self-supervision, the path to AI widespread deployment


Our technology incorporates new research advancements like generative adversarial networks (GANs) and transfer learning with forward predictors to make self-supervised AI accessible to non-engineers. 

There is a wealth of event-focused data in the enterprise like video, asset tracking logs and application telemetry. We leverage the inherent structure of event-focused data to bootstrap machine learning applications with minimal human curation.



AI Applications that Stem from Structure Discovery


Asset Tracking

Increasing volumes of real-time asset data is captured in many industries, yet it hasn't translated into operational gains. Manually interpreting this information is far too slow and expensive to be realistic.

Our AI agents can be applied to physical movement data to discover latent attributes like loss / misappropriation, theft, damage / defects.

Video is the most generic sensor and is becoming ubiquitous in our cities and work environments.

Internal structure of videos can be exploited by our AI agents to detect anomalous events and to understand behavior / sentiment.


App Telemetry

Machine generated data for application logging, debugging, security, etc. can easily add up to billions of entries per day even in a mid-sized organization.

Our AI agents can discover service quality problems and indicators of security compromise as well as highlight opportunities for streamlining processes. 


Deploy On-premise without Customization



Reinforcement learning is best known for teaching AI how to play video games. It defines an environment (that can be sampled sequentially and acted on) and an agent that wants to maximize a reward by learning to navigate this environment effectively. 

This framework provides a clean separation of concerns and is used by Adversarial.AI in all modes of learning. The domain specific information related to a particular use case or organization are encoded in the environment definition. 



Our agents are generic and self-optimizing.  They support multiple modes of learning including supervised, structured prediction, reinforcement learning, generative adversarial networks, etc.

Our intuitive user interface allows for visualization of agent performance and easy way to vet various agent types against a given environment. 


Recent Blog Posts