Has anyone had a chance to tinker with multiple major AI platforms such as TensorFlow, Cognitive Talk, Quill etc...

  • What are the strengths and weaknesses of different AI platforms?

Comprehensive articles that tackle this topic would be helpful.


In recent times different data science magazines and institutions have published their reviews of the top AI toolkits. In these reviews they tend to highlight the innovative features possessed by each platform as well as their reliability and ability to scale.

Below are a some evaluations of AI platforms that I recommend you have a look at:

KDnuggets review https://www.kdnuggets.com/2016/04/top-15-frameworks-machine-learning-experts.html

Paperspace review https://blog.paperspace.com/which-ml-framework-should-i-use/

Predictive Analytics review https://www.predictiveanalyticstoday.com/deep-learning-software-libraries/

Infoworld review https://www.infoworld.com/article/3026262/machine-learning/13-frameworks-for-mastering-machine-learning.html#slide4

Wikipedia comparison https://en.wikipedia.org/wiki/Comparison_of_deep_learning_software

However if your main interest is to get an objective and unbiased assessment of the different ML toolkits. Then the best place to get such first hand reviews is at GitHub. Here you can look at the number of forks, stars and downloads, the update frequency, the number of issues, the number of contributors, the documentation, the license and even the code itself. This approach enables you to get an impartial and neutral appraisal directly from the developers themselves.

Below is my personal evaluation of the different ML toolkits:

Tensorflow - TensorFlow is a powerful open source library for numerical computation using data flow graphs. Some of the advantages of tensorflow are: It is portable and consequently can run on GPU's, CPU's, desktops, servers and mobile computing platforms. It comes with tensorboard which is useful in visualising and fine-tuning your network. Tensorflow has a large following with lots of researchers and developers using it. This means that most issues can be dealt with easily since they are usually the same issues that other developers run into.

Microsoft CNTK - Is a unified deep learning toolkit that describes neural networks as a series of computational steps via a directed graph. It delivers state of the art results when it comes to speech recognition, in recent times it has even outperformed humans in transcribing speech to text. The external CNTK toolkit available as open-source is also identical to the internal toolkit. This means you get to access all the features that internal Microsoft developers utilize in production.

Theano - Theano is a symbolic compiler library that builds an optimized Neural Network from mathematical expressions. A major advantage of theano is that it supports Python libraries meaning that a wide range of Python libraries i.e. Numpy, SciPy and Scikit-learn can easily be integrated with theano. Additionally theano has excellent high level API's such as keras and lasagne which greatly simplify the use of the Theano toolkit.

Torch - Is a scientific computing framework and script language that provides a wide range of algorithms for machine learning. The benefits of torch are: It has a very active open source community given that Facebook is standing behind it. Additionally it is based on Lua which is a very easy language to pick up. On top of that a significant number of AI researchers publish their research findings in torch.

Caffe - Is a deep learning framework originally developed at UC Berkeley. Some of the perks of using Caffe are: It has an excellent Matlab and Python interface, it supports functions such as ReLu straight out of the box furthermore Caffe excels at CNN's for images.


Not the answer you're looking for? Browse other questions tagged or ask your own question.