Duration 26:41
16+
Play
Video

Simplify and accelerate AI for the entire data science team By Erez Barak, General Partner Microsoft

Erez Barak
General Manager at Microsoft
  • Video
  • Table of contents
  • Video
Request Q&A
Video
Simplify and accelerate AI for the entire data science team By Erez Barak, General Partner Microsoft
Available
In cart
Free
Free
Free
Free
Free
Free
Add to favorites
20
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speaker

Erez Barak
General Manager at Microsoft

Erez is currently a General Manager for Microsoft’s Cloud and AI division. Prior to Microsoft, he was the a co-founder and VP of product for Optify (acquired by Marketo) building a marketing software suite and a director of product marketing in HP’s Software business unit. Erez has been involved in creating software products in the past 20 years, with experience at Hewlett Packard, Mercury Interactive, Adwise, and Amdocs. Erez holds a B.Sc. in Computer Science and Mathematics and an MBA from The Hebrew University of Jerusalem.

View the profile

About the talk


Share

Perfect. Thank you. Thanks for the intro. And thanks for having me. I'm excited to be here as part of the global Big Data conference and I was thinking about a topic that would be most relevant to where I am. I'm seeing Maine for M. Enterprises companies in real armm thinking of that team was already started down the ml journey and is looking to do more. I'm going to share some thoughts in the next time 25 minutes or so joking about how to simplify and mostly accelerate that they are movement for everyone. They on the demon for the company.

Just one more question. If you met this is what the weather's looking like where I am. So you hear some Thunder hear some lightning, you know where it's coming from. Okay, so maybe for something about why why am I thinking about this acceleration? Why you should be considering Amendment how you do more and what aspects would lead you to do that. So I took out from IDC and just a question ask is and what kind of business impact do you think ml can make for your business now knows about an important. This is a large survey but happened back in March 2020 am

so a lot of the impact happened. We're actually seeing these drains and accelerate sense and your hearing and users of a guy saying where we believe about half of them. I don't know these things when we believe we can drive it better customer experience. And we believe we can throw a improve our employee productivity in a we believe that the basis we can accelerate Innovation using machine learning as a pretty strong belief. I am on why you were there Muse

and leverage machine learning and why you wouldn't even try to accelerate of the first and this is also an interesting View and I'd say we've been on this with a Microsoft on this path of the I-4 and that's not the general sense. Where are most most companies are so that most companies have been to in the past two to three years. Is this field what are UC applications for churning out of dicks in marketing or lead scoring in cells naturally virtual assistants for Services

risk management and demand forecasting for final scene operations and applications for a manual Workforce like a jar inside all relying on the need for ML or being based on machine learning. So it's not only about the impact you could possibly make it's also about those areas those applications and those parts of the business where ml is already providing a differentiation for ma your company. So lots going on in the question is when how do you do more? How do you push for more and again just to give some context and even if

you're not but that will work with me to do understand that this the people on the skull are already building models. The people on the school and already have some teams and resources in place allocated to machine learning and people are just call or looking to capture more value from machine learning than they are today now. If you're not here and assumption is yes, you'll get through just getting started face and you'll quickly get to the situation and I think many of us once we hit the situation a lot of people using your mouth.

OverWatch the best questions come up with how do we scale this? How do we accelerator? How do we take the assumptions? We started with and extend them to Moore sumsion. I'm looking at things that are key factors within the world of artificial intelligence and key factors for anyone who's building a machine learning model. I'm going to talk through and or quickly scan. I should say factors and starting points that relate to people resources and trust and give

some ideas on how you in the context of what you're doing can open up and try to do more and accelerator program every m Battery software program that they running machine learning a programming and then there's the set of people of our building it as a set of people are developing does a set of people are thinking about the customer scenarios serve them and many times you find yourself and you may be in this situation where this has been your assumption and

what is data science team and they're going to help out with all of this and maybe even do most of what we want to do for machine learning. And again, it's a good place to start is a good place to eat in extender and after you get started, but then the question is whether how do I accelerate this cuz it's the plan is more efficient and two kinds of applications and create the kind of impact. I talked about earlier and got them of you all know and you are probably feeling him. Pain.

Ashley the contacts of people What you wanna think about in the context of your program is how do you further extended type of people who build machines are they I give a few examples here looking at the data engineers and bi analyst how do you provide machine learning for machine learning capabilities for all skill levels for people who know the data inside out for people who are really deep into the date of times, but I'm not there. Mmmm code first time users. And how do you provide for the proper tooling for everyone in the proper

services for every1? Why you do that? How do you avoid silos and fragmentation? How do you make sure that the teams don't go into different directions and what you really want to do is also facilitated communication between all those personas and what you need to think about is not necessarily. How do I hire the next day to find us very important than continue to push on that but I Start thinking about when how do I wrap a degraded model using a wide array of algorithms you extend two more types of users you provide additional tools and now you're able

to rapidly create gravity experiment, which is really at the art of creating more. Take it from the court date of signs. You may have already building your nclu extent do more and more people using mm different capabilities to each other next question of resources, and I talked about it where you want to accelerate. You got to be thinking about this if this is your plan for a GPU and supported so desktop him under someone's asked him actually ever quite

the strong machine does 1 m equal only get you so far and it's very much a starting point for a lot of these users as you think about resources again, think of Continue what you can build okilly. Is it the local machines people are using is it the local compute cluster and your building could be a local AKA cluster that multiple users can use and what's your plan in terms of extending that to the Cloud solution or a multi Cloud solution? So at the end

of this what what I want to highlight this I see a lot of customers are accelerating their AI programs using multiple clouds and using with multiple local options to provide computer Firepower and also to provide access to data resources has read a reading they are to help you access your days off and also help use the computer. What you need to think about again in the context of your program is where you are today, which one of those Horizons you want to start pushing on and start pushing in that direction. You may

expand to a cloud vendor or you may extend your local computer capabilities. Keep in mind that the idea is to keep them connected to the idea again. You don't want to create as you want to continue until extender so you can open up the Moorea I but at the same time keep them connected. similar place where you might find yourself as hey, we're a great and I'm just using this as an example of a pipe sword shop and we're actually enabling our data scientist other users to use pytorch 4 ml but are you well know there is a lot more there in terms of

deep learning Frameworks tensorflow scikit-learn and others here on the slide them just a few to be mentioned and in today's world the ability to Merida spring works with a set of Hardware whether it's a mad mad men provided by Intel and using them abstractions like Alex run time to allow a good connection of deep learning Frameworks and the hardware solution are very much available. So again and again to it that you're going to learn what the right combination is for you but but I think the distinct in years if you're looking to accelerate you want to make sure your message does not look

like this New York team you want to make sure that your plan for driving all these business applications is not limited here, but rather opens up to these deeper and larger Frameworks. So again thinking how to break through people bitching how to break through resources last thing you really want to start establishing as you start accelerating. This program is the trust Factor know a lot has been said about doing a responsibly and a lot of still being developed in this area very interesting

framework and we can use and I've seen apply to multiple customers. Is this three-pronged framework around understanding? Controlling and protecting your assets understanding is all about putting tools and capabilities. Are there goes great open source tools for this great open-source capability for Destin. What's a great tools and applications from vendors for this to better understand what your models are about as you create more and more models that need to better understand in what they do which features matter what they push on and are they fair or

not. Are they delivering to buy us due to the training data or not that's need would become more and more pressing overtime. So what kind of models you're creating in the organization? Then there's two more chapters with one of the Run control. How do I how am I able to order the processes I have how do I able to provide the right metadata using data sheets to the daytime using so how am I able to control what I do and at the end of the day of these models deploy into production, how do I make sure they're protected from

Mmm malicious M or the protected from the wrong usage. So a lot of the deep recessions even in this conference around responsibility I touch on the different tactics, but as you think about pay, how are we going to accelerate our program against we text? Look at the type of people you're empowering and how we get more and more people into the fold. Look at what resources you put out their hand and what's your plan to extend those and then build the layer process continues to grow with the program? Know once you start

pushing through those in barriers if you make or those challenges if you make and you're you're getting into this semi mode within your organization with a lot actually going on and what you'll see which is not always great wall does the ladder of invention and innovation of Katherines that can be recognized a lot of things that can be m n m m set an alarm of learning that can be avoided for every team to take on and where you want to look around the organization of say, okay

what Catherine's degree recognized what kind of Frameworks can we give people on the team to start leaning on? So again recall that use cases and applications using ml today and I've been actively looking at a lot of customers and who built different applications where my running a lot of the analytics they were closed and looking really at 2 main address so not because a lot of activity going on. Where is it going with one eye call surround augmenting analytics with ML. Probably the organization you're in have the goods and set of data

probably has them a lot of them in understanding and usage of that data. How do we build on top of that? I didn't use machine learning to drive. Even further. The other one is the organization have the need to create more applications and solutions for your own custom. Dose Solutions open to be differentiated and to be impactful to customers will requires ml powered feature. How do we get there so I can spend the next five minutes or so talking through those two patterns first around

and run forecasting same thing could happen not just for inventory, but for your whole supply chain. Think of someone is looking to optimize their campaign. This would be an ad campaign just to be any marketing campaign how to drive that optimization. Do I get all the data is there in our system? But how do we apply in algebra? And the last one is as an example of someone running low on processing and it has to all the men the inside they have on the loan on the mmpi-2 want to take the lawn? How do we further enhance that so the

high-level is want to think of three items here and this brings me back to him. They need to extend this to a med Benito send a machine learning program to more and more people in their organizations could cuz you can see it's not The data Engineers are going to need to lean in here in the big data and data warehousing Solutions are definitely a part of this at the base of the driving this program. And then the data scientist are able to build brain on the floor models based on that data and at the end of the day if you want to make sure that the

visualization capabilities that you have in the organization's are also tied into that because there were your real delivery mechanism for all of this. So, you know, there's going to be a few and I can 0 mm examples of systems. You may be using for that or could use for that but they're also more symbolic of pay for data engineer and regardless of your big data warehouse store in your big day. Solution and something for data scientist regardless of your solution for building and training and deployment house and regardless of how you fish walleyes those need to work.

Concert in order to augment an Olympic. So I think first and foremost. I think why am I doing this doesn't matter after a fashion and if it master or bachelor in these are the people and systems you want to bring in start solving that problem. Great example of God that you could read more about it than play cat. I'm sure Raymond to stick will become available later and Cara team and was looking to open an additional retail locations and you really want to get this right? Right you want that stores to exceed the revenue plans and again by using their

Rich analytics, but using inside they're already have and buy augmenting that with capabilities and made from their machine learning team. This team was able to achieve a whole lot more of a best-performing stores in the district Steakhouse. So again, then that will be one padron where a lot of the ambuscade this map into the other one super interesting as well as existing at this could be a few examples I pulled in here could be about a time is driving

like a good example where you see a lot of ml power then I'm sure people can relate to but it's not always that as a relevant to their day-to-day fish farm monitoring is an example of took more about later but but you can think about them and generally and Monitoring Solutions whether I have a farm without a wind farm fish farm with I have another event facility. I want to monitor definitely that would be great that milk out without and predictive maintenance and app that allows me to What maintenance would be needed saving time and money? I got a great example that was around sustainable

farming and extensive applications customers. So again now it's a little different. I want to use that Eminem or that accelerated ml capability. I just put in the organization but it's not about taking my hands to get but rather building an mlo power the app. You really want to consider these as layers with free users. You want to be sure that you're serving that use cases and needs they have but but then those needs typically map to that serving layer a few recommendations and how you build that user interface what business logic you put in place. And again what they die systems of

Records, you're going to bet in systems of record you going to build this on machine learning here at the heart of the system, but this is a pattern you'll often see the last thing you want to do is have every team so don't think when we're done with the model now, how do we get this in? Vacation you'd rather want to have a more dodging eyes be okay, if you're building in ml powered up if you're building a future for an existing power for the existing out. That's why I buy ml you want to make sure you have the proper and blueprints and repeatable processes again one that fits your

organization and one that aligns with the m m quality accuracy and customer promise you're making and turns up and what's your delivery and then I'll finish up with them one more example, and this one is from an APB which I mentioned earlier. I got the full story of clicking at this link will allow you to read that. 3 a.m. This one is at the base. This is creating a new solution for their customers and their customers wanted to help Ben and maintaining and commitment to a food future somewhere and wait to see

them do this. But at the end of the day again similar, but there's a set of empowering the features their users need those ml models are done built into that pattern and serve two users. Okay, so we're just about at the time in summary. I think three things do you think about as you want to accelerate your program stay where you are in terms of which people are empowered which resources you put in place and made available. And how much can you press the game system you put in place. Once it accelerates. I touched on a few Dimensions those dimensions of the

flight difference between different organizations, but definitely still event at the island of Dimensions to think about what you got the accelerated machine going once you got through those that bomb XR as you do that really try to find okay for every use case I built for every my light at Biltmore which pattern does it fit into is it going to MLB? We got a blueprint for that. Is it going to Augmentin your analytics for that mail? Good. We got a blueprint for that and I realized what I do. More patterns that would

come up there and ask them if you work within your organization. So I think I C Enterprises end up with two maybe 5, there's always exceptions in other ideas, but having that kind of blueprint people can latch onto helps you again accelerated program and meet the demand and the city going to come your way. Especially if you're in the machine learning business. Okay, I think I got we're just about... I'm excited to maybe take a few questions if we have a I'm trying to

say is broccoli good presentation. You still have time. I know you still have a work 3 4 minutes. You can run over few more minutes because the next presentation is at 12:10. So if anything else you want to say to me or what I wanted to do really ever want to keep in mind is that them especially as you build your machine learning program and I've touched on a fuel sensor fail if you met roadblocks or bottlenecks, you may hit that idea at the ideas I get here and I seen

them generally applicable in many areas, but the way they're created in the way they remember Mamma manifested in her organization is quite different. I'm also seeing this, they were somewhat happy to see how do you say they got in jail still giving importance being honest and they'll enjoy myself. I can be rest assured that my job was not going extinct. I think I'm not the other day to find this. So I think we're seeing The strand of a lot more rolls in a lot more MSM types of background building ML and don't capabilities are now

becoming more and more available more and more widespread. So you don't or my messages. Hey, nothing about becoming a certified this and then that's the way you do in middle. But rather within the word you're doing whether your data engineer whether you're an analyst, how do you remember getting power to do ml ml driven programs as you do your job? I'm seeing a lot more happening with machine learning become inaccessible do more and more type. So that users rather than saying

well if you're not doing machine learning mmmm, you going to have to find a different. That's definitely not wear that Trend or Direction and that I'm saying is going

Cackle comments for the website

Buy this talk

Access to the talk “Simplify and accelerate AI for the entire data science team By Erez Barak, General Partner Microsoft”
Available
In cart
Free
Free
Free
Free
Free
Free

Ticket

Get access to all videos “Global Artificial Intelligence Virtual Conference”
Available
In cart
Free
Free
Free
Free
Free
Free
Ticket

Similar talks

Sudip Chakrabarti
Venture Investor at Madrona
+ 4 speakers
Ian Foley
Venture Partner at Level Ventures
+ 4 speakers
Anupam Rastogi
Partner at Emergent Ventures
+ 4 speakers
Vik Sasi
Partner at Dreamers VC
+ 4 speakers
Dennis Liu
Investor at Ford
+ 4 speakers
Available
In cart
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “Simplify and accelerate AI for the entire data science team By Erez Barak, General Partner Microsoft”
Available
In cart
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
566 conferences
22974 speakers
8597 hours of content