EpiSci

CREATING A SECURE AUTONOMOUS FUTURE

Our mission is to solve our nation’s most complex national security problems with cutting-edge artificially intelligent systems

What We Do

Our elite team of scientists, researchers and engineers are bridging the gap between rapid advances in commercial machine learning techniques and deploying for tactical applications.

MACHINE LEARNING

TACTICAL APPLICATIONS

We are tackling the fundamental challenges of adopting AI for the defense industry:

Reliability and adaptability due to limited datasets
High levels of uncertainty
Safety against “surprises” (deviations from the training environment)

We know what autonomous systems require:

Necessity to operate in rapidly shifting, complex environments
Ability to respond intelligently to unforeseen changes in real time

What Is Tactical AI?

Tactical AI is our patent-pending hybrid machine learning framework that blends three powerful machine learning paradigms together: Adaptive Autonomous Rule-based Learning (A2RL), Reinforcement Learning (RL), and Deep Learning (DL). Cross-disciplinary, Tactical AI enables unprecedented advances over single-mode machine learning techniques by dynamically exploiting the complementary strengths of each individual paradigm. Most importantly, Tactical AI’s greatest capability is to detect unforeseen, surprising, or anomalous trends of situational dynamics that were not learned during training, and adapt to them on-the-fly with real-time and local re-learning, while allowing human operators to interact with the system during the operations as needed.

How does Tactical AI work?

Through our proprietary hybrid meta learning methodology, Tactical AI simultaneously learns the structure of the underlying data and environment through A2RL while learning the features of observational, experimental, and simulated data either via DL or RL. The A2RL component also allows Tactical AI to be taught by a human in an intuitive way when data is sparse, unavailable, or surprising.

What is Tactical AI designed for?

Tactical platforms and systems that cannot autonomously adapt in real-time not only jeopardize operational efficiency and mission success, but also endanger human lives. Designed to handle the unknown and complex operational environments across a broad spectrum of mission areas, Tactical AI aims to solve problems that conventional neural networks or deep learning techniques cannot do well alone in such environments.

Why Tactical AI?

Tactical AI empowers designers, engineers, and operators to build robust and intuitive intelligent systems to be trainable through both data and human inputs naturally and autonomously. Tactical AI readily outperforms human-generated hand-crafted rules that are not well designed to adapt in real time without the labor intensive costs of optimization and adaptation. As a result, it drastically reduces the time and cost required to train future autonomous systems while significantly improving their autonomous decision-making capabilities.

Areas of Expertise

From advanced decision-making algorithms, to cutting edge SIGINT systems, to UAV swarms, we are working across disciplines to develop Tactical AI-powered systems:

1/6