Making Machine Intelligence Available To All
Posted on February 2, 2016 by andrei
Alex Kern and Nikhil Srinivasan believe machine intelligence will shape the future of humanity. Advances in data science will soon begin to reshape existing industries, while paving the way for entirely new ones. Their thesis: rather than focus on model development, the two would rather create the supporting infrastructure, tooling, and marketplace to make these technologies more broadly accessible by the wider developer community.
By striving to democratize access to the tools of data science, individuals can focus on deriving insights from their data instead of building and operating complex machine learning pipelines. With Pavlov, Alex and Nikhil plan to reduce the operational complexity associated with data science, by providing a framework to design and develop the next generation of intelligent applications.
Pavlov was started out of their frustrations building applications using modern advances in deep learning. Nikhil spent five years at Harvard Medical School, where he recognized the challenges researchers faced in operating data processing pipelines. Alex spent time at NASA’s JPL and Apple building distributed machine learning infrastructure and realized these pain points personally.
I had an opportunity to sit down with the two to talk more about what they’re building. In the arti
cle, we explore questions ranging from what they’ve built, to advice they have for others with an interest in this space.
What they’re building.
Pavlov is a framework that allows developers to focus on the “magic” of machine learning by letting them focus on implementation and predictions rather than infrastructure. Rather than managing ad-hoc data pipelines, Pavlov provides the infrastructure layer that enables teams to quickly build, deploy, and scale models. The company hopes to tackle every layer in the digital intelligence stack, from proprietary hardware to human intelligence for supervised learning, allowing users to focus on application development and model optimization.
One of the bigger challenges with the evolving landscape is that it’s deeply rooted in an academia – most of the innovation is being driven by researchers, creating a huge barrier for anyone interested in learning more. Industry messaging such as “regional convolutional neural networks” and the general lack of enterprise support for available software make data science more a craft than a hard skill.
After going through the first YC Fellowship batch, the team has been laser focused on delivering their machine intelligence solution to early enterprise customers and are focused on delivering computer vision solutions.
The problem they’re trying to solve.
Both commercial and academic research efforts are producing enormous amounts of data, yet people are still making critical decisions driven by intuition rather than data-driven insight. While significant efforts have been made building systems that can find the signal in the noise, very few have delivered robust solutions for computer vision problems – this is where Pavlov comes in. The team is building a toolkit that will allow users to build robust applications that leverage the latest advantages, termed ‘convolutional neural networks’ to make sense of everything from satellite to medical imagery.
Hardest decision they’ve had to make.
While it’s clear machine intelligence is transforming enterprise, one of the bigger challenges in this space is establishing product-market fit, one that supports clear business value despite how quickly the field is evolving. Their hardest decisions were ones where they turned down opportunities – as with any startup, your most valuable resource is your time and focus is valuable. This is especially critical with companies in this space, given the inherent research rather than product driven innovation cycle.
Why they think this is the right time.
This is the right time for them because there are a lot of exciting developments in the field. With Google opening up Tensor Flow – an open source software library for machine learning, and computer vision delivering performance improvements, Alex & Nikhil believe that this is the renaissance period for all things machine learning. There are a ton of companies who want to get into the consumer space, however Alex and Nikhil see the opportunity within the enterprise space.
Their vision is to build the infrastructure and tooling that makes it easier for enterprise customers to become heavily invested in this space and hope to serve as an abstraction for the infrastructure and human intelligence layers. While they do not have any direct competitors, companies like Palantir and Orbital Insights come to mind when trying to find analogous data solutions – that said, Pavlov is unique in it’s primary focus as a full-stack computer vision solution.
You were part of the inaugural YC Fellowship class. What was the best question a Y Combinator Partner asked you?
When pitching their product to a technical partner, during Y Combinator’s office hours, hearing the question about simplification, was a great turning point for the two founders. Realizing that they have been embedded within the data science/machine learning community so long that not even a technical partner could understand what they were saying, allowed them to take a step back and take a simpler approach when pitching in order to adequately explain their target market and how they are positioning their business.
What keeps you up at night?
The number one thing that keeps the founders up at night is the fact that training these algorithms is quite expensive. To compound the issue, the current capabilities for cloud computing is not quite where it needs to be for Pavlov to be successful. As a result the co founders will need to build their own computers and server farms – something that they will have to learn along the way. Having their own hardware will allow for much quicker turnaround times for clients, which will prove essential in the long run when working on more time-sensitive contracts.
What will you be celebrating 1 year from now?
One year from now, the team hopes to have multiple paying customers with recurring revenue. By this time they also hope to have been surprised by different use cases that customers have for their tool kit. Most importantly though they want to deliver on what they’ve both envisioned for the product, and would like to feel confident that what they’re working on has the potential to make an impact in the field.
Any advice for aspiring data scientists?
The general advice that Alex & Nikhil have offered is that you should simply jump in and start getting involved with what’s happening in the data science space. People need to be aware that there’s not a lot of conventional wisdom right now, which means there is a lot of opportunity to learn and break things.
For those looking to get early access to the beta test, feel free to check out Pavlov. To further expand your knowledge in the data science world, as always, our courses at BDU, offer you an opportunity to get free access to learn more about these concepts.