AI in our minds? Or hands? Microsoft’s new open source tool kit ensures both.
Who doesn’t want the perfect life of relaxing and simply ordering to get all work done?
There is a sky-high demand for automation which have already started affecting our lives. Deep Learning is the root of our answers. It is all about digging up the multiple levels of representation and abstraction that help to make sense of data such as images, sound, and text that might enable you to start off your day in such a dream in the near future. It is a tool whose objective has been to push machine learning to reach its original goal “Artificial Intelligence” or AI.
Microsoft has just released in the open world market, the deep neural networks that they are using in their own products as cognitive services for eg. for image captioning and speech recognition. This is a deep learning toolkit, previously known as the Computational Network Toolkit (CNTK). It has been open sourced and made it available for anyone to use in their own work on GitHub. The toolkit has far reaching implications in deep learning like medical imaging, speech configuration etc. The latest version of the toolkit, includes new functionality that lets developers use Python or C++ languages in working with the toolkit.
Microsoft’s chief speech scientist claims that its deep learning toolkit is the only publicly available one that can parallel-i take advantage of several GPUs on multiple machines for superior performance giving a new dimension to parallel computing. The CNTK tool kit was available on the Microsoft’s codeplex platform but has been moved to GitHub.
But this is not the only one of its kind of open source deep learning toolkit. Theano, the python based deep learning toolkit is another one to take note of. Named after a Greek mathematician, Theano is considered an industry standard toolkit for deep learning. It was open sourced under the BSD license and was developed by the LISA (now MILA) group at the University of Montreal, Quebec, Canada. At its core, it is a CPU and GPU math compiler.
To replace Theano Google created Most of these tool kits use a python API over C/C++ languages. You won’t be able to do heavy and large scale mathematical computations in Python due its weak interpreter. But its biggest advantage is its simplicity and flexibility. So it is not doing the heavy mathematical computations in most deep learning tool kits.
Python mainly acts a wrapper around highly optimized C / C++ / CUDA / etc. code. This wrapper to heavily optimized stuff allows you to do some awesome things for deep learning. That being said, most of the modern deep learning packages, such as TensorFlow, actually have their core functionalities implemented with C/C++. This only exposes their interfaces as python. Therefore, there should not be too much concerns on performance. You can download and install TensorFlow from here.
Microsoft’s cognitive tool kit has an exceptional ability to quickly process deep learning algorithms across multiple computers running GPUs. This vastly improved the speed at which the latest achievement in speech recognition was achieved. This ultimately helped the company’s speech scientists to configure the system to reach human parity. But from recognition to understanding is a huge step to leap through.
The tool kits like CNTK, TensorFlow, will drive the scientists in future to develop upgraded versions of Artificial intelligence (AI). These will have improved prowess to reach the supreme goal of a complete AI. They would not only have recognition capabilities but will understanding more like us. The ability to capture extremely complicated interactions between parameters makes deep learning tool kits indispensable for a myriad of applications. But whatever comes up next to surprise our progress in deep learning, Microsoft has taken a huge leap and has just given the pathway for democratizing software for the larger good a huge boost. Its deep learning tool kit is the prize meant for the same.