The best way to keep your TensorFlow installation secure is to create a unique, hard-to-guess password for your TensorFlow account. You can read more about this in our password security guide. TensorFlow is used by many folks for research and for training their machine learning models. Unfortunately, the nature of open-source software means that TensorFlow is not immune to vulnerabilities, and there have been a number of serious security issues discovered in recent months. Not only does this negatively impact TensorFlow users, it also negatively impacts the reputation of AI research. We are committed to ensuring that TensorFlow remains a secure platform and are eager to learn about any issues that you encounter. If you have any issues with TensorFlow that you believe could be of use to other users, make sure to file a GitHub issue.
Binpack Encoding and Decoding
A new, improved way of encoding and decoding binary data is available in TensorFlow 1.9.0 which is significantly faster than the old method. This new method of encoding and decoding data is called binpack encoding. Binpack was first introduced to TensorFlow in December of 2016 by @wjmckinney with the help of @lianmaogao and a few other community members. Binpack encoding has the advantage that it’s a lossless compression algorithm: You can decompress your encoded string back to its original form without losing any information, but you can also compress your encoded string again by just writing it back out to disk.
The speed improvements gained from binpack encoding are mainly due to the fact that it's an order of magnitude faster than the traditional methods used before its introduction (e.g., bzip2) since it uses fixed-sized blocks instead of variable-sized blocks like most lossless compression algorithms. If you use the latest version of TensorFlow, binpack should be enabled by default for all operations on Python objects such as tf.data .
Install TensorFlow using Anaconda
The best way to install TensorFlow is with Anaconda. You can read more about this in our installation guide but, if you haven't already, sign up for an account on the Anaconda website and download the latest version of the software from their site.
Once you have downloaded and installed Anaconda, you should be able to start a new Python environment by running the command conda create --name "TensorFlow" -n tensorflow .
After that, use pip to install TensorFlow: pip install tensorflow
What happened in TensorFlow 1.7.0?
The TensorFlow 1.7.0 release addresses a significant vulnerability in the previous version of TensorFlow, 1.6.1, that allows attackers to steal sensitive information from a machine learning model's parameters file and black-box computation on it. The new release fixes this problem by changing the way that parameter files are created and used while providing an option to suppress these changes in older versions of the software.
TensorFlow Object Detection API
TensorFlow Object Detection is a new object detection API designed to run on TensorFlow models. It provides several different pre-trained models that you can use to detect things like cars, pedestrians, and animals. The models are trained using over 225 million images from the ImageNet dataset and have been shown to perform at an accuracy of 92 percent or higher on ImageNet.
Timeline
Published on: 09/16/2022 23:15:00 UTC
Last modified on: 09/20/2022 14:42:00 UTC