ING, a large multinational and financial services corporation, is investing heavily in artificial intelligence to achieve efficiency gains and keep up with the fast changing digitization of the financial services landscape. Since machine learning and artificial intelligence practioners like data scientists are very scarce there is a need to re-educate software engineers in order for them to apply machine learning and artificial intelligence. Software engineers are an interesting group of professionals to re-educate since they already posess strong programming skills which are necessary for working in the machine learning field. Furthermore they also have a predisposition to understand the mathemtical basis principles underlying many machine learning and artificial intelligence algorithms. The more you are familiar with mathematics, especially linear algebra and optimization, the more you are able to grasp the inner working of many machine learning and artificial intelligence models, nonetheless applying machine learning and artificial intelligence models can be done without a firm knowledge of the underlying mathetmical principles.
ING asked us to prepare an applied machine learning and artificial intelligence training program for their software engineers with a focus more directed to the applied use of machine learning and artificial models instead of the mathematical underpinnings. On the basis of their requirements we developed a concise training program for ING software engineers focused on applying machine learning and artificial intelligence in a practial manner. We structured the program in such a way that we wouldn't skip over the neccesary basics underlying machine learning and artificial intelligence, but didn't fall into too much detail. We also wanted the training program to have usefullnes in their current work, so they could apply the learned knowledge to solve problems they face on a daily basis.
After carefully taking into account the different requirements we decided to structure the program in this way, starting with the basics of machine learning and artificial intelligence and working our way to applying models:
Introduction:
An overview of the current machine learning and AI landscape: where are we know and where are we going?
Architecture:
From raw data to a useful model: How does a machine learning pipeline and architecture look like? The differences and similarities with software engineering pipelines and architectures
Machine learning and AI models:
What different types of machine learning and artificial intelligence models do exist? From supervised to unsupervised learning and everything in between
Data engineering:
How to transform your raw data to be useful as input for machine learning and artficial intelligence models? Deciding which data to include and how to apply transformations on your raw data
Model training
Optimization and loss functions: what does is actually mean to train a model? The basic mathematical underpinnings underlying training machine learning and artificial intelligence algorithms
The pitfall story: spurious correlations, unrepresentative data, bias, mismatches between your training and test set, overfitting, changing data generation distrubutons over time
Model serving
Making your trained model useful: how to expose your model to the ouside world? It's not just about deploying your trained model, from data validation to making predictions, careful considerations to take into account when exposing your models
Application
Training a model yourself from start to finish: identify an use case in you current work and start appying your new acquired knowledge
During the training program we made extensive use of python programming examples to explain different topics. For this end we used the machine learning library Tensorflow.
Tensforlow is a popular and heavily used machine library in the artificial intelligence and machine learning field. Therefore it was straightforward to use this library during the training program. During the Application phase of the training program the software engineers also made use of Tensorflow.
The program was met with a lot of enthusiasm by the ING software engineers. The great reviews we received after the training program ended also showed the appreciaton the software engineers gave to the training program we constructed. We are very grateful for these great reviews and it was a pleasure to work with these software engineers. They enjoyed the learning process and in the meantime we had great satisfaction in equiping them with machine learning and artifical intelligence knowledge.
Comments
Post a Comment