Francisca, Jessica, Helton discussed on O'Reilly Data Show

Automatic TRANSCRIPT

In order to position of new the ethic -nology and imagining based solutions you also mentioned in the the one of the things that seems to hold company spent his in they might be concerned about explain ability fairness in buys. We also found this in our survey. So to what extent are you starting to hear people about this not in the sense of reading about it? But actually, this is about level issue that we when? Asked me Bill the state we have consider. Yes, I think we are saying this with some of us tumors way, you know, I mean, especially when it comes to help data or government related and bangs. I think you know, what they really want to know what actually drives the machine Honey model so in I mean, if a customer comes Magnus, why is it that I was not approved credit or something. And if it's just entirely based on a model as the model just told us that it's, you know, deny the application was says, except I think all these institutions want to actually have an answer if they actually question when why is it that this is the decision? So I think we on trying to helton's organizations and China ensure that the data that's used to build these models at as unbiased as possible, and that is difficult because historic data. May not be at data in some domain. So which office, but this is just the initial, you know, attempts by this is a real erectile men. I it is a real consultants. I think now when it comes to data, and, you know, pretty much earlier when studies detest takes deci can't as many unbiased, you know, samples in the data. But the thing is now is pretty much like some of the data is just by by nature. So you have to somehow short is unbiased before you actually build a model. So this has been a great conversation. I wanted to close by having Francisca give us a preview do they training which breaking news giant might be part of. She doesn't know what? Next year, so hard. Jessica will be each ING today training for us at the conference actually starting at San Francisco called four gassing financial time series with deep learning on Asher. So give us a high level overview of what so one what's the right backroom for someone for disrobing? What should they expect ler? So there are not too many correct receipts for these at today training. I think that experience with coding by Donna and be very useful than of resume. Busy understanding of machine. Learning nip learning affix and terminology also. And then I mean, we time serious for past also can help because again all day training today training is going to be focused on for casting financial time series, citing that these can be very helpful. So what I'm planning to do during these today training is a to the students through the course for using Azure machine. Learning service to train machine learning models, both locally and also nevermore computerese source. Then they will use it training and deployment workflow for machine learning service in icon, Jupiter know, actually, we will have a movable notable because we will start. We didn't ingestion data preparation for time series. And we will have a few notebook suggest for the modeling. Eat those. Absolutely. So you can then use the notebook as template to train, you know, your own machine. Earning Manila with your own at the app, specifically, the use case that we are going to use for abuser training is how Janury upmarket prediction with the longest short term memory Alice, the Emma and forks, so as many of you know, Radia models can use the story of a sequence of data. And then predict the one future elements of the sequence are going to seduce very helpfully many different financial use cases, for example, when you need to although prizes. So I will give any production to science the forecast than I will give any production as I said to narrow networks for time-series forecast. We will use Azure, machine learning services to build a these hand to hand the solution and apply..

Coming up next