Make Azure AI Real: Train, Debug & Deploy Responsible AI Models with Azure Machine Learning (Part 1)
- February 28, 2024
- Posted by: MainInstructor
- Category: BASIC Business Strategy Data Science Go Microsoft Azure Python
Video Title: Make Azure AI Real: Train, Debug & Deploy Responsible AI Models with Azure Machine Learning (Part 1)
Hello everyone and welcome to episode three of the make Azure AI real series and today’s lesson or Workshop or whatever you want to call it conversation I don’t know uh we are going to be covering the almost important in to topic making AI responsible using AI resp responsibly
Building with AI responsibly all of those things and we have probably one of the best best people to do it uh Ruth how are you H how’s it going hi Corey you’re too kind I’m doing good happy to be here very nice well Ruth this is your
First time even though we’re going to be doing something again tomorrow but we normally start off with a bit of a random question but not always random uh so I’d love to know in the audience but I would love to ask you as well uh what is the scariest movie you’ve ever seen
Oh really you’re going to surprise me with a question yeah this wasn’t inappropriate this is welcome to make asure and I real we have to be as real as possible it’s not all scripted okay that’s interesting um I think for me The Exodus tops everything Exodus haven’t seen it
I’m actually not uh so just in context for anyone watching this in the future this is today’s halloween for those who celebrate so we’re trying to Ste on theme Here but I’m not a big horror film person but I say the scariest movie I’ve ever seen is the ring uh this what the
Ring have you seen that one oh my God I haven’t watched that don’t don’t watch it I this is a uh backstory but I used to like if I was really scared growing up I would have the TV on to go to sleep but in that movie not a huge spoiler
Because if you’ve seen anything about it a girl actually climbs out of the TV so it kind of ruined the whole uh use the TV as a security blanket for me uh so so I’ve been now you always sleep with with the TV you make sure the TV is on yeah
Well now I have to turn it off because I don’t want any like you know ghost coming out of it but let us know let us know in the audience we’ve got a few people here watching on the stream so yeah what’s the scariest movie uh you
Have seen in the audience but second question rude for you we always ask this for a guest uh what does making AI real mean to you making AI real yeah um making AI real for me is um how can we make it practical um there are a lot of theories
Um out there um but making a real is like okay how can I take uh Ai and make make a difference make an impact so I think that’s kind of relevant to uh responsible AI um because it seemed like for the longest time it seemed like it
Was just maybe a slogan principles but at the end of the day people are like okay how do we actually do this again so for me I think once you give people practical stuff yeah nice imp yeah bring it to life exactly well said our little
MTO that we have here and I completely made it up on episode one but it’s uh make as as your AI real we don’t talk AI we do AI uh so great answer yeah great answer and uh speaking of answers we got a few uh par uh about the scariest movie so Paranormal
Activity again uh one I only watched the trailer I haven’t watched the full film because I’m already scared of the trailer like you know what even the title is scary me yeah it’s funny say that because this one doesn’t seem like a scary I don’t know if this is like
Ironic answer I’ve never seen or even heard of this movie good morning from Seattle I don’t know if that’s actual scary or not or you’re just sounds it sounds welcoming yeah I don’t know if it’s just a bad movie is what you mean by scary please please clarify Rob for us uh cool
So you know what is not a scary movie of course is that uh you know all these movies about AI taking over the world taking uh you know these things because everyone of course is going to watch this workshop and we watch our topic and they’re going to build responsible AI
Systems and we won’t have to worry about that uh but before we begin and give the floor to you Ruth uh I just want to make a quick note we actually uh just launched our generative AI for beginners course to today actually and like an hour ago and and we actually do also
Have a responsible AI uh lesson which Ruth you have gracefully uh contributed and edited my crazy writing on that topic so I really appreciate that but do check out that link below um and you can do it to 12 12 part uh lesson is a get up repo completely free so if you’re
Looking to we’re going to talk about Ai and ml activity almost in general but if you’re also looking to do uh some Genera of AI work uh that’s the link for there but that’s enough of me going on because like I said we don’t talk AI we do AI so
Ruth please tell us how can we do AI responsibly um yeah that’s a good question um I’ll make sure not to scare you guys on how yeah please don’t make a responsibly um let’s see let me get my slide up so um today I’m going to be
Talking about um how we can train debug and deploy um Mo uh AI models responsibly using responsible AI um and to do all of this uh we’re going to be doing end to end uh uh demo so that’s why this is a two-part um series um so today we’re
Going to cover how to train your model um how to use a responsible AI dashboard which is a very cool user interactive interface that um you can debug your model and you’ll be blown away the things that uh it’s able to uncover so the good thing about uh the responsible
AI dashboard is um you can train the um deploy your models in Azure machine learning and there’s also open source version as well but today we’ll focus on um Azure machine learning nice yeah so before I get into the nitty-gritty of uh responsible AI um let’s talk about the
World today so as everyone here uh is aware AI Innovation is occurring at a very rapid Pace um yeah the Innovations or breakthroughs in AI is mind-blowing especially with that generative AI um and open AI That’s thing um another thing that we’re seeing is um a lot of companies are
Accelerating their adoption of AI so it’s not just uh the uh consumer product that uh uh a lot of companies are utilizing AI but they’re also using AI internally you hear of uh co-pilot uh AI Solutions um that’s to help um enhance uh business processes and make U the everyday
Mundane task um easier for um the workforce like the employees U so they can concentrate on um things are more important another thing that is U we’re seeing is societal expectation is also evolving um I feel like in the last year or so um responsible AI has been on top
Of mine um in every conversation um also um everybody and their grandma is talking about AI responsible AI you’re like Grandma now you too yeah Grandma how do you get access to chat GPT I don’t know how you did that exactly um then another thing we’re also
Seeing is um governments um Talk um talking more about AI um about um stepping in and uh figuring out ways to regulate uh Ai and response so the question is uh um when we talk about responsible AI what do we actually mean this is perfect Ruth great transition because we got one uh
Question from the audience already uh what do we mean by responsible AI you know what I like this person I like this whole a this audience Ruth I should have told you this but since we do AI here people are always ready to just to jump
On it to get the answer start building and I love like this was like the opening question yeah that is excellent question because I had the same question three years ago it’s like okay what is this responsible AI um and you also hear of uh ethical AI uh is it just a slogan
Um that sort of thing but um at least for Microsoft when we’re talking about uh responsible AI they’re actually um six different principles that uh we look to practice and also so um some of our um clients and partners have adopted so number one is fairness uh fairness is
Basically hey when you’re building a AI solution uh making sure that uh it is uh considering um not just one set of people but um a multitude of um people that have similar um attributes um in the use case that you’re dealing with um so making it fair and
Evenly um use across uh different demographics um the next thing is reliability um and safety so when we talk about reliability uh we need to make sure that okay when we’re building AI solutions they are reliable they are safe and they are consistent so it’s not
Okay um every now in this situation um there’s a disclaimer be careful and in another dis uh situation it works well well it has to be consistent and also for developers that are um developing um these AI systems to ensure that um the AI um systems Works in normal situations and abnormal
Situations um so let’s say um you’re talking about the electric car um you test it all throughout in the daytime but did you test it at night um did you test in different weather conditions uh that sort of thing then we have privacy and security so with privacy and security is
Basically what exactly what it is uh we all know getting um data for uh training AI models is very hard so number one uh when we’re dealing with data where did we get the data uh is is it from a credible Source can we trust that data um that sort of thing then
Even within the data um what’s in that data does it have uh people’s private information pii information um are we like how we handling that then with security um it’s always there’s always uh what’s it called a vulnerability uh with um somebody coming in and messing with the data um similar to
The uh generative AI um course that um Corey just mentioned one of the parts um that um he covers is also making sure when we create um prompts somebody can actually jailbreak and mess with with the rules um and your prompts are suddenly behaving differently or grab information that they shouldn’t so um
That’s another example another example excuse my um squeezing of the screen but inclusiveness so one thing you may be thinking well inclusiveness isn’t that covered in fairness um it does kind of overlap but inclusiveness um just think of uh when you’re building your model always think okay are there other
Demographics that we’re missing so perfect example is um there are a billion people that have uh disabilities so when we’re building AI Solutions um we may consider that okay um we are checking the check box um that okay we’re covering accessibility but um some people may think access disability it only comes
With maybe somebody who’s blind or death um but there are other uh disabilities um that uh almost every one of us may have like color blindness uh being dyslexic um there’s a whole bunch of things um reading fast versus reading slow um so those are different things that um when we’re
Building AI solution we have to think of uh the make sure we’re covering all the demographics transparency is a I think there was a really good quote sorry to interrupt that I I’m gonna butcher this quote so I don’t know why I’m interrupting you to even butcher the
Quote but it was something like we’re all a part of an accessibility group it just hasn’t happened yet like you know as we age how we interact with technology is going to change right so I think like most people you know who are you know fully able-bodied or don’t
Think about that you need to bring like think one day you will probably have something that requires you to be included in a system design or a product design or user interface so it’s super important to your point to like you know build systems that are doing that
Especially in the AI side of things I like that example because that’s totally true like the way some people text so fast I’m like is this a keyboard and I can’t even do that if I wanted to um then the next part is transparency um we’re building our models um we all
Know AI models are blackbox um we need to understand um how it goes about to make predictions when something goes wrong we need to know exactly what caused it to be wrong and it’s not just how the AI is performing um it’s also when we’re deploying our AI systems we also need to
Disclose um okay what uh in what situations should we use this model um in what situations does this model not work work too well so disclosing all of that information to the general public also brings uh trust and also when there’s U regulatory things uh when you’re being audited uh transparency is
Also an area that uh is heavily involved the accountability as we’re building all these Solutions we need to be accountable for what we build um whether something is uh goes good or bad I I think that’s a great Point uh this is kind of a funny one H wrote uh it wasn’t
Me it was AI Shaggy but like that is really play onto things like right some people like to blame the model or blame the eye for generating something but obviously it’s us that needs to be accountable Whoever has built the model the model in such a fashion so you can’t
Say it just wasn’t me and run away yeah I like the comment thank you yeah I like that it wasn’t [Laughter] me okay so the question is um well we’ve already been debugging our models when we train it so let’s say we have classification um a classification model
We have all these performance metrics that we’re looking at um and when you have a regression uh model you also have other metrics so um if we already are checking performance um why do we need to debug any further so the question I’ll ask um
Us the audience and you and I is let’s say you have a model that is 93% accurate um the question is what’s going on with the 7% um that the model does not perform so well um what demographic does that fall in because that could be a crucial uh
Demographic that um you’re actually overlooking and all of these metrics that we’re looking at how do we know how to catch any let’s say data biases how do we know how to catch any security issues um how do we prove transparency and whatnot so as you can see um we need these uh
Metrics but we need to go a lot deeper into the data and the behavior of the model to truly understand how it is working and where the issues are excuse me so one of the exciting things uh that um uh we have is uh the responsible AI
Dashboard um and this is um actually uh a collaborative U solution um this is something that uh Microsoft has uh uh contributed to a lot of um companies out there kind of like uh Fair learn interpret ml um era analysis. and a lot of uh research uh
Institutions have also um come up with solutions to debug AI solution because we the data science Community understands that this is a serious crucial area and this is actually something that data scientists actually want and need um problem is there are no tools out there uh um to um help um them
Uh debug um their AI models and it’s not like um when we hear about responsible AI nobody wants their model to be uh responsible or nobody even wants their model to uh maybe be exclusive so the good thing about this dashboard is all of those different uh technologies that all of these
Institutions have put together they do one um single thing separately so even if you’re to in invoke it in your python code to debug when you do that um you need to bring in another library and do other analysis and whatnot so it’s hard to really uh merge everything together
To get a big holistic picture so what this dashboard does is it puts everything all in one so you can see how things are going so the very first part is uh error analysis so error analysis is just identifying even if uh your model says Ah it’s 99% accurate um using
Aor analysis it’s able to find areas where your model is not performing well which U data and your Dem data demographic is not performing well also data Explorer uh it looks your data representation do you have over representation or under representation that’s where a lot of biases come from
Then model overview um using the same traditional metrics that we have um fairness assessment uh model interpretability that’s um like explaining um how the model is behaving counterfactuals and what if that’s when because it’s good to see how the model is performing but you also want to test if I
Change this in the data if I change this my data a little bit how is my model going to behave so those are good ways to um test how your model is going to be behaving especially in um situations that may not be normal um that sort of
Thing but with um this you’ll be able to do those U type of debugging so um I’ll say all of this is uh things that uh Engineers would do on a day-to-day basis then on the business decision side um it’s uh a lot of business uh stakeholders who utilize even the Water
Analysis to make decisions and also causal analysis so if they’re planning to roll out uh changes uh before they roll out um they can do um analysis almost like AB testing but you’re not um spending the money or allocating the resource you can test things out before
Rolling out to business strategy um to see how the let’s say consumers or your users are going to respond to it so very quickly I just want to say I really like your point about no one like kind of intentionally goes out and you know Builds an irresponsible model like
No one’s going to like the irresponsible like irresponsible models Workshop that we had yesterday you know it’s like so but we’re you know building the tools and things like that to make responsibility much easier to achieve uh which I think that’s a really good point um to get across to everyone yeah thank
You yeah because it actually especially if you’re a business it kind of it canect effect your reputation too so for sure okay so I think I have three more slides then we’ll get to the coding um so when I was um just to give you guys a little bit
Uh visualization into what I’m talking about um this is a normal way that we train our model we train our model we do the um analysis performance analysis with with the traditional uh machine learning metrics and it will show us uh accuracy like okay this model is 89% accurate we’re
Like yay finally but realistically this is how your model looks um there’s a demographic um let’s say the one in red you don’t know that this demographic is only 40% accurate uh this one is only 59% accurate this one the model is only uh uh accurate uh 79 79% of the time so
That’s the reason why it’s good to use uh a debugging tool um so this is a snapshot of the dashboard um one of the um sections of the dashboard so um this is how it anal it um analyzes your model and the red hot red spots are the areas
Where there are issues so the darker the red the higher the error rate another thing I talked about is um data representation so I think this image is um a pretty good one so imagine you have a whole this is all your data set there’s um there’s some
Demographics uh in this um big circle over here um that’s a representation of some of your data and some demographics fall in this little data so as you can see there’s an imbalance in your data representation so guess what when you train your model your model is probably
Going to be more favorable to the popular uh demographic the demographic that falls into the bigger ball and the ones that are under represented guess what um your model is not going to be favorable so the good thing that this um dashboard does is you’ll be able to um
Analyze your data um and see um how the model is behaving um with certain demographics like uh and looking at U the data distribution um of its behavior and knowing that okay um the model is favoring this demographic versus that demographic so we can go to the final
Slide so the last U part is kind of like how I mentioned um AI models tend to be a black box so um it’s hard to uh really explain um how um it’s going about making it uh predictions so what the dash uh board actually does is
It make it makes it like a glass box so it’s you can see and understand exactly what is going on so uh one of the features is um let’s say you have a whole bunch of features um these are the data sets um that you have um you have
The ability on a global level to see okay what are the top 10 features that um the model is uh uh that are driving the model’s prediction so what are the top um I’ll say five uh features that are influencing the model’s uh prediction you’ll be able to see this so
Let’s say this is like a loan uh applic model and you see that okay the top five driving factors that uh the model uh takes into consideration to approve or deny somebody’s um loan is their gender and their race so if you see that okay maybe like one of those uh sensitive features
Are one of the top features automatically you’re going to know that okay this model is not beh behaving the way it should be because it is uh uh basing its uh predictions on a bias uh feature another good thing is even though you see these um top uh five or top 10
Features when you look on U the dashboard gives you the ability to also go drill down on the individual level and be able to identify exactly for this individual what made let’s say the model reject their um loan versus another uh individual that has the same characteristics so um maybe the
Top uh 10 or top five criteria are not always going to be the same as this Global one because each individual is different so um each individual is uh the way the model is uh making it predictions different different so you’ll be able to Deep dive and this
Really comes in handy uh if you’re being audited um so I use financial services but let’s say you’re in health care and you’re using uh AI to make um diagnosis uh and you get audited uh okay what were the driving uh features that made you diagnose this patient to have this or that
Um so we can uh go to the code let’s do it I’m bringing out the code you have any questions yeah let’s know if you have any questions we just dropped the link there in the uh in the chat as well as on below I can’t if
I point down I’m just pointing the youth so that’s not official Ruth can you point down to the link so everyone knows that that’s the link right yeah there we go there we go that’s the link so I know it’s probably hard to see on the maybe
On the slide there but there’s a big link there uh so we will walk through uh this repo but this is uh we always keep this question when we do these like is this being recorded or is this recorded so the answer is yes this goes live
Again on the reactor YouTube channel uh so you don’t have to necessarily follow along now like don’t get stressed you can obviously watch this back again uh as you walk through this uh demo but that’s where the repo is and the code I see your screen is now up re so
We’re getting started okay awesome so what I’m gonna do is um show you guys because I’m sure you’re wondering okay I have my model but how do I get my model from point A to Hero like how do I get however way I went about uh training my
Model how can I get it uh to the dashboard so the very first um um script we’re going to run all is doing is is grabbing data and training um a classification model so why I’m explaining we can have this running so the data that we’re using is
Uh data from uh UIC University of California um I think Irvine um and basically is uh data about U diabetes uh patients so what the classification model what is trying to do is predict if a a patient is discharged from the hospital what what are the chances of them being readmitted
Back to the hospital within 30 days so after we train this model is predicting okay this uh patient is going to be reemitted um back uh maybe after 30 days or um not or they’re going to be reemitted within 30 days which is not good so the very first thing that we’re
Doing is um um calling reading the input data um that we’re already split into training and test data and Save in the paret file um so if you guys notice um we’re um extracting the Target Field so there’s a app attribute called um or a feature called Reit status and that’s a status
That says yes or no somebody’s going to be reemitted in uh within 30 days um later on you see how this is important because since this is a classification we need to tell uh the dashboard um what field we’re trying to decide on then the r is uh pretty straightforward we’re
Getting our training and test data um then um normalizing um the data so we’re doing our encoding since we have a very mixed audience can you tell us a little bit about why do we normalize data um so your data can come into different forms um can come into text it can come
Into uh large numbers small numbers so you know when you’re training your model you want it to be in a certain range that um your model understands um so that’s why um you need to and it also needs to know okay this is numerical this is categoric uh a string that sort
Of thing so those are some of the things are as good to um normalize your data um when you’re training it and it helps U the training process go faster as well um so these are the pre-processing um Step we’re um doing the encoding um and we’re putting it it
Into uh a pipeline so each time we um and we’re using a logistic uh regression so each time we try to uh train the model it already has all of these sign coding steps um and it will just run the pre-processor whenever we need to train additional data so um after training the
Model uh we can see that oh uh for this classification model um the accuracy score is uh 0 839 uh which is pretty good it’s almost like the slide um so the next thing is um the Azure machine learning comes with the SDK um as you can see I’m running this
Uh from my desktop so the good thing is um you can um install this uh the SDK for Python and it uses uh the ml client so if you need to authenticate into your adml workspace um this is how to go about authenticating um it use it basically um
Depends on a config um Json file which um when you’re running it um yourself um you need to copy from the Azure machine learning workspace um you download it and it has three fills in there your subscription um ID your Azure resource uh name and uh Resource Group and um the
Azure machine learning uh workpace that you created so all is doing is connecting to that workspace um in the cloud the next thing that um the aure machine Learning Studio also does it puts everything organizes everything for you so you just read your data um it has uh a
Data uh function and basically what it does is uh it takes your data and stores it into a data set the good thing about that is let’s say you’re working on a team and um you all need the centralized data to do run different experiments good thing is the data is cleansed it’s
In one place you guys don’t have to move the data around each time somebody needs to run a different um experiment or training with it um all you can do is um Point um use the pointer um to where this um data set is stored so that’s uh
Very useful um the next thing is we’re um creating a compute um cluster um that we’re actually going to be using to run our training model so that’s pretty straightforward um spinning up a VM uh instance um the next thing that um AER has is something called jobs so jobs are
Very handy um because as we’re um training our models we usually do multiple things so say you grab the data um you cleanse it next thing is okay you have to do the preprocessing steps um encode it then train it and all of that so the good
Thing about um creating a job is you can have a pipeline a job Pipeline and each task is a component so the data cleansing part is a component uh the training part is a component the deployment part is component so um for this one um since we have the
Data uh already saved uh what we’re doing is declaring what the experiment is um also specifying what the compute is um also specifying okay where is the data located what type of uh data uh are we passing so we’re uh passing a URL um or URI um path um where’s the source
Code located like um when we train that U model where are we going to store it um then you also have you have the option of um creating your own environment or a you also creates some curated environments um it’s just uh whatever uh dependencies that you need you can put
It in the environment then uh what we’re doing here is executing the command to actually um take the python training training code um with our training data um the Target Field that we’re using for classification and also the output that we want where we want that model to be
Saved so once we have that um you have the client um you have the job um so you submit the job um to actually um train the model um so this would take a few minutes but while we’re waiting let me see if we can see the activity on the
Dashboard okay so when it’s running um these this actually creat different artifacts um in your as you’re training so you can actually monitor um the progress of the run so the good thing is it just uh finish training then the next thing we need to do is register um the model
Because okay we’re satisfied with this model so we need to register it in the Azure uh machine learning studio so each time we need to communicate with the machine Learning Studio we’re using this ml client um then specifying okay this time I’m dealing with the model um and
These are the things that um in order to store the model you need to pass in so the name the path um and also um use mlflow um attributes as well so U we’re basically creating the model so that’s um done so everything that we just did here um really is not that
Different from what you do uh because you can easily um train your model and once you have your model you can just uh take it and just uh register it into the aure machine Learning Studio and that would be good as well but this one is a
More structured organized way of um um dealing with your model so the next thing is because this is a big question that a lot of people have I have my model how can I get it to the dashboard so now we do have a model the next thing question
Is how can I get um create the responsible AI dashboard and how can I utilize uh the dashboard to debug U my model so s similar thing um you authenticate um to the Azure machine Learning Studio so we’ve done that before you have the model name um so we’re dealing with hospital um
Reemission so we give it the name um we also are getting remember we the other um time we called um the ml client with models were actually creating now we’re actually retrieving that model back um so we retriev the model the next thing that uh the dashboard needs to do is
Figure out that okay out of all your data sets I want to know which ones are categorical which ones are strings um so basic Bally this function all is doing is going through your data set and figuring out uh which fields are numeric and which ones are categorical which are
Strings and um identifying those fields um so that’s pretty standard this is a part that we’re actually uh starting to um Define our dashboard so the very first thing you need to do is the dashboard comes in different sections so you can pick and choose uh which
Sections you want uh to be included or excluded from the model um it comes with a component um and remember I mentioned component are one task it does one task so it’ll create one module of thing of something so we have a comp component um for the construction
Constructor uh we also have uh a component for explanation that’s explainability so you’re basically telling the dashboard that I want to be able to use the EXP explainability um section of the dashboard to debug my model um the next component is uh error analysis so you also want to do um error
Analysis um then um Gathering insights is something that uh it also does uh especially for the uh model overview so these are the key components that you need to specify that um you want to be included in uh the dashboard so once you get those because mind you all of these
Components are already pre-created within the Studio you’re just uh retrieving those uh components so you don’t need to write any code to um actually build that individual component Okay cool so we’re covering up on the hour Mark Ruth I know we have uh tomorrow as well we’re going to uh
Continue on here so I don’t know if this stop and then uh encourage everyone to join us tomorrow the link is below Ruth I can’t get to the pointing uh but uh can I I have two more minutes yeah okay go ahead all right so the last part is
Actually um the you need to build a pipeline so the pipeline is what uh is going to build your dashboard so basically you’re passing in the the column that you’re classifying your input uh are the data uh this is um defining um what the um dashboard is
About so remember each of the different insights the components that um you selected um these are just um different um um parameters that you need to um specify so um you need to tell it as a classification what type of model it is um you give it a unique name then you
Know you pass in the model the D data set the Target Field um then the rest of the fields are pretty standard um so you just passing them and um all you did was uh Define what your dashboard is um you submit it into
The data pipe line so this is a job to actually create um the dashboard um then you’re also retrieving um um some of the metadata um like U the uiux U metadata to also will be included in the dashboard so that is it um also one thing to point out is as
Your model is running um you can also pull up U the pipeline and look at U the progress of the model so everything that we’re talking about components pipeline visually you see exactly what is going on so that concludes um today’s session uh we learned about training your model
Uh we learned about uh creating the dashboard so configuring the dashboard with the test data your model and the Target Field if you’re dealing with the the classification model so tomorrow we’ll actually go into details of uh going through the dashboard and actually using all the different fields to debug your
Model perfect thanks a lot Ruth everyone in the chat looks like appreciate it there’s uh oh I uh we had a little Fu there but they say thank you in the chat if you see it uh and yeah we have a a little clip finger so uh see you
Tomorrow uh where we continue on building responsible Ai and like I said we don’t we’re talking AI we’re doing Ai and we’re doing AI responsibly so thanks everyone for joining you could have been anywhere on the internet right now but you spent an hour with us so for that
I’m truly appreciative and thank you Ruth thanks everyone bye bye
Video Keywords: Microsoft Azure
-
Sale!
Wireless WIFI Repeater Extender Amplifier Booster 300Mbps
$29.99$14.99 Add to cartWireless WIFI Repeater Extender Amplifier Booster 300Mbps
Categories: Electronics, Wi-Fi Router, Wireless Wi-Fi Extender Tags: 300Mbps, 802.11N, Amplifier, Booster, Extender, mobile wi-fi booster, Remote, WIFI, Wireless, Wireless WIFI, Wireless WIFI Repeater, Wireless WIFI Repeater Extender, Wireless WIFI Repeater Extender Amplifier, Wireless WIFI Repeater Extender Amplifier Booster, Wireless WIFI Repeater Extender Amplifier Booster 300Mbps$29.99$14.99 -
Sale!
Full RGB Light Design Gaming Headset Headphones with Mic
$24.99$14.99 Add to cartFull RGB Light Design Gaming Headset Headphones with Mic
Categories: Electronics, Gaming, Gaming Headsets Tags: Design, Full, Full RGB Light Design Gaming Headset, Full RGB Light Design Gaming Headset Headphones, Full RGB Light Design Gaming Headset Headphones with Mic, Gamer, Gaming, Gaming Headset Headphones, gaming headset wireless, Headphone, Headphones, Headset, Light, Mic, Package, RGB$24.99$14.99 -
Sale!
Wireless BlueTooth Multi-Device Keyboard Mouse Combo
$39.99$19.99 Add to cartWireless BlueTooth Multi-Device Keyboard Mouse Combo
Categories: Electronics, Gaming, Gaming Keyboards, Keyboard Mouse Combos Tags: Combo, Keyboard, keyboard mouse combos, Mouse, MultiDevice, Set, WireKeyboard Mouse Combo, Wireless, Wireless BlueTooth Keyboard Mouse Combo, Wireless BlueTooth Keyboard Mouse Combos, Wireless BlueTooth Multi-Device Keyboard Mouse Combo, Wireless BlueTooth Multi-Device Keyboard Mouse Combos$39.99$19.99 -
Sale!
High Back Leather Executive Adjustable Swivel Gaming Chair with Headrest and Lumbar
$199.99$139.99 Add to cartHigh Back Leather Executive Adjustable Swivel Gaming Chair with Headrest and Lumbar
Categories: Gaming, Gaming Chairs Tags: Adjustable, Chair, computer chairs, Desk, Executive, Gaming, Girl, Headrest, High, High Back Leather Executive Adjustable Swivel Gaming Chair, High Back Leather Executive Adjustable Swivel Gaming Chair with Headrest, High Back Leather Executive Adjustable Swivel Gaming Chair with Headrest and Lumbar, High Back Leather Executive Adjustable Swivel Gaming Chairs, Leather, Lumbar, Office, Racing, Swivel$199.99$139.99 -
Sale!
Professional LED Light Wired Gaming Headphones with Noise Cancelling Microphone
$29.99$19.99 Select optionsProfessional LED Light Wired Gaming Headphones with Noise Cancelling Microphone
SKU: N/A Categories: Electronics, Gaming, Gaming Headsets Tags: Cancelling, Gaming, Gaming Headphones with Noise Cancelling Microphone, gaming headset, Headphones, Headset, LED, Light, Mic, Microphone, Noise, Professional, Professional LED Light Wired Gaming Headphones, Professional LED Light Wired Gaming Headphones with Noise Cancelling Microphone, Wired, Wired Gaming Headphones, Wired Gaming Headphones with Noise Cancelling Microphone$29.99$19.99 -
Sale!
Gaming Desk with LED Lights USB Power Outlets and Charging Ports
$349.99$249.99 Select optionsGaming Desk with LED Lights USB Power Outlets and Charging Ports
SKU: N/A Categories: Computer Desk, Gaming, Gaming Desk Tags: and Charging Ports, Charging, Desk, Desks, Gaming, gaming desk with led lights, Gaming Desks with LED Lights, Home, LED, Lights, Monitor, Office, Outlets, Port, Power, Room, Stand, USB, USB Power Outlets, White, Workstation$349.99$249.99 -
Sale!
Wired Mixed Backlit Anti-Ghosting Gaming Keyboard
$99.99$79.99 Add to cartWired Mixed Backlit Anti-Ghosting Gaming Keyboard
Categories: Electronics, Gaming, Gaming Keyboards Tags: Antighosting, Backlit, Blue, brown, Gaming, Gaming Keyboard, gaming keyboards, gaming keyboards and mouse, Keyboard, Laptop, Switch, Wired, Wired Mixed Backlit Anti-Ghosting Gaming Keyboard, Wired Mixed Backlit Anti-Ghosting Gaming Keyboards, Wired Mixed Backlit Gaming Keyboard$99.99$79.99 -
Sale!
Wireless Bluetooth 5.3 ANC Noise Cancellation Hi-Res Over the Ear Headphones Headset
$119.99$59.99 Add to cartWireless Bluetooth 5.3 ANC Noise Cancellation Hi-Res Over the Ear Headphones Headset
Categories: Electronics, Gaming, Gaming Headsets Tags: 5.3 ANC Noise Cancellation Hi-Res Over the Ear Headphones Headset, ANC, Audio, Bluetooth, Cancellation, Ear, Earphone, gaming headset, Headphones, Headset, Hi-Res Over the Ear Headphones Headset, HiRes, Noise, Wireless, Wireless Bluetooth 5.3 ANC Noise Cancellation Hi-Res Headphones, Wireless Bluetooth 5.3 ANC Noise Cancellation Hi-Res Over the Ear Headphones Headset, Wireless Bluetooth 5.3 ANC Noise Cancellation Hi-Res Over the Ear Headphones Headsets$119.99$59.99 -
Sale!
Wired Sports Gaming Headset Earbuds with Microphone
$19.99$9.99 Select optionsWired Sports Gaming Headset Earbuds with Microphone
SKU: N/A Categories: Gaming, Gaming Headsets Tags: Accessories, Earbud, Earphone, Earphones, Gaming, gaming headset with microphone, Headphones, Headset, IOS, Microphone, Sports, Wired, Wired Sports Gaming Headset Earbuds, Wired Sports Gaming Headset Earbuds with Microphone, Wired Sports Headset Earbuds$19.99$9.99 -
Sale!
150W Universal Multi USB Fast Charger 16 Port MAX Charging Station
$49.99$29.99 Add to cart150W Universal Multi USB Fast Charger 16 Port MAX Charging Station
Categories: Charging Stations, Electronics Tags: 150W, 150W Charging Station, 150W Universal Multi USB Charging Station, 150W Universal Multi USB Fast Charger 16 Port MAX Charging Station, 150W Universal Multi USB Fast Charger 16 Port MAX Charging Stations, 150W Universal Multi USB MAX Charging Station, 16 Port MAX Charging Station, 3.5A, Charger, Charging, Fast, laptop charging stations, Max, Multi, Port, Stand, Station, Universal, USB$49.99$29.99
🎯 Key Takeaways for quick navigation:
00:06 🎬 Introduction to the Responsible AI Topic
– Discussion about the scariest movies.
– Defining what "making AI real" means.
09:17 🔍 Understanding Responsible AI Principles
– Explanation of the six principles of responsible AI, including fairness, reliability, safety, privacy, security, inclusiveness, and transparency.
19:13 🛠️ Introducing the Responsible AI Dashboard
– Overview of the Responsible AI Dashboard.
– How it combines various AI debugging tools.
24:08 📊 Visualizing Model Errors and Data Representation
– Illustration of model accuracy issues.
– Explanation of data representation imbalances and their impact on model training.
26:27 📊 Understanding Data Bias in AI Models:
– The dashboard allows you to analyze data behavior and demographic distribution, helping identify model biases.
27:22 🔍 Making AI Models Transparent:
– The dashboard transforms AI models from black boxes to glass boxes, making predictions more understandable.
– It highlights the top features influencing predictions, aiding in model interpretation.
30:35 🖥️ Training and Deploying a Model:
– The process of training a classification model and deploying it to Azure Machine Learning Studio.
– The importance of normalizing data and using Azure Machine Learning SDK for Python.
42:23 🚀 Registering and Using the Trained Model:
– Registering the trained model in Azure Machine Learning Studio.
– Using Azure Machine Learning SDK to authenticate, retrieve, and utilize the registered model.
46:57 📈 Building and Configuring the Responsible AI Dashboard:
– Configuring the components of the Responsible AI Dashboard, including Constructor, Explainability, Error Analysis, and Gathering Insights.
– Creating a pipeline to build the dashboard and monitoring its progress.
Made with HARPA AI