Choice Of Data Representation Research Articles

Advances in Data‐Driven Analysis and Synthesis of 3D Indoor Scenes

AbstractThis report surveys advances in deep learning‐based modelling techniques that address four different 3D indoor scene analysis tasks, as well as synthesis of 3D indoor scenes. We describe different kinds of representations for indoor scenes, various indoor scene datasets available for research in the aforementioned areas, and discuss notable works employing machine learning models for such scene modelling tasks based on these representations. Specifically, we focus on the analysis and synthesis of 3D indoor scenes. With respect to analysis, we focus on four basic scene understanding tasks – 3D object detection, 3D scene segmentation, 3D scene reconstruction and 3D scene similarity. And for synthesis, we mainly discuss neural scene synthesis works, though also highlighting model‐driven methods that allow for human‐centric, progressive scene synthesis. We identify the challenges involved in modelling scenes for these tasks and the kind of machinery that needs to be developed to adapt to the data representation, and the task setting in general. For each of these tasks, we provide a comprehensive summary of the state‐of‐the‐art works across different axes such as the choice of data representation, backbone, evaluation metric, input, output and so on, providing an organized review of the literature. Towards the end, we discuss some interesting research directions that have the potential to make a direct impact on the way users interact and engage with these virtual scene models, making them an integral part of the metaverse.

Guest Editorial Introduction to the Special Section on Representation Learning for Visual Content Understanding

Representation learning methods allow a system to automatically learn robust and discriminative features from raw data for given goals, which play an important role in various visual content understanding applications, such as visual object segmentation, detection, tracking, recognition, and search. The performance of visual content understanding tasks is heavily dependent on the choice of data representation (or features) on which they are applied. Conventional feature representation methods usually employ transformations of data that make it easier to extract useful information, such as scale-invariant feature transform (SIFT), local binary patterns (LBP), and histogram of oriented gradients (HOG). In recent years, deep learning techniques have been widely applied to learn data-driven representations with supervised annotations and achieved great success in different visual content understanding tasks. Representative methods include the ResNet method for image classification, the DeepFace method for face recognition, and the feature pyramid networks (FPNs) method for object detection. Despite recent progresses on deep representation learning with a great amount of annotated data, how to effectively learn visual representation with limited data annotations still requires many efforts. This special section focuses on data-effective representation learning methods for visual content understanding.

Compressed Data Structures for Binary Relations in Practice

Binary relations are commonly used in Computer Science for modeling data. In addition to classical representations using matrices or lists, some compressed data structures have recently been proposed to represent binary relations in compact space, such as the $k^2$-tree and the Binary Relation Wavelet Tree (BRWT). Knowing their storage needs, supported operations and time performance is key for enabling an appropriate choice of data representation given a domain or application, its data distribution and typical operations that are computed over the data. In this work, we present an empirical comparison among several compressed representations for binary relations. We analyze their space usage and the speed of their operations using different (synthetic and real) data distributions. We include both neighborhood and set operations, also proposing algorithms for set operations for the BRWT, which were not presented before in the literature. We conclude that there is not a clear choice that outperforms the rest, but we give some recommendations of usage of each compact representation depending on the data distribution and types of operations performed over the data. We also include a scalability study of the data representations.

Data system design alters meaning in ecological data: salmon habitat restoration across the U.S. Pacific Northwest

AbstractAs an increasing variety and complexity of environmental issues confront scientists and natural resource managers, assembling the most relevant and informative data into accessible data systems becomes critical to timely problem solving. Data interoperability is the key criterion for succeeding in that assembly, and much informatics research is focused on data federation, or synthesis to produce interoperable data. However, when candidate data come from numerous, diverse, and high‐value legacy data sources, the issue of data variety or heterogeneity can be a significant impediment to interoperability. Research in informatics, computer science and philosophy has frequently focused on resolving data heterogeneity with automation, but subject matter expertise still plays a large role. In particular, human expertise is a large component in the development of tools such as data dictionaries, crosswalks, and ontologies. Such representations may not always match from one data system to another, presenting potentially inconsistent results even with the same data. Here, we use a long‐term data set on management actions designed to improve stream habitat for endangered salmon in the Pacific Northwest, to illustrate how different representations can change the underlying information content in the data system. We pass the same data set comprised of 49,619 records through three ontologies, each developed to address a rational management need, and show that the inferences drawn from the data can change with choice of data representation or ontology. One striking example shows that the use of one ontology would suggest water quality improvement projects are the rarest and most expensive restoration actions undertaken, while another will suggest these actions to be the most common and least expensive type of management actions. The discrepancy relates to the origins of the data dictionaries themselves, with one designed to catalog management actions and the other focused on ecological processes. Thus, we argue that in data federation efforts humans are “in the loop” rationally, in the form of the ontologies they have chosen, and diminishing the human component in favor of automation carries risks. Consequently, data federation exercises should be accompanied by validations in order to evaluate and manage those risks.

A provably sound TAL for back-end optimization

Typed assembly languages provide a way to generate machine-checkable safety proofs for machine-language programs. But the soundness proofs of most existing typed assembly languages are hand-written and cannot be machine-checked, which is worrisome for such large calculi. We have designed and implemented a low-level typed assembly language (LTAL) with a semantic model and established its soundness from the model. Compared to existing typed assembly languages, LTAL is more scalable and more secure; it has no macro instructions that hinder low-level optimizations such as instruction scheduling; its type constructors are expressive enough to capture dataflow information, support the compiler's choice of data representations and permit typed position-independent code; and its type-checking algorithm is completely syntax-directed.We have built a prototype system, based on Standard ML of New Jersey, that compiles most of core ML to Sparc code. We explain how we were able to make the untyped back end in SML/NJ preserve types during instruction selection and register allocation, without restricting low-level optimizations and without knowledge of any type system pervading the instruction selector and register allocator.

Options in physical database design

A cornerstone of modern database systems is physical data independence, i.e., the separation of a type and its associated operations from its physical representation in memory and on storage media. Users manipulate and query data at the logical level; the DBMS translates these logical operations to operations on files, indices, records, and disks. The efficiency of these physical operations depends very much on the choice of data representations. Choosing a physical representation for a logical database is called physical database design. The number of possible choices in physical database design is very large; moreover, they very often interact with each other. We attempt to list and classify these choices and to explore their interactions. The purpose of this paper is to provide an overview of possible options to the DBMS developer and some guidance to the DBMS administrator and user. While much of our discussion will draw on the relational data model, physical database design is of even more importance for object-oriented and extensible systems. The reasons are simple: First, the number of logical data types and their operations is larger, requiring and permitting more choices for their representation. Second, the state of the art in query optimization for these systems is much less developed than for relational systems, making careful physical database design even more imperative for object-oriented database systems.

Quadtrees as a representation for irregularly shaped fields in radiotherapy applications

The choice of data representations in treatment planning software impacts on the accuracy of the treatment plan and the total computation time. We present the quadtree data structure as one such representation for irregularly shaped fields in a three-dimensional electron pencil-beam algorithm. The quadtree subdivides the irregular field into pencil-beams of unequal size and considerably reduces the number of pencil-beams needed in the dose calculation while preserving the calculational accuracy. Small pencil-beams are concentrated around the field edges to provide optimal resolution and computational accuracy in penumbra regions. The quadtree simplifies the evaluation of the pencil-beam fluence at a given depth, and removes systematic errors present in alternative implementations. The quadtree algorithm results in a ten-fold performance improvement when compared to alternative implementations for the irregular field.

Optical Representations For Symbolic Logic

Artificial intelligence problems are solved on electronic computers by techniques that make heavy use of address calculation and dynamic management of data storage space. Optical computing, on the other hand, is normally associated with numerical problems in which the size of the data space is fixed and addressing may be handled in a predictable manner not affected by actual data values. In this paper we categorize expert system problems by the amount of dynamic storage management required and discuss several methods for eliminating unnecessary address manipulation by careful choice of data representation. Major emphasis is placed on the implementation of the mathematical technique of resolution. Various resolution strategies are analyzed, and the effect of these strategies on storage management is assessed with a view to minimizing the complexity of processing. Finally, electro-optical implementations are considered for the major functional elements of an optical resolution system.