Spotlight

Space Talent Spotlight: Appu Shaji

The Space Capital Podcast |

December 7, 2020

"AI is in a state where you have enough starting resources to get your hands dirty and build things that work without too much of an initial investment."

Spotlight

Space Talent Spotlight: Appu Shaji

|

December 7, 2020

GO TO WEBSITE

"AI is in a state where you have enough starting resources to get your hands dirty and build things that work without too much of an initial investment."

Spotlight

Space Talent Spotlight: Appu Shaji

"AI is in a state where you have enough starting resources to get your hands dirty and build things that work without too much of an initial investment."

One of the rare common factors of my life as a scientist and life as an entrepreneur is that in both disciplines, I encountered quite a few scenarios with uncertain outcomes.

Appu Shaji, CEO of Mobius Labs

A Space Talent Spotlight Series Interview with Appu Shaji, CEO of Mobius Labs, former founder of Sight.io and Croppola

What is your background?

My first serious foray into computer vision was during my bachelor thesis where I made a small electronic braille, but with a twist that it could display images. Computer vision part came in transforming a normal image to a line/edge drawing that can be realized in the braille and is a good summary of the original image. Though seemingly straightforward, this turned out to be really tricky, and made me realize that the whole field of computer vision was under-solved and hard.

This kindled the interest in me to learn further and I pursued my PhD at Computer Science Dept. at IIT Bombay. My thesis was on Computer Vision, specifically, 3D reconstruction of shapes from single view images. It was a very math oriented thesis, and I was trying to solve 3D reconstruction problems by using ideas from differential calculus and optimization methods.

In a field like computer vision, most problems are under constrained (i.e. there are no pure mathematical instruments to pick the right solution from thousands or more of other plausible, but incorrect solutions). However, most object motions (for example a person walking) and camera movements have geometric constraints in it, and the hope was that by modeling these implicitly using mathematical structures from differential geometry, we can find the right solution. This search remained elusive and was deeply frustrating for most parts. But it was great when I got it to work, albeit only in a subset of types of scenes and motions.

Towards the end of my PhD, I had a mental shift that computer vision is very much an empirical domain rather than a mathematical domain. So I also took a 180 degree turn from theoretical viewpoint to a deeply empirical and pragmatic viewpoint, and started applying ML to some of the problems I was trying to solve. Back in the 2000s, when data and compute was really scarce compared to current times, machine learning was mainly treated as a theoretical domain, and limited to toy datasets.

This learning influences a lot of our current work at Mobius Labs today. For example, nowadays I argue for more loose and simple formulations, and use messy datasets to train algorithms on. Our observation is that when trained with enough data such methods learn to generalize better, rather than one with complex formulations or with a dependency on clean data.


What have been your top three career accomplishments so far?

Grad school was very important. It was a five year journey with multiple ups and downs that tested me on multiple occasions. I discovered a technique, which is probably one of the first works that teaches machines to associate an aesthetic score to image. My dad is a filmmaker in India. While growing up, I overheard many conversations he was having with his friends and others in visual aesthetics and its importance to storytelling and filmmaking. Since, I was more a geek first, rather than a visual storyteller, there always has been a sense of mystery and deep reverence towards people who can spot and identify aesthetics.

I co-founded Mobius Labs where we make it easy to add superhuman computer vision to any application, device or process. The opportunity to start a company from the ground up, hire some of the most talented people on the planet, find partners like Space Capital to support us and create an organization that is motivated with a sense of purpose is something that gives me tremendous joy and satisfaction.



What were the critical steps/choices that helped you get ahead?

Having a long term perspective always helped me. One of the rare common factors of my life as a scientist and life as an entrepreneur is that in both disciplines, I encountered quite a few scenarios with uncertain outcomes ( for eg: an experiment does not work, or a business strategy is not working). A mental model that helped me in these situations is to treat them as learning phases and to keep exploring to find a path towards your goal. It is often a marathon, and not a sprint, and some miles might be hard, but having a goal helps you to keep moving.    


What part of your education had the most impact on your career?

I always had a passion towards computer vision even from my high school days, and grad school helped me explore and refine my thoughts about the topic. 


What about your career have you enjoyed the most and least?  

For some reasons, I had the most fun and most misery within the same projects. For example, there are days while I am building Mobius Labs where I am super exhilarated and days when I can be super deflated. This was a similar story during my stint in academia. 


Where do you see the most promising career opportunities in the future?

For someone working in computer vision and AI, the next 5 to 10 years has many things to look forward to. Mainly due to the fact they have got out of the incubation phase and are making its journey to mass adoption. I think this is the same in an industry like the space sector (maybe a bit earlier). The fundamentals of technology, product and commercialization, is still to be invented and is ripe for innovation. I will personally recommend to someone who wants to spend time on AI and computer vision on figuring out the following: 

  • From a technological perspective, how to make machines learn as automatically as possible from as little data will be the key.
  • From a product perspective, how to deploy technology effectively might be the key.
  • From a commercialization perspective, nailing the core pain points that technology can solve above human grade performance, and will also result in strong ROIis the holy grail. 


What advice/resources would you share with the next generation following a similar career path?

It’s a great time, since technology (I am talking related to AI specifically) is in a state where you have enough starting resources to get your hands dirty and build things that work without too much of an initial investment. Try to get feedback, and use this as an empirical test bed to identify your core strengths and passions. Rinse and repeat! 


Is there anything else you would like to share?

Have fun, seek excitement, and discover new horizons!


The Space Talent Spotlight is our blog series focused on the leaders and builders at the intersection of space and tech.The Space Talent Spotlight is our blog series focused on the leaders and builders at the intersection of space and tech.





BACK TO ALL

Show notes

Episode Transcript

DOWNLOAD TRANSCRIPT

:

Space Talent Spotlight: Appu Shaji

BACK TO ALL