Experience Transformation Through the Power of AI

SFL SCIENTIFIC’S UAV-BASED AI SYSTEM DEPLOYS ALONGSIDE DOD OPERATIVES IN FIELD EXERCISES, RECEIVING POSITIVE EVALUATIONS FROM WARFIGHTERS

Cutting-edge technologies compete at the Defense Threat Reduction Agency’s annual Chemical Biological Operations Analysis event. SFL Scientific’s drone-mounted machine learning system, OptoAI, deployed alongside warfighters to test its capabilities in the field. OptoAI attaches to existing UAVs, which then become semi-autonomous: Multiple drones can be dispatched on patrol, inspecting or mapping flights using one-click autopilot software. Multi-modal sensors recognize entities on EO, IR, and CBRN spectrums. Embedded AI models are custom trained to detect mission-specific targets of interest even under low-data constraints. Drone feeds and detected targets are displayed in real time on 2D touchscreens and holographic 3D maps, giving users direct insight and control into the area of operations.

OptoAI’s flexible situational awareness is dual-use: Its adaptive tracking and real-time BVLOS capabilities support ISR needs, while defect inspection workflows benefit from its high accuracy and LiDAR-based 3D digital mapping functionality. Dr. Mike Segala, OptoAI’s director, reflected: “OptoAI isn’t a replacement; it’s a tool to empower people to get things done more effectively, cheaper, and safer.”

PROTEIN PREDICTION AND SURFACE MAPPING

This white paper describes SFL Scientific’s custom-built protein prediction tool, called ProMap (Protein Mapping), which combines recent advances in machine learning and bioinformatics algorithms to predict an allergen’s protein structure and identify its most likely immunodominant epitopes. Our tool facilitates rapid and low-cost allergen research for diagnostic and therapeutic applications, eliminating the time and cost needed to collect a large amount of patient data and experimentally locate epitope sites. As a result, using ProMap, pharmaceutical companies can iterate on allergic disease therapies more quickly and at a lower cost, decreasing the time to market and the cost of development for novel treatments.

IMPROVING PREDICTION VIA OPTIMAL MODEL SELECTION

We present here a novel optimal model classifier: a deep learning model that can accurately predict the preferred model for structure prediction using only the amino acid sequence, with consideration given to both the runtime to generate the predicted structure and final structure accuracy. The optimal model classifier allows clients to specify a trade-off between predictive accuracy and inference runtime, allowing the predictions to be customized for each use case’s needs. We show that the use of this model to select the optimal folding model achieves a new state-of-the-art predictive accuracy when focused solely on maximizing prediction quality.