Security News

### JSR: What We Know So Far About Deno’s New JavaScript Package Registry

JSR, a new package registry from the Deno team, aims to address npm’s limitations but the JavaScript community is concerned about ecosystem fragmentation.

graph-compression-google-research

Matrix compression for neural networks.

- 0.0.4

- Maintainers
- 1

Readme

This document describes an experimental API that facilitates matrix compression of a neural network's weight tensors. The API helps inject the necessary tensorflow operations into the training graph so the model can be compressed while it is being trained.

Full documentation can be found here.

- Library Overview
- Model creation
- Hyperparameters for compression
- Adding compression ops to the training graph
- Example

**MatrixCompressorInterface**- used to implement any matrix compression algorithm in the method**CompressionOpInterface**- used to create a tensorflow operator-like object that injects any matrix compression method dynamically into a tensorflow layer.**ApplyCompression**- convenience wrapper class that can be used directly or extended for novel compression operator types; used to repeatedly invoke the compression operator to different layers in a model.**CompressionWrapper**- wrapper module used to create the proper ApplyCompression implementation for the compression_option (method) of choice.

The first step involves creating an ApplyCompression object, with the desired compression parameters. This object then is used to compress the model weights and use these compressed weights during the forward execution of the graph. Matrices are compressed to the rank specified in the compression parameters, provided at the start. To apply the compression, the weight tensor of the layer should be wrapped with the compression object's 'apply_compression' method, provided in compression_op.py.For an example, see the section below.

The pruning library allows for specification of the following hyper parameters:

Hyperparameter | Type | Default | Description |
---|---|---|---|

name | string | model_compression | Name of the compression specification. Used for adding summaries and ops under a common tensorflow name_scope. |

alpha_decrement_value | float | 0.01 | Real number by which alpha is decremented at each update. |

begin_compression_step | integer | 0 | Global step at which to begin compression. |

end_compression_step | integer | -1 | Global step at which to terminate compression. Defaults to -1 implying compression continues till the training stops. |

compression_frequency | integer | 10 | Intervals at which compression is applied and compression parameters updated. |

compression_option | integer | 0 | Indicates what type of factorization/compression to use (see the list below for the algorithm options). |

rank | integer | 100 | Factorization rank (r), where if A = BC. See definition below of how rank (r) is used to compute final weights matrix dimensions. |

update_option | integer | 0 | Indicates how update logic is being run: 0 - use tensorflow operations for updates; 1 - use python functions for updates. |

use_tpu | boolean | False | Experimental flag - training using TPUs |

- Low Rank Approximation
- Simhash
- Dictionary Learning
- Kmeans Quantization

The hyperparameter rank (r) is used to compute the new ranks as such: (rank of A) * (100 / r) + 1. For simhash compression, the value r provided should be the ratio value you would like divided by 8 (i.e. 300 / 8 -> same as using r = 300 in the equation above). This is because simhash compression represents values as bits (rather than bytes) therefore the true rank is the size of the array divided by 8.

If the original weights were m-by-n and the compressed decomposition B*C is (m-by-k)*(k-by-n), then the compression ratio is (m*k + k*n) / (m*n).

A gradually increasing alpha value is used to smooth the compression from start_step to end_step. This way the model gradually moves from the full weights matrix to a compressed one. For example, in the low-rank approximation scheme, the weight matrix that is used in the training process is W = (alpha) * A + (1 - alpha) * BC. This alpha value is decremented over time from alpha = 1 to alpha = 0, using the alpha_decrement_value at intervals of compression_frequency.

```
# Parse compression hyperparameters
compression_hparams = compression.CompressionOp.get_default_hparams().parse(
hparams)
# Create a compression object using the compression hyperparameters
compression_obj = compression_wrapper.get_apply_compression(
compression_hparams, global_step=global_step)
# somewhere in the model, compute the compressed weights
local = tf.nn.relu(
tf.matmul(reshape, compression_obj.apply_compression(weights, scope)) +
biases,
name=scope.name)
all_update_op = [apply_gradient_op, ...] # all existing model updates
# Run compression update steps with all the other updates. Example below is
# assuming update_option=0.
all_update_op.append(compression_obj.all_update_op())
with tf.control_dependencies(all_update_op):
train_op = tf.no_op(name='train')
```

Ensure that `global_step`

is being incremented, otherwise compression will not
work!

As an example, the cifar10 model provided in Tensorflow’s Advanced Convolutional Neural Networks (see page for more details) has been modified to incorporate the compression library:

- cifar10_compression.py creates the deep CNN and adds the weight compression to the fully-connected layers.
- cifar10_train.py creates the compression object and provides it to the training graph (described above) to use.

To train the compression version of cifar10 (make sure you're working in a properly configured virtualenv - as setup using the run.sh script):

```
$ python cifar10_train.py --compression_hparams=name=cifar10_compression,alpha_decrement_value=0.005,begin_compression_step=40000,end_compression_step=100000,compression_frequency=100,compression_option=1,use_tpu=True,update_option=0,rank=200 --max_steps 120000
```

Eval:

```
$ python cifar10_eval.py --compression_hparams=name=cifar10_compression,alpha_decrement_value=0.005,begin_compression_step=40000,end_compression_step=100000,compression_frequency=100,compression_option=1,use_tpu=True,update_option=0,rank=200 --run_once
```

Eager execution example.

An eager execution example is provided at compression_lib/examples/mnist_eager_mode/mnist_compression.py. To train the model, run:

```
$python mnist_compression.py
```

Authors: Rina Panigrahy (corresponding author -- email: rinap@google.com), Lucine Oganesian, Sudeshna Roy, Xin Wang, with support from: Badih Ghazi for helpful contributions, Rasmus Pagh (pagh@google.com, doc) for the simhash code, Zoya Svitkina for code reviews, and Suyog Gupta for consultations.

FAQs

Matrix compression for neural networks.

We found that graph-compression-google-research demonstrated a **healthy** version release cadence and project activity because the last version was released less than a year ago. It has **1 open source maintainer** collaborating on the project.

Did you know?

Socket installs a GitHub app to automatically flag issues on every pull request and report the health of your dependencies. Find out what is inside your node modules and prevent malicious activity before you update the dependencies.

Security News

JSR, a new package registry from the Deno team, aims to address npm’s limitations but the JavaScript community is concerned about ecosystem fragmentation.

By Sarah Gooding - Feb 22, 2024

Security News

International law enforcement organizations have disrupted LockBit, the world’s largest ransomeware gang, seized their operations and infrastructure, and indicted some of the perpetrators.

By Sarah Gooding - Feb 20, 2024

Research

Security News

Socket discovered two malicious Python packages, enchantv and vibrant, imitating popular packages and targeting victims via a base64 encoded payload in their setup files.

By Socket Research Team, Sarah Gooding - Feb 16, 2024