Learn more about Enabled Intelligence.
Through supporting the AI data requirements of our US Defense and Intelligence Community customers, several things have become clear:
- End-user requirements are evolving rapidly in response to new applications and challenges;
- No one AI tool or platform is future proof or can “do it all”: instead, there are many excellent commercial, government developed, and opensource AI/ML Ops tools out there, some better suited for certain requirements than others;
- Given the importance of being able to share and integrate data, and given how quickly requirements can change, US Defense and IC customers want to avoid “vendor-lock”, i.e. being forced to rely on a single vendor’s proprietary tools, interfaces, and data formats.
This understanding has led to Enabled Intelligence’s new EnkI™ (“n-key”) AI/ML Ops Platform, which represents a fundamentally better approach for mission-critical AI.
Enkl maximizes scalability and quality, while limiting “vendor lock” and ensuring that data remains fully owned and accessible to our customers.
To do this, EI has built proprietary APIs, scripts, and software that automate data set management and preparation. EI uses these proprietary tools in combination with rigorously tested COTS and government AI data annotation and model development solutions.
Overview
EnkI is an integrated, fully developed, and commercially available software suite of multiple AI development tools designed to support U.S. Government and commercial AI data annotation and model development requirements. The main technology and workflows for developing automated image analysis software are immediately operational and deployable for US government and commercial customers. The EnkI AI platform supports multiple data types including various forms of imagery, video, text and audio, and manages all phases of AI model development including:
DATA INGEST: The platform can ingest data from cloud or local data stores via API’s or direct data links. The platform is currently used by government customers for electro optical (EO), synthetic aperture radar (SAR), hyperspectral imagery (HSI), full motion video (FMV), infrared imagery, telescope imagery, ground-based video, drone and airborne imagery, social media data, and almost any other form of imagery. The ingestion process captures and maintains all metadata including geocoordinates, time stamps, image formats, types, quality, sensor origin, etc. The platform works with most common data formats including jpeg, NTIFF, GeoTIFF, etc.
BATCH DATA PREPARATION: Enkl can handle full imagery and video and automatically batch prep data into image chips or down sampled video segments. The platform also automatically prepares data for labeling and/or model training, including image chipping and restitching, video down sampling, and meta data characterization. EI’s orchestration engine automatically creates labeling campaigns and assigns and manages labeling resources depending on data type, security permissions, ontologies, and annotation methodology (e.g. machine assisted labeling, human labeling, etc.)
CV TASK AND ONTOLOGY GENERATION: EI’s Enkl platform allows for fully customized ontologies and labeling structures. The platform can create flat or nested ontologies specific to each project, labeling campaign, or model use case. The platform also supports linked Quality Control ontologies related to specific labeling campaigns or model testing processes to identify and measure labeling or model errors.
LABELING: Enkl is primarily built for labeling campaigns and is focused on ensuring high levels of label accuracy and work with a local and/or distributed workforce. EI’s platform is open-architecture and includes additional modules for machine-assisted labeling, synthetic data, and automated quality control tools. The platform can handle imagery, text, audio, radio frequencies and other data types. The platform supports multiple annotation geometries including, image aligned bounding boxes, object aligned bounding boxes, polygons, center points, vectors, audio segments, and automated transitions between them (e.g. bounding box to polygon).
AUTOMATED LABEL AUGMENTATION: Enkl fully integrates third party tools to assist in data augmentation. EI’s platform can automatically evaluate data sets to ensure full representation of sensors type, data quality, biome, object class imbalance, label quantity, etc.). The EI platform includes tools to synthetically augment or adjust existing data or improve data set representation.
MODEL TRAINING: Enkl allows operators and data science team to create and segment training and test data sets and automatically train models from these sets. Models may be trained using cloud compute resources or locally connected compute/GPUs’.
MODEL SCORING: Enkl includes a model testing harness that evaluates models for typical model metrics (precision, recall, F1). The platform also provides deeper evaluations identifying differences in precision and recall per object type or characteristic. For example, a model may score well on Recall for Syrian ground vehicles. But the platform will note that the data set includes very few Syrian T-72As and the model missed all of them. While the Recall is sufficient, the model will need retraining to ensure comprehensive and reliable performance. The EI platform also assesses the “weight” of models and can determine how much compute is needed to adequately operate them. This is critical information determining if models can be “run at the edge” or need access to cloud or local compute.
MODEL CONTAINERIZATION: All models developed on Enkl are fully containerized using standard Docker and Kubernetes formats and protocols.
Learn More
Enabled Intelligence is using the Enkl platform today to support classified and non-classified US government and commercial AI projects involving various types of geospatial, aerial, and ground-based imagery and video. To request a demonstration of Enkl, go to https://enabledintelligence.net/contact/
/
