Deep Learning vs Traditional Machine Learning AI Tools
Philippe Touati
Management Board | Corporate & Institutional Banking | Tech & Communications
The December Forbes Issue describe well the difference between ML and Deep Learning.
https://www.forbes.com/sites/bernardmarr/2016/12/08/what-is-the-difference-between-deep-learning-machine-learning-and-ai/#44b82d516457
The question I have in my mind for a while, is it worth investing time on traditional ML Tools when Deep Leaning seems to be winning. When you look at Google works on tensor Flow or Facebook work with Images, you have the impression that CNNs or RNNs with Deep learning are the only way to go.
Neural Networks Models have a lot of bias, limited ability to explain conclusion or non AI specialist and difficult to train. For bank, for instance it may be an issue. They may be good at fraud detection or credit approval but it may be difficult to explain to regulators what process was used and what criteria were selected and more importantly why....
A good analysis from DatascienceCentral web site give some good answers: Has Deep Learning Made Traditional Machine Learning Irrelevant?
https://www.datasciencecentral.com/profiles/blogs/has-deep-learning-made-traditional-machine-learning-irrelevant
Would love to get your views.
Owner, IN5K Consulting
8 年Seems to me that deep learning is a particular case of machine learning (or, similarly, neural nets) - only with more layers, or "deeper" - so anyone claiming to be an expert in deep learning first needs to have a grasp of machine learning and neural nets. But definitely, deep is where the action is now for two reasons: it works better, and we can (i.e. GPUs and TPUs are now available to handle the massive parallelism). That is, until someone comes up with a new hyped buzzword... Haven't thought much about the use cases for banking (besides the everyday applications as a customer), but there are interesting thoughts to be had, as a parent or mentor: what skills will be in high demand in the near future? It could be that teachers and (data) gatherers make a comeback, whereas the aristocracy of engineers, lawyers, bankers and doctors (dare I add artists, inventors and of course politicians?) are headed towards obsolescence!... Entertainers and social workers would be needed to take care of the newly unemployed, but then again a well trained machine will be more efficient at that too!...
Hello Philippe, I happen to work currently on this subject and I would like to stress two things about Deep Learning: first deep learning needs a monstrous (really) volume of data before producing useful results, second you do not know why the results are what they are, as in not being able to understand the contribution of factors in the production of predictions by the network. So most of the time, when you do not have really that much data (think for example macro-economic series) and you would very much like to understand the underlying model build by the machine learning algorithm, you are better with traditional ML like for example random forests. In conclusion, deep learning is a very promising technology but that is stil in a R&D stage with a lot remaining to understand while traditional ML are better understood statistics tools.