Getting into data science and AI

Bandwidth for changelog is provided by Fastly. Learn more at fastly.com we move fast and fix things here at Changelog because of Rollbar. Check them out@rollbar.com and we're hosted on Linode servers. Head to linode.com changelog this episode is brought to you by DigitalOcean. They now have CPU optimized droplets with dedicated hyper threads from Best in class Intel CPUs for all your machine learning and batch processing needs. You can easily spin up their one click machine learning and AI application. This gives you immediate access to Python 3R, Jupyter Notebook, TensorFlow, SciKit and PyTorch. Use our special link to get a 100 credit for DigitalOcean and try today for free at the.co changelog once again.co changelog. Welcome to Practical AI, a weekly podcast about making artificial intelligence practical, productive and accessible to everyone. This is where conversations around AI, machine learning and data science happen. Join the community and slack with us around various topics of the show@changelog.com community. Follow us on Twitter. We're at Practical AIFM and now onto the show.

Welcome to Practical AI. How you doing Chris?

I'm doing very well. How's it going today? Daniel?

It's going really great. As you know, in my background, I started out in academia and then I moved into industry to become a data scientist and I'm really excited to have Himani Agrawal here with us who has also made a similar transition and we're gonna kind of talk about some of that today and also what she's working on currently with AT&T. So welcome.

Himani hi Chris and Daniel. Thank you very much for having me. I'm excited to be here.

Awesome. Yeah. Well, could you just give us a brief sketch of your background and what you're doing?

Sure. So my name is Himani Agrawal. I'm a machine learning engineer at AT&T where I work on predicting the network outages and avoiding customer dispatch. My background is very interdisciplinary. I pursued a bachelor's in the area of civil engineering where I was using optimization algorithm to find optimal slope of a dam. During my PhD I was working on solving interdisciplinary problems in the area of HIV infection, applied mathematics and computing.

Awesome. Yeah, that's a really great background and I'd love to dive into the individual pieces of that. So you mentioned kind of starting out more on the engineering side. Specifically in civil engineering was kind of data analysis and machine learning and data science always something that you were interested in pursuing or how did that develop during your Schooling, Yeah.

So I believe I was meant to be a machine learning engineer. And I see machine learning is an optimization problem. My tryst with optimization began in my undergrad school, which is located in the flat regions in the foothills of the Himalayas. A lot of dams have been built in the upstream of the city. And determining the optimal slope of these dams is a very, very important optimization problem that the engineers have to solve. I got really interested in that problem as a civil engineer and I started working on developing a model to determine the optimal slope based on features like the type of soil, the water retention of the soil, precipitation conditions, vegetation, and so on. And to solve these problems, I used genetic algorithms, which is based on Darwin's theory of evolution of human beings. And what interesting thing happened while I was solving that problem? That not just computations and applied maths and optimization, I also fell in love with computational biology. So I decided to pursue a PhD to solve biological problems in the area of computational biology. And during my PhD I was working in a very interdisciplinary area encompassing fields like mechanics, biophysics, applied mathematics. I collaborated with applied mathematics lab at Rutgers University and material science lab at Rice University to solve a very interesting problem of how can soft. How can rigid HIV proteins can make a cell membrane softer?

So that sounds pretty amazing. It sounds like you've really known you wanted to be in this field for a long time. And so I guess, you know, I was, I was going to ask you if you were kind of thought about getting into data science as you were doing your PhD, but it sounds like from your undergrad you already knew you wanted to do that. How young were you when you decided that this was the right path for you and that this is the way you wanted to.

I have been really into mathematics and computing since I was a teenager. I decided to pursue engineering because I thought that involved a lot of mathematics and computing.

Yeah, as you were kind of like having those passions for mathematics. When did you start first hearing about data science and AI?

During my undergrad, I got a chance to attend the Grace Hopper celebration of Women in Computing in India and I presented a poster at the conference. I was also a Grace Hopper scholar for the Indian version of the conference. And that's where I got in touch with the technology industry. So when I attended the conference, I came to know how my engineering and computing background can be directly applied to solve a variety of problems in the technology industry. And one thing I was always interested is that I have a very varied interest, a lot of diverse interests, and I wanted to solve a variety of problems I never wanted to solve or get stuck in solving just one problem. So when I saw in the area of technology, people work on multiple use cases and acquired the domain knowledge as they go forward with the project, I was really excited about the nature of solving problems in the technology industry.

Awesome. Yeah. So for those that aren't familiar, what is the Grace Hopper Celebration?

Grace Hopper Celebration of Women in Computing is the world's largest gathering of women technologists. Last year, 22,000 women attended this conference in Houston. It sells out within minutes. It's an amazing conference. I've been attending the conference for the last five years. I have been Grace Hopper Scholar twice when in 2013 for the Indian conference and in 2014 when I attended the Phoenix Conference. And it has a wonderful community of women technologists and I really love so.

What does it mean to be a Grace Hopper Scholar? Could you kind of share what you've done it twice now, but for those of us who are not familiar with the details of it, could you tell us a little bit about it?

As a Grace Hopper Scholar, the Anita Borg Institute sponsors you to attend the conference. Many of the scholars are also either speaking at the conference or presenting a poster at the conference. There are very exclusive Grace Hopper Scholar networking events that happen at the conference where we get to meet with mentors and industry sponsors. We have roundtable discussions. So apart from the General Conference, Grace Hopper Scholars have access to have a unique networking event at the conference. Apart from that, we have a Grace Hopper Scholar Facebook group where we keep in touch with all the Grace Hopper Scholars. And yeah, it's a wonderful community. We help each other and it's a great networking platform.

Awesome. Yeah. And I mean, it sounds like it had a really huge impact on your life in terms of you knowing about the engineering field and knowing that you really enjoyed mathematics, but then seeing how those skills could be applied in so many different areas. It sounds like it was really a great inspiration for you. So that's awesome. We'll definitely put a link in the show notes so people can find out more. I'd be interested to hear a little bit coming from a PhD myself. I know it was a really kind of, I guess I should say weird experience going from a PhD into industry. And I know I had to learn a bunch of different jargon and find my own way through that. I was wondering if you could speak to that yourself. Did you find that to be a similarly weird transition or how did you go about going from academia to industry?

My first experience with industry was when I Got an internship with Microsoft Research. That's the first time in 2015 I interned with Microsoft Research. There was a group in Microsoft Research which works in the area of computational immunology and I joined them as a HIV domain expert. That was the problem I was solving during my PhD. I was working with a machine learning group at Microsoft Research, although my background was not in the area of machine learning and data science in the traditional sense. So it was great to contribute my domain knowledge to the projects at Microsoft Research. But I was also very inspired to see the kind of problems that the researchers there were solving, not just in the area of immunology, but in a variety of domains. And again, I said the very fact that technology, data science and machine learning helps me solve a variety of problems, that's what excites me. So apart from Grace Hopper conference, that was a great platform where I got a firsthand experience seeing how machine learning can be applied to a variety of projects.

Oh, that's pretty cool. So you also did an immersive data science program with Galvanize, right? You know, why did you do it and would you do that again and you know, could you just tell us a little bit about what that's like?

Yeah, sure. Galvanize Data Science Immersive program was a very transformational experience for me. When I finished my PhD, I knew that I wanted to join the technology industry. I had all the technical skills, I had great computational skills, but I just didn't know how to really be a data scientist in the technology industry. And Galvanize made that happen. Galvanize Data Science immersive is a 12 week, 500 hour program that teaches industry standard data science tools and knowledge and statistical analysis, machine learning, algorithms and data engineering. So it's during that program it equipped me with the right skills which made me transition into the tech industry.

That sounds cool.

Did it?

So I'm kind of curious when you got to the end of that and you know, thinking about the fact that you are now an AI engineer at AT&T, did that Galvanize training help you bridge that to the real world so that you could enter that job and be productive?

Yes. So during the Galvanize program we went through several projects, applied projects, and I also did an internship with machine learning for genomic medicine based startup Sympatica Medicine where I worked on machine learning for early diagnosis of Parkinson's disease from RNA sequencing data. So the galvanized program provided a great platform for me to not just learn the the tech skills, but also apply it to A real world problem with a company. So that experience was very valuable. And then on the side I also joined Jeremy Howard's fast AI deep learning class at University of San Francisco which was a great program. Jeremy Howard is a great teacher. I really enjoyed that class and I built my deep learning skills during that program. I was also working at Augment Solutions which is machine learning for customer experience based company. Does that sound right?

Yeah.

Where I was working on churn prediction from Frontier customer chats data. So this whole experience was possible only because of the galvanized program. And then I transitioned into a full time role at AT&T.

Yeah, it's really great for me to hear that from you because sometimes I think I myself I feel pretty. I'm self conscious in the industry because I also came from a non kind of CS and machine learning focused background. I'm always learning new jargon that I wasn't exposed to before. I remember coming from academia. It's like when I realized that when people were talking about these regressions it was really just ordinary least squareds. I remember having these light bulb moments when I understood, hey, I did this back in science, but now it applies here. I was wondering if you've also kind of felt that way coming through this transition and what advice you have for students coming from engineering or science backgrounds and wanting to transition into data science and AI.

I definitely felt that way. I realized that during my PhD I was working on solving optimization problems which are very similar to the problems in the data science industry. But I was not using the same jargons that are being used in the tech industry. So by being part of galvanized program, I got to learn data science in the tech way. So that really helped me a lot and it helps me even today in my job at, AT&T. Coming to your second question, how can people from different backgrounds can enter into data science? I believe that machine learning and data science is very ubiquitous right now. There is a huge scarcity of machine learning and data science expertise. So it's great if people from different backgrounds can enter into that field because that would really spark creativity.

I think it's great that you're kind of exposing some of these things. I think some of people from engineering and science feel like, oh, you should just kind of know all this jargon and all of that stuff. But most people are trying to pick up either the computer science and programming pieces or the science pieces or the optimization pieces. I don't know if you know this Chris, but went through a similar transition and utilized the thinkful course, which I think is now kind of like an online boot camp to get up on some of this jargon.

So I had no idea.

Yeah, I never knew that I would similarly recommend. I think Hamani hit the nail on the head that, you know, it's a great time to get into this field and don't let that kind of lack of jargon scare you away. But there's a lot of resources out there that you can use.

So you two have made me want to go and get in a boot camp right now after hearing both of you talk about it.

And what sort of events and opportunities are there with women who code women in ML, these different organizations, what sort of things are available?

So there are a lot of machine learning communities out there. For example, Women in Machine Learning and data science organizers. There are a lot of women in machine learning and data science groups all over the world and they organize technical talks at companies where we get to meet with machine learning technologists from varied backgrounds. I remember I attended a talk at the Chan Zuckerberg initiative and that was great. Apart from that, I know vimlts in San Francisco. They organize a code and coffee session every Friday. I think it's a great chance for aspiring data scientists to be involved in that community so that they can receive mentorship from experienced women. I've also been a part of Women in Machine Learning. It's a community that organizes one day workshop and luncheons at conferences like NIPS and icml. I attended Women in Machine Learning workshop last year at nips. It was a great experience for me.

Yeah, that's. That's awesome. It's kind of one of my goals to go to NIPS sometime. I haven't been yet. It's another one of those that sells out like a Taylor Swift concert. But I know you were mentioning before the show as well. Himani, aren't you involved with mlconf? Maybe you could share a little bit about that. I know you've shared a lot of great community things, but I think you're pretty active also in the greater kind of machine learning community. So tell us a little bit about ML Confident and what you're doing with that.

I'm excited to tell you about the upcoming ML conference in San Francisco that's happening on November 14th. It's a single day, single track conference, has amazing programming in the area of applied AI from top industry AI experts. If anyone wants to go, I have a 20% discount for them. They can use Imani 20 and get 20% off the conference. Registration.

Awesome. Yeah. And we'll put that in the show notes as well. It sounds sounds like a great opportunity. I appreciate your efforts in helping to organize that.

All right, as we come back out of our break, Himani, I'm going to turn us toward telecom. Since you are an AI engineer at AT&T, I wanted to start with kind of a general question and just ask what are some of the main uses of machine learning and artificial intelligence in telecom?

I would like to talk about telecom industry from AT&T perspective. AT&T for me is first and foremost a modern media company which is empowered by telecommunication engineering, television engineering and advertising analytics. Along with our subsidiaries hbo, cnn, Turner, Warner Brothers Entertainment, Xander Aden, Vault and Magic Leap. Truly sky is the limit with what machine learning problems can be solved at AT&T within ATT, I work with the organization called Chief Data Office. It's a really wonderful organization and very young too where I work with AT&T business units to develop automation solutions. We collaborate extensively with AT&T Labs and use the research innovations that spin out of the labs in our applied AI projects. In the Chief Data Office. As a machine learning engineer, I work on data analysis and pattern recognition of telecommunication devices and streaming alarms data to predict network outage and avoid customer dispatches. Furthermore, our devices have been impacted due to the recent hurricanes, so I'm working on utilizing the weather flood and power data in conjunction with the streaming alarms data to predict the optimal dispatch time to restore the devices. Apart from that, AT&T has been a pioneer in the area of 5G and I believe 5G when combined with Magic Leap in conjunction with machine learning is truly game changing for personalized customer engagement for TV streaming.

Well, I am just super impressed to hear all. I mean, coming from optimizing dams to working in computational biology to helping with disaster related recovery within telecom, all of this is definitely super inspiring to me. It sounds like your team, this Chief Data Office, it's kind of like you said, it's positioned between research and the rest of the company. So do I have it right that you're kind of more on the applied side? That you take kind of some of the things that are coming out of research and kind of try to figure out how to apply them within the rest of the company? Is that right?

Yeah, that's quite right. AD&D Chief Data Office is on the production side of things. We want to deploy machine learning solutions at scale.

Awesome. Coming kind of from more the academic research side. Have you been surprised at all by some of the challenges that are Involved in kind of taking and applying those research things to a larger scale within a company context.

I've seen a lot of commonalities between my research experience and my experience right now as a data scientist. For example, during my research, when we come up with a research problem, it's so hazy. We go out there, read all the research papers and try to figure out the problems that we have to solve. Similarly, at AT&T, when we work with the business units, we get tons of data and we have to figure out what kind of problems that can be solved. That's very similar to what we do as researchers. First figure out the problems that we can solve and execute them.

So I have a follow up question you had just mentioned. When you work with, with business, while you're doing that, I was wondering if you could actually share with us some of the things that you are working on. Kind of tell us what you're doing on a day to day basis.

Yeah, sure. So as I mentioned, I'm working on the data analysis on the streaming alarms, data that come out of the telecommunications devices. And from that data I try to figure out the root causes of the network outage. And by doing that, I'm preventing a dispatch to a customer's house.

That is super cool.

Yeah, it is really cool. So the data that you're processing is actually kind of streaming off of all of the AT&T related devices and then you're kind of detecting network outages, is that right?

Yeah, I have a topology, the network topology with me. And from that, from all the alarms that are being generated from the network, I try to find out where exactly in the network the problem arises so the dispatch can be sent to that particular location.

That is cool. So I didn't mention at the beginning of the show, but about 20 years ago I worked for AT&T in network engineering and honestly we did not have any machine learning, at least that I was aware of at the time. So I can tell you, going back and thinking of the pain of putting together networks, I wish we had had someone like you there then to help us get through these kind of difficult things. So I was just, I couldn't help but think about that as you were describing your daily duties.

I didn't know you worked at AT&T.

Yeah, and actually I don't know that I've mentioned it on this show before, but my first job as a data scientist was with a VoIP startup, VoiceOver IP. And I know that we had in our mind kind of envisioned some of the things that you're talking about. So it's just really cool to see people actually kind of putting this into practice. I know it's not an easy problem to solve at all.

Is that when you got into AI?

It was. Yeah, I was in my PhD. I came out into industry. I was actually working at an IP firm and that's when I was doing the thinkful program. And then right after that, I started my very first data science job with this VoIP company called Telnyx, which is still around. Actually is kind of voiceover IP and doing number porting and all of that stuff via API. Yeah. So I kind of wanted to dig into one more piece. You mentioned kind of magic leap and machine learning and other things. I was wondering, as you look forward into the future of kind of the media and telecom industry, what are you excited about in the future that you could see maybe AI enabling?

Yeah. So I'm really passionate about music. I'm an opera singer.

Oh, wow. That's awesome.

Are you going to sing for us now?

No, not now.

Oh, okay. I had to ask. Okay, I had to ask.

Well, if you have any. If you have any videos or links to you singing, we would love to include them in the show notes. It's awesome.

I do have my performances and actually I also currently train with a University of North Texas voice professor and it Italian singing as a mezzo soprano.

So that is super cool. I actually there's another connection we have. I'm not currently a singer, but when I was young, grew up in the Atlanta Boy Choir and it's not something I was ever expecting to say on this podcast, but there's a little connection.

Yeah, I'd love to hear more.

I'm continually amazed with all of the things that you're able to do. You do the opera singing and then you were kind of mentioning, you know, I think that was leading to something with the question of what kind of is inspiring to you in media with AI?

Yeah. So the reason I'm so much interested in the media company is because I'm passionate about music and I'm very interested in exploring research in the area of reinforcement learning and score following. So that's my side research passion.

That's awesome. Have you seen the stuff coming out of. I think it's Project Magenta.

Yeah, that's a very cool project. I was particularly impressed with this project called AI to it, in which the random notes played by the user on their interface are turned into a beautiful melody. That was a very cool project that came out of Magenta.

Oh, that's amazing. Yeah. So do you ever see yourself in the future kind of devoting more of your time to music and AI?

I'm definitely very interested in it. I am very much interested to explore the realm of how reinforcement learning can be applied to music. Very excited about that area.

So when you delve into that, you'll have to come back on the show and share with us a reinforcement learning within music topic with us so that we can learn a little bit about that. That is one topic. We are hitting reinforcement learning on the show, but we've never combined that with music.

That would be great. I will keep you updated.

Sounds great. I had another kind of question about your current work and you mentioned that you're an AI engineer. And I know that there's a lot of different kind of labels that people are putting on those in industry doing AI and machine learning. Sometimes they're called data scientists or, or analysts or machine learning engineers or AI engineers. I was wondering from your perspective, is there a difference in those things? Do you work with data scientists as well, or how does engineering fit into AI?

Yeah, that's a great question. So my formal position is that of a data scientist, but I do apply machine learning. I do use deep learning, and I solve challenging problems in this area, which is the concept of AI. AI is a concept and a dream that the machines can see, machines can hear, and machines can be creative. So I apply all of that in my role. So that's how I see myself as an AI engineer. I believe that although there are a lot of titles that have come up recently, they are one and the same in my opinion, because as my job in this whole realm, I have to use all the tools and technologies in multiple domains. So even though different terms have come up, a data scientist or a machine learning engineer has to be well equipped with multiple skills.

That sounds really cool. I have a question because as you're kind of talking about how those are going together when you're working with coworkers that are not in data science and not in AI, I'm curious, what is their perception of you now bringing AI? Because I mean, with this being relatively new to the industry and certainly in kind of a production role, as you talk to these co workers, how do they perceive you and the job you're doing?

So my coworkers, who are not data scientists, who are not machine learning engineers, believe that machine learning is magic because all the work that was being done by human beings are now being done by machines. Machines are getting more intelligent, and I feel it appears like magic to a lot of people. But I really believe at the very end machine learning is mathematics, deriving patterns from data and it's only as good as the data that we have.

Yeah, that's a great point. I think that's a really great way to end up. Our conversation here is really with that emphasis on the applied side and really emphasizing that AI is a set of methods that we apply in a predictable way. I was wondering if you want to share any about where people can find you online and maybe either on the AI side or the opera side or wherever. And we'll kind of end up after that.

Yeah. So I'm on LinkedIn, Twitter. I have two YouTube videos of my performance. I would be happy to share the links.

Awesome. Great. Well, thank you so much for joining us, Himani. It's been fascinating to hear about your journey and what you're working on. I know me for one. I'm super impressed and so thank you so much for being on the show.

Thank you very much.

Thank you so much for having me.

All right, thank you for tuning into this episode of Practically High. If you enjoyed this show, do us a favor. Go on itunes, give us a rating. Go in your podcast app and favorite it. If you are on Twitter or social network, share a link with a friend. Whatever you gotta do. Share the show with a friend if you enjoyed it and bandwidth for Changelog is provided by Fastly. Learn more@fastly.com and we catch our airs before our users do here at Changelog because of Rollbar. Check them out@robot.com changelog and we're hosted on Linode Cloud Servers. Head to linode.com changelog check them out. Support this show this episode is hosted by Daniel White Knack and Chris Benson. Editing is done by Tim Smith, the music is by Breakmaster Cylinder and you can find more shows just like this@changelow.com when you go there, pop in your email address. Get our weekly email keeping you up to date with the news and podcasts for developers in your inbox every single week. Thanks for tuning in. We'll see you next week.

I'm Tim Smith and my show Away from Keyboard explores the human side of creative work.

You'll hear stories, sometimes deeply personal, about the triumphs and struggles of doing what you love.

I need to give myself permission to not overdo it if I know that the weather forecast is really good tomorrow and I don't have to do a podcast tomorrow and I could go to the beach. Maybe I go to the beach maybe I do something that is not working work.

New episodes premiere every other Wednesday. Find the show@changelog.com afk or wherever you listen to podcasts.