Detect
Find behavior that departs from the expected pattern in its context. An outlier is a reason to examine the evidence, not an automatic verdict.
One primitive.
Three ways to act.
Behavioral Outlier Optimization began while coding models for fraud discovery. It connects detection, attention, and surprise.
DetectionAttentionSurprise
From fraud to
adaptive defense
The starting question was practical: which changes in behavior deserve a closer look? While coding models for fraud discovery, our founder discovered a primitive that connected finding departures, directing attention, and recognizing surprise.
A primitive is a reusable building block. Behavioral Outlier Optimization gives those three jobs a common foundation, from the first unusual signal to the decision about what deserves more computation.
Attend
Direct computation toward the observations and specialist models most relevant to a changing situation. Attention becomes a choice about where to spend effort.
Reconsider
Use unexpected change to challenge an expectation. Surprise can trigger another look, a different specialist, or a revised assessment.
Geometry gives
change meaning.
Information geometry
Information geometry studies the structure of probability models. It asks how distinguishable two states are, rather than treating every numerical difference as equally meaningful.
For our work, that provides a language for sensitivity, uncertainty, and behavioral change. The public story is the role it plays; the implementation remains part of the technical evaluation.
Spiking neural networks
Spiking neural networks represent activity through events over time. The timing and accumulation of those events can matter as much as their magnitude.
We explore how that temporal view can work alongside geometric surprise and specialized models. Efficiency is something to measure on the target hardware.
Machines set
the tempo.
Machine speed is needed to act against machines. When the other side can observe, adapt, and execute automatically, a defensive decision has to arrive while it can still change the outcome.
That makes the path from observation to action an engineering problem: local decisions, bounded work, and clear operator policy.
Explore OuijaBackground reading: Information geometry — Frank Nielsen · Neuronal Dynamics — Gerstner and colleagues