookup table;
training an autoencoder using the Lagrange multiplier;
implementing the trained autoencoder into a network; and
training the network including the trained autoencoder with quantization.

### Claim 2 (depends on 1)

The method of claim 1, wherein a value of the Lagrange multiplier is selected from a predetermined set of values.

### Claim 3 (depends on 1)

The method of claim 1, wherein the network produces a compressed output with a varying compression rate and distortion depending on a conditioning value of the Lagrange multiplier.

### Claim 4 (depends on 1)

The method of claim 1, wherein the autoencoder is trained using conditional convolution.

### Claim 5 (depends on 1)

The method of claim 1, wherein training the network further comprises training the network with mixed quantization bin sizes.

### Claim 6 (depends on 5)

The method of claim 5, wherein training the network with mixed quantization bin sizes further comprises drawing a uniform noise for various noise levels for random quantization bin sizes.

### Claim 7 (depends on 1)

The method of claim 1, wherein training the network further comprises encoding a latent representation of an input.

### Claim 8 (depends on 7)

The method of claim 7, wherein training the network further comprises quantizing the encoded latent representation with a selected quantization bit size.

### Claim 9 (depends on 1)

The method of claim 1, wherein training the network further comprises encoding a quantized latent representation with entropy encoding.

### Claim 10

A system, comprising:
a network; and
a processor configured to:
determine a modulation coding scheme (MCS) from among a plurality of MCSs included in a lookup table,
selecting, from the lookup table, a Lagrange multiplier that corresponds to the determined MCS in the lookup table;
train an autoencoder using the Lagrange multiplier;
implementing the trained autoencoder into the network; and
train the network including the autoencoder using quantization.

### Claim 11 (depends on 10)

The system of claim 10, wherein a value of the Lagrange multiplier is selected from a predetermined set of values.

### Claim 12 (depends on 10)

The system of claim 10, wherein the network produces a compressed output with varying compression rate and distortion depending on a conditioning value of the Lagrange multiplier.

### Claim 13 (depends on 10)

The system of claim 10, wherein the autoencoder is trained using conditional convolution.

### Claim 14 (depends on 10)

The system of claim 10, wherein the processor is further configured to train the network by training the network with mixed quantization bin sizes.

### Claim 15 (depends on 14)

The system of claim 14, wherein the processor is configured to train the network with mixed quantization bin sizes by drawing a uniform noise for various noise levels for random quantization bin sizes.

### Claim 16 (depends on 10)

The system of claim 10, wherein the processor is further configured to train the network by encoding a latent representation of an input.

### Claim 17 (depends on 16)

The system of claim 16, wherein the processor is further configured to train the network by quantizing the encoded latent representation with a selected quantization bit size.

### Claim 18 (depends on 10)

The system of claim 10, wherein the processor is further configured to train the network by encoding a quantized latent representation with entropy encoding.