For anyone curious whether they should download and read the book, this is a high level summary, the cliff notes of applying ML/AI, aimed at managers. It won't do much for a programmer looking to learn neural networks, but it will give you an overview of what they're being used for and some things to watch out for if you have ML programmers on your team.
i cringe every time i see "machine learning" and "ibm" in the same sentence. it's not because ibm research is subpar (if anything, id consider them second to none) but because their marketing efforts represents everything wrong with the current ai cycle - that is, selling magic beans to nontechnical and helpless buyers.
I knew of a few people who left our company to work in the Watson healthcare division sales side, post purchase of Merge Healthcare.
Everyone in the industry, assisted diagnosis, thought this was gonna be a game changer, with IBM with merge dominating.
They lasted about 9 months before coming back or going to another company.
Even what they were briefed to sell on, IBM couldn't exactly do all it was marketing.
If someone wants a gentle and high quality introduction to Machine Learning, I can't recommend "
Hands-On Machine Learning with Scikit-Learn and TensorFlow" by Aurélien Géron [1] enough. I wish I had this book when I was starting. It explains everything from data engineering, through how algorithms work, to practical applications. Everything in Python 3, covering pandas, scikit-learn, tensorflow. It is absolutely wonderful!
I think one of the hardest parts of ML is not learning how to write it but learning how to pick features and clean data. So, assuming this is a run down of types of ML algorithms (which is how most "learn ML" things seem to go) this is not going to be much more helpful for someone who is still trying to figure it all out. It's important to know these things but knowing the actual algorithms is kind of the low hanging fruit. Data organization and features are the real complex pieces of most ML.
This is absolutely correct! After spending a lot of time trying to understand why building machine learning models is so difficult, I came to the same conclusion that "feature engineering" is the key to building high performing models.
While feature engineering's importance is generally recognized [0], it's unfortunate that there aren't more tools and formal methods for applying it. Personally, I am a developer of an open source python library called Featuretools (https://github.com/featuretools/featuretools/) that is trying to change this for tabular and multi-table datasets. We are working hard to make automated feature engineering available to everyone and have a list of demos for people to try here: https://www.featuretools.com/demos.
It's also worth noting that deep learning is changing the need for feature engineering. However, it primarily works in cases where you don't need interpretable features and you have plenty of data. This means that it's biggest success have been in images, audio, and text problems. For all other use cases feature engineering is still a necessary step for applying machine learning.
What do you find as vaporware? I don't think they are offering vaporware, but their marketing is doing lot of hype hoping to enter into the startup ecosystem, but are failing miserably due to their traditional focus on selling to large corporates. I have attended IBM sponsored conferences and I was floored by the the reach they have for their cloud offering, especially on large firms and Govts. Agree that Bluemix, as an offering is nowhere as polished as AWS is, but they do have a good business going around it.
I was speaking of Watson. It seems like every time it comes up someone who was tasked with investigating it for their company or someone who actually worked at IBM comes along and talks about how big of a joke it actually is.
I was one of those people tasked with investigating Watson and I had the exact experience you described. It was a huge joke. Tried out Bluemix as well, it wasn't a huge joke but it wasn't great either.
What was lacking in Watson? Lack of features or lack of accuracy? I had integrated Watson in on of the libraries in a platform I was building and found it reasonable. Haven't compared it extensively with other offerings from Microsoft or AWS, so interested in knowing if you have..
