Every project has its own story. Most engineering projects are not just delivered and forgotten. Project needs to be monitored and maintained with every aspect. Like a symphony, the project needs to be orchestrated. But when it comes to AI . . .The story becomes fairy tale. 📖
First things first we need a problem but this problem needs to be promising and precision to solve. In industry, it’s called a business case or business problem. Most of the time, projects start with data to solve mistakes. It means that project managers or stakeholders look at their resources to generate problem data that fits the data. Most meetings start with ‘So we have data, what should we do with it ? ‘. . . In a proper approach, problems need to be determined at the phase of the big bang of the project. Every blast should have a reason. Additionally this reason should be, as we mentioned above, precious. What are the metrics of the value ? That question is the topic of another story. In summary, we need a reason, problem to solve with data.
Yes, we are at the level of Data. By the way, the header is a reference for the paper called “Attention All You Need”. Let’s go on with data. Data is the second part of AI based project development. It’s important to mention that not every problem needs a data based (learning) solution. But we are talking about AI projects that include Machine Learning and Deep Learning.
In the data phase, there are important points that should stand on it. First of all, data availability. Data needs to be available or easily and cheaply becomes available. For example you have tons of data that can be stored geographically and technologically separated and it’s expensive to get them and arrange them in one. In this situation, the cost of the project can go beyond the estimated earnings. Feasibility dropped down. We need to remind ourselves that, we will investigate further more every step with different stories . . .🙃
After that we need to develop and research a model that solves our problem and generates business value. This phase is a relatively small and easy step. I can say easy because it depends on technical quality and knowledge of the team. Of course there are a lot of metrics like hyper-parameter tuning, feature extraction and selection etc. For the moment we just focus on mainstream lines.