What if neural networks could tell stories too?
We tell stories all the time. You probably just finished telling your co-worker about your amazing journey inside a volcano. It was epic.
What if neural networks could tell stories too? That would be wild. Well, look at this neural walk and talk video. It’s happening already.
It’s things like this that inspire Anton, a brilliant data scientist here at Beeline Russia. In this interview, we meet him, and learn more about what he does — for work, fun, and inspiration.
You are a data scientist by day, but a master of Kaggle competitions by night. Can you tell us about some recent competitions?
Absolutely. I spent a lot of time on a competition for Homesite Quote Conversion. It was a classification task with a binary outcome. The metric for the competition was the area under the ROC curve (receiver operating characteristic). During this competition, I made a lot of progress with classification ensembles, as I found it was the best way to get an effective quote. It was a process, building different models, then building new models on top of the previous models by blending and stacking. I finished in 19th place out of 2000.
Are there certain methods you use for each competition?
Every competition is really different. Some are focused on the customer experience, while others may be about conversion rates and data security. So there is no recipe for success. I like the challenge of trying lots of methods and creating from scratch. I like to do Kaggle competitions at night, usually after my baby and wife go to bed!
What do you learn from a business perspective about these competitions?
These Kaggle competitions are all about engineering. It’s all about working with data. The same goes for my job: when building these data models, a lot of the time is spent collecting data and preparing it. Then you start with analysis and find insights. From there, you can build effective models. In the end, the goal is to solve a problem — which is what the business is trying to do.
Is your work here similar?
In Kaggle competitions, they give you a dataset to work with, which is different from here at VEON. Here you get mountains of data, and you have to decide what you need to solve the problem.
What else do you do for inspiration?
Well, one of my favorite books is Building Machine Learning Systems with Python.
I’m constantly looking in the Python data analysis library and seeing what I can do with pandas in terms of exploring and preparing data. I’m currently really into recurrent neural networks and their potential as storytellers. My field has a lot of inspiring things going on, and I want to have an impact. I also enjoy music, like Sum 41 and Offspring (can we play that after the interview?).
Can you tell more about neural networks and storytelling?
Yes, it’s very cool what’s happening with this technology. It’s basically a model that creates descriptions of images. These architectures are being trained every day. The neural networks can generate sentence descriptions that are pretty great already. For example, if it was capturing us right now, it would say, “two people are talking in an office.” These models are truly getting better at connecting image content with natural language. There is still much training to do, but it’s all very exciting.
If you were trying to convince someone to study data science, what would you say?
Firstly, I would emphasize that data science is a really cool field, and it’s only getting more popular. Data is increasing, and the things we can do with it are increasing too. Finding good and exciting work is very possible. Data science teaches you how to apply different mathematical algorithms, and what it can accomplish is amazing.
If you were trying to convince someone to join your team, what would you say?
We are a growing team. As we have over 200 million customers globally we have a lot of data to analyze. We are organizing this data, discovering characteristics about our customers, and building solutions. We need the best talents. You can really improve yourself by joining our team.
The story of Anton tells us how great things can be achieved when we are scrappy and resourceful. It also tells us much about the potential of data science to transform the world. Anton has a PHD in Theoretical Physics, which gave him lots of options. But it’s innovations like neural network storytelling that drove him to data science. And it’s the hope to pioneer more great innovations that keeps him going.
