AI or die: the race to AI solutions implementation

May 29 2019

Using artificial intelligence became almost mandatory for businesses, to the point that it is more important to claim that you apply AI than to do it usefully. Of course, there is nothing bad in wanting to integrate machine learning algorithms in your company. Businesses have all the right to look for the most advanced solution to their problems. Not mentioning that the private sector is playing a crucial role in research and development when it comes to machine learning, creating an almost unprecedented synergy with academia. So, if you want AI, go for it! Use machine learning if you need it. Just do it in the right way.

 

 

Have a clear understanding of which problem you want to solve. Moreover, if you, after some consideration, see that there is a non-AI more efficient solution to it, choose that. Don’t get too attached to the “AI-hype”, when unnecessary.

When you are confident enough that you would benefit from machine learning, then is time to choose your machine learning weapons. On internet we can find a great deal of precooked algorithms, pretrained neural networks, and cheat sheets, that give an indication on how to use the different tools. And for some specific problem these tools might be very effective, especially simple classification problems or image recognition problems (image recognition algorithms nowadays are unbelievably powerful). But for more involved problems they might perform miserably. The solutions out of the internet box are dangerous if used without knowledge, they might even seem to work until they don’t anymore. Most of the time your only hope is to write, or have written for you, a customised solution or finding a proprietary solution that does the job for you.

What does it take to write a machine learning algorithm?

 

 

Python, R, TensorFlow, GPU, Spark, Hadoop, linear algebra, optimisation: these are some of the key words that one might hear when it comes to machine learning algorithm development. If you want to develop your own customised solution you must get familiar with at least some of them (I would say that linear algebra is mandatory, but that is only my theoretical physicist opinion). Apart from linear algebra, also the programming language Python is a favourite of mine, partly because it can be used as a starting point to understand most of the other key words. It is open source and there are uncountable contributions to it. The libraries for machine learning are very extensive. And once you are accustomed to that, writing and training a machine learning algorithm can be pretty fast. If you are thinking to enter in the world of data science, Python will be a good starting point. If you are instead interested in hiring a date scientist to take care of your machine learning algorithms, then Python should probably appear in the job announcement. Finally, remember that a Data Scientist is not the only bit you need to build your AI powered company (see here for more details).

 

Tags: Blog

Integration monitoring tool checklist

How does your Azure, BizTalk or SQL monitoring stack up?