Transparency and accountability through machine learning
Today, officers spend 65% of their time dealing with paperwork, leaving only 35% of their time available to build deep relationships with the communities that they serve. Axon's mission is to improve transparency, accountability, and productivity in law enforcement via technology.
Axon Research's goal is to use deep learning to accelerate the many manual human workflows in public safety data, so that officers can spend more time with what matters most: the community. We focus on workflows, while continuing to entrust decision-making to officers.
Our product is a machine learning platform that uses deep learning to understand video footage and power three key use cases:
Accelerated footage review to enable supervisors to better understand their officers' behavior and provide positive or negative feedback in order to improve training and police/community relations.
Automated redaction to speed up the process of sharing footage with the public while protecting the privacy of citizens captured in video.
Automated reporting to populate factual records directly from video and audio, so officers can spend more time serving the community.
By removing the burden of manual notes and endless hours of keyboard work, officers can be present, build relationships and be more human. We want our technology to enable more personal interactions to help build safer and stronger communities.
How we do it
Axon Research is a group comprised of machine learning researchers and developers. We balance an aggressive roadmap of engineering challenges with a portfolio of research bets. All of our engineering work is in service of ML, and all of our research work will make major impacts to product if successful. We understand the importance of collaboration for rigorous research and our whole team is encouraged to publish and open-source. As the deep learning technology and team came from the acquisition of Dextro, we maintain a highly startup-centric culture.
Like Apple, Axon is a vertically-integrated company that builds the entire stack, from camera hardware to firmware to cloud evidence management solutions, and now all the way up to AI. This tight coupling enables incredible scale with machine learning across both training and inference.
Our work requires much more care than typical AI applications such as targeting ads or tagging photos; therefore, we are investing time into a whole new level of applied AI security, including understanding adversarial examples, as well as a wholly independent AI ethics board that guarantees that our work will be used only for transparency and accountability, never for surveillance.
Footage from body worn cameras and in-car cameras is no ImageNet, or even a Youtube-8M for that matter. The domain is relatively unconstrained; the footage is extremely shaky, blurry, and spans day and night; the concepts we need to understand span all sorts of physical sizes and time scales; and most importantly, the cameras are never being “focused” on anything because there is no operator intentionally filming things. We have the challenging task of determining what a video is about as a whole, as well as creating a fine-grained temporal segmentation of all sorts of actions and small objects that are present on-screen at the same time.
Though it's difficult and unsolved, the impact of our work is immense-unlike most ML research in industry, the success of your models directly impacts the status quo of transparency and accountability in law enforcement.
We're building a team that spans diversity of all kinds, which we think is critical in our mission to bring officers and the communities they serve closer together. If this is something you care about, we would love to have you join us.
Your Day to Day
- You will design new models/techniques for understanding video content.
- You will pair with our research engineers to iterate on ideas as well as scale them out to production once they are ready.
- You will continually read the state of the art of research and publish novel work yourself.
- You will see that generalizable parts of your work get open-sourced.
- Ideal Scientist would have done original deep learning research pertaining to videos (categorization, summarization, captioning, tracking etc).
- Demonstrable research rigor via a PhD or a long publication track record in machine learning, computer vision, robotics, or a similar field working with unstructured data
- You have an unimpeachable grasp of deep learning theory and are comfortable designing new models, not just modifying existing ones.
- You have a good understanding of the complete workflow from collecting and annotating the appropriate data for the task at hand to experimenting with and comparing different ML techniques to analyzing results.
- You're able to execute on your ideas and have competence in frameworks like Tensorflow/Keras/Torch/Caffe/Darknet.
- You care about getting to the “truth” of things and will dig deep to prove or disprove a hypothesis
Compensation and Benefits
- Bonus plan
- Stock options
- Supportive parental leave policy
- Unlimited paid-time-off
- Commuter options
- Stocked kitchen
- Awarded 14th best place to work in the state of Washington and the geekiest office space in Seattle
- Opportunities to ride along with real US police officers in real life situations, see them use technology, and get inspired.
- And much more...