Creating AI(Deep Learning) solutions for business ? What You need to know.
Having a good deep learning model that solve a real problem can take a lot of time and resources. There is a lot of materials in the internet about AI, but from my observation the focus is still more in area of science not business.
Here I want to focus more on production related problems which maybe of interest equally for business owners and IT developers.
The first papers about deep learning were published by Alexey Ivakhnenko in 1967 but only recently we are witnessing extreme growth in AI development. Computing power available at the time of most fundamental discoveries in AI was not enough to put development further.
The deep learning computational complexity problem was addressed with usage of Graphical Processing Unit(GPU) for Deep Learning algorithms. GPU is in general a specialized computer processor responsible for rendering graphics for display on an electronic device.
The game changer was introduced by 2012, when GPUs had evolved allowing on very efficient parallel computing. Thanks to GPU, big and time consuming problems can be know spitted into multiple problems and be solved at the same time.
As a business owner who wants to have AI solution You should be specially interested about cloud server that will be used to train deep learning model. Depending on algorithms used to train deep learning model and volume of data, a professional developer will give You approximate time needed to train the model.
What is CUDA ? CUDA is the parallel computing platform enable the computing power of GPUs for general purpose processing. CUDA was created by NIVIDIA which leads in development and production of highly efficient GPU processors.
CUDA is a parallel computing platform and programming model that makes using a GPU for general purpose computing simple and elegant. The developer still programs in the familiar C, C++, Fortran, or an ever expanding list of supported languages, and incorporates extensions of these languages in the form of a few basic keywords.
So what really drives AI development at the moment is hardware. The current deep learning algorithms continue to solve more complex problems but the demand on extra computational powers is still growing.
We have a GPU which is hardware, CUDA which is a platform with libraries that know how to utilize GPU power and Deep Learning frameworks. Deep Learning frameworks allows developer to focus on solving the problem, all needed algorithms are already implemented, what is even more important those frameworks are already built to be highly efficient by using GPU and CUDA.
Developing AI applications start with training deep neural networks with large datasets. GPU-accelerated deep learning frameworks offer flexibility to design and train custom deep neural networks and provide interfaces to commonly-used programming languages such as Python and C/C++. Every major deep learning framework such as TensorFlow, PyTorch, and others, are already GPU-accelerated, so data scientists and researchers can get productive in minutes without any GPU programming.
PyTorch is a deep learning framework I used so far. It is great because of its pythonic structure and simplicity.
PyTorch is integrated with NIVIDIA deep learning library cuDNN.
The NVIDIA CUDA® Deep Neural Network library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks. cuDNN provides highly tuned implementations for standard routines such as forward and backward convolution, pooling, normalization, and activation layers.
Did You know that CUDA is not only for Deep Learning ? As developer You may gain incredible performance boost in Your existing code by using CUDA. If You work with Python, check Numba, it is developed by Anaconda developers and supported among others by NIVIDIA.
If You are developer and want to know more about CUDA, check this article.
PyTorch is supporting ONNX. ONNX is a general format how to represent deep learning model. So You could train Your deep learning model on high performance cloud Amazon Web Servers and download it and try predicting on Your local PC.
How to release AI solution for business and satisfy Your client ? First and the most important inform client about costs, if it is too much for client try to find existing solution, there is many pre-trained models that You could use.
If Your client will be specially interested about performance I would first start looking after solution proposed by hardware manufacturers like NIVIDIA as they will for sure know best how to use hardware with deep learning to achieve best results.
If You are interested about how to use deep learning for business and not to exhaust Your client budget and patience check my last project :)