DBTB INT Ricardo Baeza Yates

I'm Ricardo Baeza-Yates and right now I'm an independent consultant. I think I'm not really frequently talking to industry audience. Usually I talk to more research audience and I think that makes it very interesting because the type of questions are different and also the expectations may be different. Well I think because we have much more data and we have tools like deep learning it's much more easy to obtain better results. However, in my talk I wanted to basically to get people aware of all the possible biases in data and other biases that algorithms can introduce. So at least we take the results with a bit of a grain of salt to make sure that they are realistic. I guess the main insight is that we have many biases that come from people. These are positive and negative and depending on the case they have different views

In some case something positive may be negative and vice versa. So if you are aware of them you can take them to your advantage and also try to, in the case of the negative ones, try to unbiased them or basically try to minimize their impact or at least if you know them to understand better your results. That's a hard question. I think you don't only need, not only you need to say a great programmer and know all the tools that you can use to analyze data. I think you need to really, uh, lab data, understand data. That means that every time you need to deep dive in data, try to basically look at many samples, look at distributions and really understand what is there before you do anything. I see many people don't look at data before they start. They use data as an input and not as the key thing in the problem.