skip to Main Content

Until We Have Quantum Computers for Data Science

For analytics professionals there was a time where model building decisions were nearly entirely related to methodology and algorithm selection.  These decisions also informed the choice of software to use.

 

Today, the availability of algorithms and software has exploded, making choice a daunting task

 

But even more recently, there is a new dimension that has been added to that complexity:  hardware.  We have seen the emergence of chip technologies such as GPUs, FPGAs and ASICs. An example of an ASIC is Google’s TPU (Tensorflow Processing Unit), specifically designed for machine learning.  And Intel will soon be shipping Nervana, a so-called neural network processor (yes, they are calling it an NNP, so add that to the alphabet soup of chip architectures).

 

Before the advent of these new chip architectures we had to devise clever ways of programming algorithms to make the best of limited available processing power (cores and threads).  In fact, the team at First Analytics enjoys collecting CPU utilization screenshots, such as the one below with 48 cores.   We have put all of the cores on a box to work, making an algorithm parallel that isn’t inherently so.

 

, Until We Have Quantum Computers for Data Science

 

Fortunately, our team finds themselves less often needing to come up with programmatic solutions to the performance challenge.  For example, in some recent machine vision projects we have been pleased with the performance of GPU-based systems.

 

Even so, we may be in a transition period, if you are to believe the proponents of quantum computing.  Quantum computers, with their qubits, represent an entirely different approach to computation versus the current binary bit system.  There are promises of performance gains of more than 100 times conventional computing power.  Such computers, if they make it to market, are years to a decade off in the horizon. 

 

In the meantime, what can one do?  Fortunately, in today’s cloud computing world it is not very difficult to set up a hardware environment based on some of the latest chip architectures.  One example is Microsoft’s Data Science Virtual Machine, available as pre-configured images on the Azure platform.  First Analytics is equipped and experienced to help you set one up, and evaluate your problem, through our Collaborative Analytics Lab service.

 

It may be time to revisit those problems you previously thought too large to solve.  Many are now feasible with new hardware and algorithms that can take advantage of these new chips.

 

 

 

 

 

 

Back To Top