OGC Testbed-14: Machine Learning Engineering Report

This OGC Engineering Report (ER) describes the application and use of OGC Web Services (OWS) for integrating Machine Learning (ML), Deep Learning (DL) and Artificial Intelligence (AI) in the OGC Testbed-14 Modeling, Portrayal, and Quality of Service (MoPoQ) Thread. This report is intended to present a holistic approach on how to support and integrate emerging AI and ML tools using OWS, as well as publishing their input and outputs. This approach should seek efficiency and effectiveness of knowledge sharing.

This engineering report will describe: experiences, lessons learned, best practices for workflows, service interaction patterns, application schemas, and use of controlled vocabularies. It is expected that the description of workflows for geospatial feature extraction will be more complex than the implementations found in the deliverables.

The following normative documents are referenced in this document.

For the purposes of this report, the definitions specified in Clause 4 of the OWS Common Implementation Standard OGC 06-121r9 shall apply. In addition, the following terms and definitions apply.

Active learning is a special case of semi-supervised machine learning in which a learning algorithm is able to interactively query the user to obtain the desired outputs at new data points. There are situations in which unlabeled data is abundant but manually labeling is expensive. In such a scenario, learning algorithms can actively query the user/teacher for labels.

An unambiguous specification of how to solve a class of problems. Algorithms can perform calculation, data processing and automated reasoning tasks.

Manual image annotation is the process of manually defining regions in an image and creating a textual description of those regions. Automatic image annotation (also known as automatic image tagging or linguistic indexing) is the process by which a computer system automatically assigns metadata in the form of captioning or keywords to a digital image.

Conceptual schema for data required by one or more applications [SOURCE: ISO 19101‑1:2014]

Artificial Intelligence (AI) is the ability of a computer program or a machine to think and learn. An ideal (perfect) intelligent machine is a flexible agent which perceives its environment and takes actions to maximize its chance of success at some goal. An extreme goal of AI research is to create computer programs that can learn, solve problems, and think logically.

An artificial neural network is an interconnected group of nodes, akin to the vast network of neurons in a brain. The signal at a connection between artificial neurons is a real number, and the output of each artificial neuron is computed by some non-linear function of the sum of its inputs. Artificial neurons typically have a weight that adjusts as learning proceeds. See Neural Network.

A subdivision of a dataset into number of batches. A batch is required when the entire dataset is to large to be passed to a neural network.

Total number of training examples present in a single batch.

To bundle software means to sell it together with a computer, or with other hardware or software, as part of a set.

A class is the category for a classifier which is given by the target.

Classification is the problem of identifying to which of a set of categories (sub-populations) a new observation belongs, on the basis of a training set of data containing observations (or instances) whose category membership is known.

A visual representation of the output of classification, often viewed as dense, gridded imagery. See Classifier.

An algorithm or a method that processes to classification of an input. See Classification.

Generally, a group of data objects. Typical cluster models include connectivity models, distribution models, density models and neural models.

Clustering is the task of grouping a set of objects in such a way that objects in the same group (called a cluster) are more similar (in some sense) to each other than to those in other groups (clusters). A "clustering" is essentially a set of such clusters, usually containing all objects in the data set. Additionally, it may specify the relationship of the clusters to each other.

Controlled vocabularies provide a way to organize knowledge for subsequent retrieval. They are used in subject indexing schemes, subject headings, thesauri, taxonomies and other forms of knowledge organization systems.

A convolutional neural network (CNN) uses convolutions to extract features from local regions of an input. CNNs have gained popularity particularly through their excellent performance on visual recognition tasks.

Identifiable collection of data.

Deep Learning is a class of machine learning algorithms that: use a cascade of multiple layers of nonlinear processing units for feature extraction and transformation; learn in supervised and/or unsupervised manners; learn multiple levels of representations that correspond to different levels of abstraction.

A deep neural network (DNN) is an artificial neural network (ANN) with multiple layers between the input and output layers. See Artificial Neural Network.

Detection includes methods for computing abstractions of image information and making local decisions at every image point whether there is an image feature of a given type at that point or not. The resulting features are subsets of the image domain, often in the form of isolated points, continuous curves or connected regions.

A thing with distinct and independent existence. An entity is something that exists as itself, as a subject or as an object, actually or potentially, concretely or abstractly, physically or not. An entity-relationship (ER) model describes interrelated things of interest in a specific domain of knowledge. A basic ER model is composed of entity types (which classify the things of interest) and specifies relationships that can exist between instances of those entity types.

Learning iteration on a training dataset.

In machine learning, feature extraction starts from an initial set of measured data and builds derived values (features) intended to be informative and non-redundant, facilitating the subsequent learning and generalization steps, and in some cases leading to better human interpretations. Feature extraction is related to dimensionality reduction. See Feature.

Abstraction of real-world phenomena. A feature can occur as a type or an instance. Feature type or feature instance should be used when only one is meant [SOURCE: ISO 19101‑1:2014].

Feature engineering is the process of using domain knowledge of the data to create features that make machine learning algorithms work. The need for manual feature engineering can be obviated by automated feature learning. See Feature Learning.

Feature learning is a set of techniques that allows a system to automatically discover the representations needed for feature detection or classification from raw data. This replaces manual feature engineering and allows a machine to both learn the features and use them to perform a specific task. See Feature.

Fine-Tuning refers to the technique of initializing a network with parameters from another task, and then updating these parameters based on the task at hand. Fine tuning is a common technique is to train the network on a larger data set from a related domain. Once the network parameters have converged an additional training step is performed using the in-domain data to fine-tune the network weights. This allows convolutional networks to be successfully applied to problems with small training sets.

Generative adversarial networks (GANs) are a class of artificial intelligence algorithms used in unsupervised machine learning, implemented by a system of two neural networks contesting with each other in a zero-sum game framework. One network generates candidates (generative) and the other evaluates them (discriminative). Training the discriminator involves presenting it with samples from the dataset, until it reaches some level of accuracy. See Sample.

In machine learning, the term "ground truth" refers to the accuracy of the training set’s classification for supervised learning techniques.

In machine learning, a hyperparameter is a parameter whose value is set before the learning process begins. By contrast, the values of other parameters are derived via training. Given these hyperparameters, the training algorithm learns the parameters from the data. The time required to train and test a model can depend upon the choice of its hyperparameters. A hyperparameter is usually of continuous or integer type, leading to mixed-type optimization problems.

In machine learning systems, the number of batches needed to complete one epoch.

A knowledge base (KB) is a technology used to store complex structured and unstructured information used by a computer system. A knowledge-based system consists of a knowledge-base that represents facts about the world and an inference engine that can reason about those facts and use rules and other forms of logic to deduce new facts or highlight inconsistencies.

In traditional feed-forward neural networks, a hidden layer neuron is a neuron whose output is connected to the inputs of other neurons and is therefore not visible as a network output. See Layer.

See Class.

Basic unit of geographic information that may be requested as a map from a server. In the context of neural networks, a layer is an ensemble of neurons in a network that processes a set of inputs and their results.

Machine learning is a subset of artificial intelligence in the field of computer science that often uses statistical techniques to give computers the ability to progressively improve performance on a specific task with data, without being explicitly programmed.

Data that provides information about other data.

MLlib is Apache Spark’s scalable machine learning library. It provides tools such as ML Algorithms, Featurization, Pipeline, Persistence and Utilities.

A statistical model is a mathematical model that embodies a set of statistical assumptions concerning the generation of some sample data and similar data from a larger population. A statistical model represents, often in considerably idealized form, the data-generating process.

A computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs.

The OpenAPI Specification is an Application Programming Interface (API) description format for REST APIs.

In computer programming, a parameter is a special kind of variable, used in a subroutine to refer to one of the pieces of data provided as input to the subroutine. In convolutional layers, the layer’s parameters consist of a set of learnable filters or kernels.

Set of one or more base standards and - where applicable - the identification of chosen clauses, classes, subsets, options and parameters of those base standards that are necessary for accomplishing a particular function [ISO 19101, ISO 19106].

PyTorch is an open source machine learning library for Python, based on Torch, used for applications such as natural language processing or image processing. PyTorch provides GPU-accelerated Tensor computation and Deep Neural Networks.

Remote sensing is the acquisition of information about an object or phenomenon without making physical contact with the object and thus in contrast to on-site observation. It generally refers to the use of satellite- or aircraft-based sensor technologies to detect and classify objects on Earth, including on the surface and in the atmosphere and oceans, based on propagated signals.

In computer vision, image segmentation is the process of partitioning a digital image into multiple segments.

A conceptualization of the implied meaning of information that requires words and/or symbols within a usage context.

A semantic class contains words that share a semantic feature. According to the nature of the noun, they are categorized into different semantic classes. Semantic classes may intersect. See Semantic Feature.

Semantic features represent the basic conceptual components of meaning for any lexical item.

Separation between the visual content in a digital image and semantic descriptions.

Semantic interoperability that assures that the content is understood in the same way in both systems, including by those humans interacting with the systems in a given context.

Semantic segmentation describes the process of associating each pixel of an image with a class label.

Supervised learning is the machine learning task of learning a function that maps an input to an output based on example input-output pairs. It infers a function from labeled training data consisting of a set of training examples.

Synthetic aperture radar (SAR) is a form of active remote sensing – the antenna transmits radiation that is reflected from the image area, as opposed to passive sensing, where the reflection is detected from ambient illumination. SAR image acquisition is therefore independent of natural illumination and images can be taken at night. See Remote Sensing.

Detection of targets is a specific field of study within the general scope of image processing and image understanding. From a sequence of images (usually within the visual or infrared spectral bands), it is desired to recognize a target such as a car.

A system or controlled list of values by which to categorize or classify objects.

A test dataset is a dataset that is independent of the training dataset, but that follows the same probability distribution as the training dataset. The test dataset is a dataset used to provide an unbiased evaluation of a final model fit on the training dataset. A test set is therefore a set of examples used only to assess the performance (i.e. generalization) of a fully specified classifier.

A rectangular representation of geographic data, often part of a set of such elements, covering a spatially contiguous extent which can be uniquely defined by a pair of indices for the column and row along with an identifier for the tile matrix [source: OGC 07-057r7].

A training dataset is a dataset of examples used for learning. A model is initially fit on a training dataset, that is a set of examples used to fit the parameters of the model. The model is trained on the training dataset using a supervised learning method.

See Fine Tuning.

Two type of uncertainty can be identified, epistemic and aleatoric uncertainty. Epistemic uncertainty captures ignorance about which model generated the collected data. Aleatoric uncertainty relates to information which collected data cannot explain.

Unsupervised machine learning is the machine learning task of inferring a function that describes the structure of "unlabeled" data.

A validation dataset is a set of examples used to tune the hyperparameters of a classifier. Fitted models are used to predict the responses for the observations in the validation dataset. The validation dataset provides an unbiased evaluation of a model fit on the training dataset while tuning the model’s hyperparameters.

A language user’s knowledge of words. See Controlled Vocabulary.

Section 5 provides background information found in the state-of-the-art of AI, ML, and DL that is relevant to the ML task.

Section 6 describes the proof of concept developed by the task participants. The concept, scenarios, and use cases are presented. The design of each component, deliverable by deliverable, is described. Implemented systems, processes, and workflows are reported. Metadata and application schemas are described.

Section 7 reports various approaches and good practices as identified by the participants through their own implementations or by references of authoritative material. The section presents best practices of workflows, service interaction patterns, and application schemas.

Section 8 demonstrates the proof of concept. The demonstration scenarios and relevant material are presented.

Section 9 lists the recommendations from the task participants to OGC. Recommendations addressing the creation or extension of OGC standards to support ML and AI geospatial algorithms are presented, as well as recommendations of good practices for the use of tiles or grid structures for the analysis, including DGGS.

Annex A provides an XML WPS 2.0 process description of ExecuteML as presented by the ML system.

Annex B provides an XML WPS 2.0 process description of RetrainML as presented by the ML system.

Annex C provides an XML WPS 2.0 process description of TrainML as presented by the ML system.

Annex D provides a JSON file of the default response of the Controlled Vocabulary Manager.

Annex E provides a JSON file of the full response of the Controlled Vocabulary Manager.

Annex F provides a components table for the ML Task of the MoPoQ thread of Testbed-14.

Annex G illustrates a sample Pleiades VHR image of Paris used to train models and infer classes.

Section 5 provides background information found in the state-of-the-art of AI, ML, and DL that is relevant to the ML task. This landscape study puts into context key elements from the CFP, supports future work recommendations found in the summary, and lays the groundwork for discussion.

For additional background, the CFP also lists the following sources:

Section 6 describes the proof-of-concept developed by the task participants. The concept, scenarios, and use cases are presented. The design of each component, deliverable by deliverable, is described. Implemented systems, processes, and workflows are reported. Metadata and application schemas are described.