Reema, Daraghmeh, Arriva discussed on Data Skeptic

Data Skeptic
|

Automatic TRANSCRIPT

Reema over a remus sin. Like i didn't do. You need more data for it to become effective. Or what's the use case there. If any i remind tyrannize actual needs a very crisis. Komodos what we knew. Canada is actually just do countries at the mortar. The modern approach of iron is removed from elisa. Angola thanks to get some incremental improvements. This idea and the we're going to disarm problems if you use a remorse daraghmeh are you may source than those problems are unpredictable and we saw those problem may not be unpredictable. Based on our lances this kid point. This is studying point. Well i was going to say we're going to replace i- remorse target just want to make some incremental contributions and into kansas in from arriva. I'm gonna so some real problems. This how precision another reason maybe to use the more traditional techniques is because arena sin requires a bit more training. You should know about spectral analysis and linear systems analysis. A bit in order to properly use it. Do you think that will always be a barrier or do you envision a day when machine learning will do the fine tuning into nine google machine learning or artificial intelligence in the end. I actually believe that. The machine learning motor official intelligence border. Huma do listen to fight tiny that models this see the true but how can will choose. Aren't i think the process is all very deng on the cruises and this moment Women not achieve that. But i think we can. Vacate is the end after the efforts of the researchers from the hull. Would maybe we come make. And this moment is not possible to use machine learning to fight time anymore though are any focused program any trainee anymore to catch the time where i i. Don't think that is possible this moment. But i think that can be a cool to pursue to achieve. Well it's always good to hear someone say that it's not possible to do something with machine learning 'cause then usually six to twelve months. We have a paper published showing it. Yes yes it's bad to actually kind of the aero time-series civil nobel is work on time series but the problem is this area is that many many beautiful servicemen many beautiful results but the problem is we still cannot predict royston postings very dukes verse symbol for example you cannot predict the twin of stock market cannot pigment reveal if you can predict where where everyone can be very rich but we cannot predict roy and many other problems. Simple problems look simple but you still cannot predict where we will and the machinery Should be very important technologies but there is still a lack distance from rooney appliances models to solve real. Problems gives some real benefits. I think you still like difference. Not like these or you make it. A point are many many papers. Van's cab your paper. Machinery focused. he. Actually most of the papers cannot be used in practice for example. Let's say this very interesting problem if you ask many researchers even professor as we can we predict a demand for s q stock capping unique of everyday usually the this is not predictable but actually in companies. They have this demand. The how this request alibaba or amazon. I think because if you can predict this room will and then they can make decisions. Will you winter decisions. Assortment decisions so subtle scenes. So i think the arosa many papers can be published but this does not equals the problems especially the focusing problem concerts focused purpose is reality soft for practically a union practice. This vacation and this problem. I think in that direction. You're right there are many papers published and not as many of them can be practically used. What is your vision for. Making armagh sin usable. How can people get their hands on the tools or software and start trying out the technique language sees when you make some ideas or make some. You have some jurors to solve some problems. You make this paper on available for example put the archive or just the first thing because you can share your ideas with everyone. One is a word. Count approach that year. The sickness were important and also recruiters. You should open and public your source code because for papers because this is quite simple the code is quite simple but if this is very complicated is not easy wishes related. Make it available under topic accessible because this will attract the waste of the hilbert to push this area forward. You'll cannot be as small minded which means you only keep your own scenes and developing developing. That's not good. The the better thing is open make more open source available and then we can push this area forward especially for forecast. Because i'm very interesting focused. Actually this i'm muslim many also of this paper. You can chick i. The first is made june national university of singapore while walk to their budget. What this Problems this forecast the first hour. The may australia's very interested in this digital flittering spectral chances but i'm very interesting forecast so we find the communists under developed this paper but go back to your question. I think the first thing is sharing rider sickness. Share your coat surtees shero problem which is beta. If we can't make this these everyone can put effort on this three. I think the machinery and the most specific Tanzer is focused com. Develop the fast. I think this is very important. Share your idea. Sugar baked code shero data. Actually that they cannot be shared too much and this moment especially those eight from real would. From companies there are some equal issues in some countries are budget. I think is necessary for researchers. Are either people quite interested in this area. It's very important to reducing great points in maybe on the last one we say share your data responsibly. And that'll cover it broadly. Yes that's good. Well maybe the wind up. I think we've covered the paper pretty well. Would you spend less time with a speaking a little bit more. Broadly about your work in forecasting in general are there other techniques or things that are going on. Now you're excited about you can shoot them. Google scholar page and propose new a framework focusing which is hierarchy before custody at coit out hierarchy learning this hierarchical forecast hierarchy on murray of fries down two scenarios firstly regressions taking these vacation the is just -tuation described. Just now well. I approached retail problems. Their point of sales problems. We look at the data. Viz find rich data which is because the data from point of sales. Machines is very vaguely. intel's u. s. k. u. is stood stock unique very specific description for item is still you once. They can on one specific store. You feel alive since that data is very difficult for you to alliance direction usually do some aggregate but if you the daytime high angrily will you can't focus. The independents were near. The problem is with this forecast results. You cannot make the decisions were because if you want for example you can focus their hurt. The total amount of the her us for particular product is useful. Forecasting results cannot approach your specific decisions. Such as human decisions for specific stole and demand fulfillment decisions are settlements. Taking sort of sense. You cannot make better if you focused very new angra gauge problems for causing problems for of this grow into aggregates focused bravo were reveals good but there is a relationship between this aggregate and high aggregates naval if you sum up all this new aggregate data. You should be equal. Chooses haagen nate how to use this relationship to develop some new focused in from framework and learning framework. This is my area and this regression problem for the problem is the same. Will you'll new candidate the item for example. You've kind of people better. If you just look on his body parts for exam look on his finger. You may sort okay. This is not his finger. This may be just a seminar products but if you look at the background of this as finger finger there is a hierarchy will which is the high anger they will close quench label should be consistent with no angry because in probably you cannot say you. Under the label of small awol. I ended up to highlight with started problem the raccoon learning or hiraoka focus. You can be applied time series because in times there is damage dementias. You can do the hirakawa scenes in first time. Second location or space survey will. He's the i two because this cease. I'm walking on very interesting. I'll be sure to ever lincoln the show notes to google scholar to follow up there. Thank you so much for coming on the show and sharing your work. Thank you thank you back. Includes another installment of data skeptic time-series. Thanks to our sponsor today quantum metric our guest chung show lee myself. Cloudy armbruster as associate producer. Vanessa bligh guests coordination and our host kyle polish..

Coming up next