Video Presentation: Los Angeles CTO Mixer

Video Presentation on Machine Learning and Artificial Intelligence.

Transcript

0:00
I’m Barbara Bickham, a little bit about me. My fun project is a Hyperloop. I do work with Hyperloop. I do all their strategic software. I also work on their AR VR experience.

0:13
So I do know about that.

0:17
I’m still pretty strong in Java C and C++. Although I do program in multiple languages. Everybody asked me What language do I know? It’s for what language do I do not know at this point? My largest kind of fundraise also help companies raise money. 13. Now, I learned budget management about five minutes ago. Has any song? Has anyone saw seen the movie her? Anyone See that? Okay, I’m trying to figure out where we’re at. anymore seeing the movie Ex Machina? Honestly, I like. Okay, but also my TV shows. Has anyone seen the TV show Person of Interest?

1:01
No, yes. That was a great show. So why did I mention all these shows? Because this is relevant to the topic. They all had something to do with kind of robots and AI and kind of machine learning and all the machines taking over kind of the world. So those are all good children. All right. So how did I get introduced to AI? When I was in Berkeley, I got introduced AI, there was a class called cs 188. And I learned lists. And so that was a program language that kind of did structural AI. And then I also took a semantics class. So I combine those two and kind of wrote some NLP, like for very basic early AI for your class project. So that was my very first introduction into artificial intelligence. Again, so I took a break. For a while I was that company flex 5.5 was a big data, and still is a big data analytics company. So we did big data analytics for Fortune 500, companies, banks, and all these things. And so we took all their data and put it and made it into these nice apps that they could see. And these dashboards, everybody has those Splunk was one of our competitors, and tableau. But we have something very special that we did, we created a language called fel flexlite engine language, why we didn’t use MATLAB or R, I don’t know.

2:38
But we could do predictive modeling based on our little language. And we could do it across time, because we also have what they call time varying data. So we take the data, we put it in there, and we could crunch numbers and predict kind of what’s gonna happen in the future. And we could also do regression analysis on what happened in the past. So that’s my predictive model. Alright, so then can’t get rid of it. It’s back 2015. So I worked for a company that did emergency response for mascara.

3:13
We use what this was, I’m a REST API expert forgot to say that. So we use a lot of nerd API, we use their natural language processing and their sentiment analysis. It was written in Node. And I like happy. And we were using Twilio in a kind of a our own mobile app in order to get the responses. So this is hard to read. But this is this is happening. So this is the whole entire server in 32 lines.

3:51
At the very top here, if you see dialogue equals Watson Gladwell, which is in a very tough definitely. But it’s it’s very at the very top, that’s that’s the call off of Watson, you connect to the server, you make a server and basically that’s it.

4:08
That’s the whole server 32 lines. Hapi, it’s very good. All right. This is more more stuff. This is the output. So basically, this is the setup totally on. This is the setup where so what we’re doing was we were getting the inputs in from the people we were putting into the NLP get the sentiment, like I’m having a heart attack versus I need a beer, I need a beer really mercy. So we will recording that. And then yeah, we need to immerse people in my be but for us it was and then we will we will go through here and this dialog pieces where it will go into the Watson and say, Hey, is this an intent that we intend to have and if it was, we’d say We said, the word is so better with store, the key value pair like these are a head this user, this is just a test Hello.

5:13
And then let’s see, if it was fine, then it would send an SMS. And then Watson is just like, I don’t know what to use messenger for. I don’t know if the microphone is. But once it is triggered, so I had to use this post nice faces in order to make sure that I can get the response back, or else it was good to refer to the note didn’t like it.

5:46
Same things Facebook does all those cross. Alright, so that was that was the extent of that code. So I mean, in about, I’d say that was about maybe 125 lines, you had a full server with all the response back and forth to chat, and some back and forth. This is how you programmed Watson. So all kind of API’s need to be trained. So we need this data. This is all the data. And you could label it certain things. And we expected things to come back. So at the top, it says, you know, sorry, I understand I didn’t quite get there, grasp what you mean on my phone. So those were some of the contents that come back, based upon what we put in here. And then if you did pass one of these questions that would come back to the I don’t know. So that was what this was more.

6:55
So Watson was Watson, it was very interesting to work with is slow for emergency response. So don’t use it for that calm. And otherwise, it was all kinds of decisions.

7:09
So this was a next project I did. I wrote the decision engine. So the project for that was we were voting in groups. So we would ask, like, Do you want to come to the CTO mixer meetup? And so I’d said that the five of my friends, and they would all vote yes or no, that was one way to vote. Or we could say, Do you want a? What are they like those actions versus something else? And then you, your friends could vote on that. So it is it is always also descent. And that’s written in Python? Oops. So this is the Python for that, for that one point. So what I did here, this is all my code. What I did here was there’s this shows the algorithm that creates a decision tree that’s waiting. And why would we wait the decision tree, because eventually, one of our friends may have more influence over someone that another. And so initially, everyone had equal weight. So we used to attack. But this shows the algorithm was smart enough to say, if we had somebody that was like, influential in the questioning, then their weight would be higher than someone that was listening to this room was very special, for two reasons.

One, it created the tree itself. So create this new tree, go to a bowtie, and we were voting in groups kind of polling, that we want to make sure that, you know, final decision could be made. And then also, it was very hard to kind of, you can kind of play around with voting, it was very difficult to get free rides from that. So that’s all the code for that was more coding. So why did I use the single transferable vote, you could do proportional voting, which will equal weighted and avoids the free reign. So you know, if you vote for a president, or anybody May, or whoever, you can kind of say, Well, I’m going to try and vote for the opposite person, and therefore my vote will have more weight, but that doesn’t work. And then or if you don’t want to vote for anybody, that doesn’t work for you. And then that algorithm could do groups of votes. And then the decision tree is in this algorithm design manual, which is pretty good for all different kinds of algorithms.

And then came Alexa. So I asked within node and I use lambda for that. So that was my first introduction into serverless architectures. So this is the entire Alexa skill, which is what they call it, They have some free the five and 10 steel. They have a chance to just like most of these other guys, this was branded with guitar fats. So there’s a whole list of the best tower coding, export it out. And if you wanted to get the facts, you get effect, basically you pull the thread and it says, and then you can say, Alexa, give me a guitar fact, tell you here’s your photographer.

10:35
So back to the extent of the Alexa fairly simple.

10:42
So if you want to do Alexa, there’s more. There’s Alexa developers on YouTube. That’s how I went to like one intensive workshop and then learn to rest on video. So what else am I doing here? napalm entertainment is a is a game company. They do video games.

11:05
They do massive multiplayer MMOGs. We’re branching out and we’re going to put in a minute game like when the investigation not that sophisticated, but semi sophisticated. onto the Facebook Messenger and Alexa. So that’s coming out in two weeks. p3 is in June, June 13 1415. Last month. So Facebook is special or not so special. Because you can, you know, you can do the chatbot with it. They use witty AI, I use the API that’s Amazon, not Amazon. Google’s Google’s version of the NLP that’s pretty beautiful. And I looked into the Lex and Lex is still which is a waxman’s version of kind of Watson or Facebook’s NLP or something. It’s still pretty primitive. So it’s requires a lot of training all these require along with the Lexus and Lexus the engine behind you know the, the text part of Alexa.

12:18
So that’s it pretty fast.

Transcribed by https://otter.ai

You've successfully subscribed to Trailyn Ventures | Unbound
Great! Next, complete checkout to get full access to all premium content.
Error! Could not sign up. invalid link.
Welcome back! You've successfully signed in.
Error! Could not sign in. Please try again.
Success! Your account is fully activated, you now have access to all content.
Error! Stripe checkout failed.
Success! Your billing info is updated.
Error! Billing info update failed.