5 min read
Tractable teams up with Torr Vision Group to push the boundaries of uncertainty
How do you teach AI systems to make accurate decisions in the face of uncertainty? It’s a seriously exciting area of investigation for the right person.
From day one, Tractable has been focused on taking novel academic research and theories and applying them to real world scenarios.
In fact, we’re one of the only companies in the world successfully fusing and applying artificial intelligence (AI) disciplines including computer vision, natural language processing and machine learning (ML) to commercial problems at scale.
As a business, we are focused on visually assessing cars and homes that have typically been damaged. In so doing, our AI acts as a trusted standard that connects everyone involved in insurance, repairs, and sales of cars and properties. Our solutions aim to help people work faster and smarter, while reducing friction and waste – better for businesses, their customers and the planet.
Evidence of our success can be seen in our customer list. Some of the most successful companies in the Fortune 500 want to buy our technology. And the list is only growing.
The heart of our technology is our AI. It’s been built on a bedrock of research and patented intellectual property (IP) rather than cookie cutter techniques, as can be the case in industry. It powers all our products across every industry we touch. We’re now focused on building the next generation of our AI.
To help, we are sponsoring a Post Doc role with the world-renowned Torr Vision Group, which, driven by the needs of society, conducts state of the art research into computer vision and AI. It’s a unique opportunity for the right person to solve some of the toughest pure research problems related to uncertainty and robustness. Ken Chatfield, Vice President of Research Science at Tractable, and Prof. Phil Torr – who leads the multi-award winning Torr Vision Group – told us more.
"This role is about chipping away at the big, fundamental question related to uncertainty and robustness: how to teach AI systems to understand better when they don’t know? It’s a seriously exciting area of investigation for the right person.”
Q&A with Ken Chatfield
What do the research team at Tractable do?
“We are in the business of building AI systems that can make or assist in the making of complex decisions, much like an expert would, when assessing damage to houses and cars – often in the face of uncertainty. Our research teams work on creating novel ideas to some of the most complex problems in our space, validating their thinking in real scenarios which of course informs production. We think that’s pretty unique.”
What is the focus of this Post Doc role?
“One of the big open challenges is that AI systems don’t predict or make accurate decisions in the face of uncertainty. And they don’t know those decisions are inaccurate. So this role is about chipping away at the big, fundamental question related to uncertainty and robustness: how to teach AI systems to understand better when they don’t know? It’s a seriously exciting area of investigation for the right person.”
So who is the right person for this role?
“Aside from a PhD and a strong track record of research publications, the right person will already have a good understanding of uncertainty and robustness. And they are itching to put their theories to the test and use our resources to validate their thinking. We’re looking for someone with a can do, energetic, entrepreneurial spirit.”
What happens after 12 months?
“Well, if they enjoy their experience in the commercial world, there’s an option to join Tractable permanently. If not, they can of course stay in academia having solved and validated complex research problems that relate to uncertainty and robustness at one of the world’s most renowned computer vision research groups. There’s nothing to lose.”
Interested in applying?
Read more about the role here.
Q&A with Prof. Phil Torr
Why is your team best suited to this problem?
“Apart from a good personal relationship with Tractable, the Torr Vision Group has already undertaken a vast body of research within the area of uncertainty quantification for neural networks along with robustness and reliability. That means we’ll be in a strong position to help develop and progress any ideas on the table. Our approach is strongly grounded and based on a bias to empirical validation rather than building theory in a vacuum. With Tractable’s resources and expertise in applied AI, this person’s ideas can be quickly put to the test. It really is a great opportunity to contribute to a fundamental missing gap in today’s ML systems.”
Why is it such an interesting problem to crack?
"There are two main reasons. Firstly, neural networks have driven a great deal of progress across many challenging visual assessment tasks, but they are still incredibly brittle. This is a problem for applications, because outputs from models can be unreliable. However, the work that TVG has been doing on uncertainty and robustness provides a potential route to address this problem.
Secondly, Tractable’s mix of real-world assessment tasks are far more challenging than digit recognition, but also much more tangible than solving general intelligence. Tractable provides a great context in which to develop and validate novel algorithms bringing real value whilst moving the field forwards."
How will this work between Torr Vision Group and Tractable?
"This role will report to one of my PIs but there will be frequent interaction with the research team at Tractable. Given this is open research, the primary output will be research publications documenting the algorithms developed. As for what it’s like to work at Torr Vision Group, I like to give a lot of autonomy and promote an open working culture. So aside from someone with a deep curiosity in expanding the capabilities of modern ML systems where they are most limited, a bias for action is strongly preferred."
What about Oxford as a location?
"Locations tend to matter less after COVID but it’s helpful that we have a world-class engineering department on our doorstep as well as a complimentary ecosystem of academics. More importantly, it’s often quicker to get back to Oxford after a night or day in London than using London’s public transport system to get to wherever your flat is in Zone 3!"
Interested in applying?
Read more about the role here.