1. Description
In machine learning, ontologies are increasingly used to provide ML models based on similarity analysis and scene knowledge.
In the traditional label-based definition, objects are often isolated, poor scalability, there is a possibility of duplication, and the relationship between objects cannot be reflected. In ontology-based definition, objects no longer exist in isolation, and functions such as scene search, ontology fusion, and ontology recommendation can also be realized through relational annotation.
Ontology was first introduced by Xtreme1, the world's first open source multi-sensory training data platform, to abstract the definition of AI problems from various model requirements. It can be reused and extended to build a knowledge base of AI algorithms to accelerate model development.
2. What is Ontology?
An ontology is a structured way of describing everything in the world and consists of three elements:
class — represents a type, label, or abstract class representing an instance;
Relation - Represents the relationship between descriptions, which can be a directed or undirected graph. For example, in an autonomous driving scenario, cars may have "parallel", "overtaking", etc. relation;
Attributes - Represents the attributes of a node or relationship. For example, attributes of a "car" might include "color", "windows open/closed", etc., while attributes of a "pedestrian" might include their "gender" or "mask on/off".
Autonomous driving is one of the most promising and challenging research topics for AI companies and the automotive industry. At present, mainstream self-driving cars are equipped with some high-sensitivity sensors, such as cameras, lidar and radar equipment. While these sensors may already be able to accurately identify specific objects, such as cars or traffic signs prohibiting turns, without a comprehensive understanding of the data context, the vehicle cannot understand the meaning of the driving environment. Therefore, a machine-friendly knowledge representation is needed to bridge the gap between perceiving the driving environment and processing knowledge.
3. What is the role of ontology in data management?
After defining classes and attributes in the ontology center, users can easily search for scenes such as "Chage Lane". The ontology hub can also infer new annotations based on rules among classes, attributes, and relations. As the amount of ontology data increases, the ontology center can also recommend ontology models with better performance in different fields.
3.1 Scene search
Scenario Search addresses the problem of how to define and find data that occurs in a particular scenario.
Traditionally, when we manage data by defining it through a label hierarchy, a problem arises that data labels are too general to pinpoint specific problems in the data scenario. Meanwhile, in traditional label-based definitions, objects are too isolated to avoid repetition or show connections to other objects.
The schema search function defines objects by classes and attributes, and schemas by relationships and attributes between objects. Scenarios such as lane changes, stops, turns and runway incursions are easily defined and looked up.
3.2 Ontological reasoning
In annotation and quality control, new labeling results or problematic annotations can be inferred based on rules between attributes, classes, and relationships. For example, in an autonomous driving scenario, red lights, green lights, and yellow lights belong to the same traffic light. If the status of the red light is "on", then it can be inferred that the green and yellow lights must not be on. If both red and green lights are marked as "on", there may be a problem with the marking results.
3.3 Destruction of main body
In annotation, it is very common to encounter different definitions within the same data batch. Ontology fusion can help users resolve these inconsistencies.
3.4 Ontology recommendation
In the SaaS version of the open-source Xtreme1, when the data is accumulated to the PB level, higher performance ontology models can be provided for common model requirements in various fields to facilitate customized solutions.
The core ontology of safe autonomous driving
4. Highlights of Xtreme1 v0.5.5:
The new ontology center aims to manage ontology and data across data sets, and improve industry templates and solutions for model training based on scenarios;
Classification and classification of CRUD ontology in ontology center;
Ontology fusion between classes in the dataset and classes in the ontology hub;
· Export and import ontologies in ontology centers and datasets;
Copy classes and/or taxonomies from ontology hub;
Push/pull of classes in datasets and ontologies;
Search for solutions of the same data type across datasets;
· Export search results as a JSON file or as a new dataset.
Planned features in future releases of Xtreme1 include:
Annotate relationships and search by class and scene;
· Attribute search (in schema search) by class, relationship and/or attribute.
Website | Xtreme1.io
Documentation | docs.xtreme1.io
GitHub repository | github.com/xtreme1.io/xtreme1
slack | xtreme1io.slack.com
5. Reference:
The core ontology of safe autonomous driving: https://ceur-ws.org/Vol-1486/paper_9.pdf