When I first started down the path of selling DL and ML infrastructure I took the decision to know all I could. I am not a Data Scientist or have ever been taught anything to do with ML, I can’t code, and I used to think a Python was a snake… But I was determined to know all I could so that I could offer the best service to my clients.
Well as you can imagine the learning curve has been very steep! I can't tell you how many books I had to read…and how little they made sense at the start!! However, after many failed attempts I can now hold my own in pretty much any situation. I have worked on deployments with some of the biggest names in the industry like SAP, I have given board level presentations to F1 teams, trained a very successful business team, launched what we think is one of the biggest and best ML and DL sites around (www.novatech.co.uk/deeplearning/) deployed systems to Academia, Defence, Research, Retail, manufacturing and AI start-ups. I had to understand Docker, containers, optimised frameworks, source codes, batch sizes, different nets one can use and their benefits (DNN, CNN, DCN, DCIGN, BM, DAE, SAE, DFF, FF, RNN, GRU’s) I started to learn the benefits of using NN's in Gas Turbine Diagnosis, or say NN's for fabric defect classification. I am even working on a project based around general AI (not narrow) application creation (not very good for small talk at the family dinner table) but clients do want to know that I know what I am talking about, so that they can trust my recommendations and that of my team.
In short after working in this space for around 2/3 years it is only now that I feel I know enough to say I am knowledgeable in a very limited capacity. And even then, compared to the audience that Andriy Burkov has, my skills are little more than useless compared to what a real Data Scientist has; more talking points than anything else. But I was never trying to be a Data Scientist, I just wanted to be better at my job. The reason for this was my desire to prove someone wrong! An old boss of mine once said all I ever would be is simply a sales person no more no less and that I was not adaptable. It hurt and made me want to prove him wrong, so I said I would be the best at what I do, even if it meant learning about ML and DL (not the easiest of things to do if you can imagine with zero experience) oh, and I left the company where the boss said that I was no good :)
So maybe some of you are starting to ask yourselves how this is relevant to the article and challenge Andriy posted : https://www.linkedin.com/feed/update/urn:li:activity:6384544024790908928/
On a daily basis I am speaking to clients from companies that are the biggest and some that are the smallest, CEO’s CIO’S down to System admins and Data Scientists and even government members and investors. And pretty much they are all interested in purchasing systems and hardware that we offer – of course. But the thing that is really interesting is that they have no idea as to what it is they need vs what they are being told they need! They all think they have to spend hundreds of thousands to get the solution the team needs!
But "why?", I hear you ask is this the situation
AI, ML, DL are all massive buzz words at the moment. Type Deep Learning into Google to see what I mean. The interest levels in this are completely out of control, and of course with all this attention in DL and ML means the marketing people go into overdrive producing content telling the world they need this system or that system. Of course all these systems cost a fortune, and in lots of cases are complete overkill for what is needed. And because most skillsets are focused on X86 infrastructure, the moment you start talking GPU based systems, Ubuntu, Frameworks, cuDNN, flash arrays or anything else they aren’t comfortable with then they typically look for advice from there trusted reseller of choice. But unless your reseller of choice is specialised in this area of DL and ML, the chances are they will simply say you need this system (X) that I have seen advertised lots, and there is a constant machine gun assault by the marketing chaps to say you should be buying product (X) or in the worst case the reseller (with lack of skills) will try and build you a system – I’ve seen this a few times and trust me when I say it’s not pretty!! I had a conversation last year with a Financial Service house in London . They had spent nearly £1M on new infrastructure from a South American company. After 9 months they replaced the system completely! I see this a lot. Not always on this scale, but it is a fairly normal.
“Developing deep learning models is a bit like being a software developer 40 years ago. You have to worry about the hardware and the hardware is changing quite quickly… Being at the forefront of deep learning also involves being at the forefront of what hardware can do.” - Phil Blunsom, Oxford University and DeepMind
If you are in a position of trust within a company and your company is looking for new ML/DL platforms then reach out to a specialist who can correctly advise you. I have seen so many clients ...spend hundreds of thousands of pounds on hardware equipment... when all they really needed was some guidance and help, and a small pot of money to make your DL dreams come true.
We work in a space that is centric to data and learning, but there seems to be such a big skills gap from resellers and even the bigger global partners to correctly position a platform for a company. Its like if you don’t know simply sell them a bunch of V100 GPU’s in a big box!
Deep Learning is all about data and learning. If I can adapt and learn then so can anyone - not JUST the sharpest of knives in the draw!
Give the best service you can to your clients, train yourself to be better, make the best team you can with what you have learned, empower them and don’t hold them back, then start the process again and repeat always!