arXiv subjects

Thomas Ortner

Publications and source records attributed to Thomas Ortner.

15 recordsLinked to original sources

arXiv2026

Task-driven Processing with Coarse-to-Fine Glimpse-based Active Perception

State-of-the-art vision models process images in their entirety, lacking the ability to selectively zoom in on relevant regions. This limitation is particularly acute in scenarios where processing must be conditioned on a specific task - such as instance detection, which requires localizing a specific object in a high-resolution, cluttered scene. In such settings, critical details are easily lost as images are often resized to match the model dimensions and computational constraints. We introduce Coarse-to-Fine Glimpse-based Active Perception (CF-GAP), a task-driven front-end that enhances high-resolution processing of existing instance detectors. CF-GAP selectively directs a sequence of limited view glimpses across the scene, utilizing task information to iteratively refine focus on the most relevant regions. These localized regions are then processed at high resolution by a downstream instance detector. By avoiding full-image processing and eliminating irrelevant confounding information, CF-GAP improves Average Precision (AP) by up to 20% across various state-of-the-art instance detectors on the HR-InsDet and Robotools benchmarks, while further enabling lightweight detectors to outperform their larger counterparts.

cs.CV↗

arXiv2025

FlowState: Sampling-Rate-Equivariant Time-Series Forecasting

Existing time series foundation models (TSFMs), often based on transformer variants, lack adaptability to different sampling rates, struggle with generalization across varying context and target lengths, and are computationally inefficient. We introduce FlowState, a novel TSFM architecture that achieves sampling-rate-equivariant forecasting through a unified design that pairs a state space model (SSM) encoder with a functional basis decoder (FBD). This design enables continuous-time modeling and dynamic time-scale adjustment, allowing FlowState to inherently generalize across all possible temporal resolutions, and dynamically adjust the forecasting horizons without retraining. We further propose an efficient pretraining strategy that improves robustness and accelerates training. Despite being one of the smallest TSFMs, FlowState achieves state-of-the-art results on the widely used GIFT-Eval benchmark, while demonstrating superior adaptability to unseen sampling rates. Our detailed analyses confirm the effectiveness of its components, and we demonstrate its unique ability to adapt to varying input sampling rates.

cs.LG↗

arXiv2024

Mind the GAP: Glimpse-based Active Perception improves generalization and sample efficiency of visual reasoning

Human capabilities in understanding visual relations are far superior to those of AI systems, especially for previously unseen objects. For example, while AI systems struggle to determine whether two such objects are visually the same or different, humans can do so with ease. Active vision theories postulate that the learning of visual relations is grounded in actions that we take to fixate objects and their parts by moving our eyes. In particular, the low-dimensional spatial information about the corresponding eye movements is hypothesized to facilitate the representation of relations between different image parts. Inspired by these theories, we develop a system equipped with a novel Glimpse-based Active Perception (GAP) that sequentially glimpses at the most salient regions of the input image and processes them at high resolution. Importantly, our system leverages the locations stemming from the glimpsing actions, along with the visual content around them, to represent relations between different parts of the image. The results suggest that the GAP is essential for extracting visual relations that go beyond the immediate visual content. Our approach reaches state-of-the-art performance on several visual reasoning tasks being more sample-efficient, and generalizing better to out-of-distribution visual inputs than prior models.

cs.CV↗

arXiv2024

BEMTrace: Visualization-driven approach for deriving Building Energy Models from BIM

Building Information Modeling (BIM) describes a central data pool covering the entire life cycle of a construction project. Similarly, Building Energy Modeling (BEM) describes the process of using a 3D representation of a building as a basis for thermal simulations to assess the building's energy performance. This paper explores the intersection of BIM and BEM, focusing on the challenges and methodologies in converting BIM data into BEM representations for energy performance analysis. BEMTrace integrates 3D data wrangling techniques with visualization methodologies to enhance the accuracy and traceability of the BIM-to-BEM conversion process. Through parsing, error detection, and algorithmic correction of BIM data, our methods generate valid BEM models suitable for energy simulation. Visualization techniques provide transparent insights into the conversion process, aiding error identification, validation, and user comprehension. We introduce context-adaptive selections to facilitate user interaction and to show that the BEMTrace workflow helps users understand complex 3D data wrangling processes.

cs.HC↗

arXiv2024

The State of the Art in Visual Analytics for 3D Urban Data

Urbanization has amplified the importance of three-dimensional structures in urban environments for a wide range of phenomena that are of significant interest to diverse stakeholders. With the growing availability of 3D urban data, numerous studies have focused on developing visual analysis techniques tailored to the unique characteristics of urban environments. However, incorporating the third dimension into visual analytics introduces additional challenges in designing effective visual tools to tackle urban data's diverse complexities. In this paper, we present a survey on visual analytics of 3D urban data. Our work characterizes published works along three main dimensions (why, what, and how), considering use cases, analysis tasks, data, visualizations, and interactions. We provide a fine-grained categorization of published works from visualization journals and conferences, as well as from a myriad of urban domains, including urban planning, architecture, and engineering. By incorporating perspectives from both urban and visualization experts, we identify literature gaps, motivate visualization researchers to understand challenges and opportunities, and indicate future research directions.

cs.HC↗

arXiv2024

Learning-to-learn enables rapid learning with phase-change memory-based in-memory computing

There is a growing demand for low-power, autonomously learning artificial intelligence (AI) systems that can be applied at the edge and rapidly adapt to the specific situation at deployment site. However, current AI models struggle in such scenarios, often requiring extensive fine-tuning, computational resources, and data. In contrast, humans can effortlessly adjust to new tasks by transferring knowledge from related ones. The concept of learning-to-learn (L2L) mimics this process and enables AI models to rapidly adapt with only little computational effort and data. In-memory computing neuromorphic hardware (NMHW) is inspired by the brain's operating principles and mimics its physical co-location of memory and compute. In this work, we pair L2L with in-memory computing NMHW based on phase-change memory devices to build efficient AI models that can rapidly adapt to new tasks. We demonstrate the versatility of our approach in two scenarios: a convolutional neural network performing image classification and a biologically-inspired spiking neural network generating motor commands for a real robotic arm. Both models rapidly learn with few parameter updates. Deployed on the NMHW, they perform on-par with their software equivalents. Moreover, meta-training of these models can be performed in software with high-precision, alleviating the need for accurate hardware models.

cs.NE↗

arXiv2023

Intrinsic Biologically Plausible Adversarial Robustness

Artificial Neural Networks (ANNs) trained with Backpropagation (BP) excel in different daily tasks but have a dangerous vulnerability: inputs with small targeted perturbations, also known as adversarial samples, can drastically disrupt their performance. Adversarial training, a technique in which the training dataset is augmented with exemplary adversarial samples, is proven to mitigate this problem but comes at a high computational cost. In contrast to ANNs, humans are not susceptible to misclassifying these same adversarial samples. Thus, one can postulate that biologically-plausible trained ANNs might be more robust against adversarial attacks. In this work, we chose the biologically-plausible learning algorithm Present the Error to Perturb the Input To modulate Activity (PEPITA) as a case study and investigated this question through a comparative analysis with BP-trained ANNs on various computer vision tasks. We observe that PEPITA has a higher intrinsic adversarial robustness and, when adversarially trained, also has a more favorable natural-vs-adversarial performance trade-off. In particular, for the same natural accuracies on the MNIST task, PEPITA's adversarial accuracies decrease on average only by 0.26% while BP's decrease by 8.05%.

cs.LG↗

arXiv2023

Online Spatio-Temporal Learning with Target Projection

Recurrent neural networks trained with the backpropagation through time (BPTT) algorithm have led to astounding successes in various temporal tasks. However, BPTT introduces severe limitations, such as the requirement to propagate information backwards through time, the weight symmetry requirement, as well as update-locking in space and time. These problems become roadblocks for AI systems where online training capabilities are vital. Recently, researchers have developed biologically-inspired training algorithms, addressing a subset of those problems. In this work, we propose a novel learning algorithm called online spatio-temporal learning with target projection (OSTTP) that resolves all aforementioned issues of BPTT. In particular, OSTTP equips a network with the capability to simultaneously process and learn from new incoming data, alleviating the weight symmetry and update-locking problems. We evaluate OSTTP on two temporal tasks, showcasing competitive performance compared to BPTT. Moreover, we present a proof-of-concept implementation of OSTTP on a memristive neuromorphic hardware system, demonstrating its versatility and applicability to resource-constrained AI devices.

cs.NE↗

arXiv2023

Feature-assisted interactive geometry reconstruction in 3D point clouds using incremental region growing

Reconstructing geometric shapes from point clouds is a common task that is often accomplished by experts manually modeling geometries in CAD-capable software. State-of-the-art workflows based on fully automatic geometry extraction are limited by point cloud density and memory constraints, and require pre- and post-processing by the user. In this work, we present a framework for interactive, user-driven, feature-assisted geometry reconstruction from arbitrarily sized point clouds. Based on seeded region-growing point cloud segmentation, the user interactively extracts planar pieces of geometry and utilizes contextual suggestions to point out plane surfaces, normal and tangential directions, and edges and corners. We implement a set of feature-assisted tools for high-precision modeling tasks in architecture and urban surveying scenarios, enabling instant-feedback interactive point cloud manipulation on large-scale data collected from real-world building interiors and facades. We evaluate our results through systematic measurement of the reconstruction accuracy, and interviews with domain experts who deploy our framework in a commercial setting and give both structured and subjective feedback.

cs.GR↗

arXiv2020

InCorr: Interactive Data-Driven Correlation Panels for Digital Outcrop Analysis

Geological analysis of 3D Digital Outcrop Models (DOMs) for reconstruction of ancient habitable environments is a key aspect of the upcoming ESA ExoMars 2022 Rosalind Franklin Rover and the NASA 2020 Rover Perseverance missions in seeking signs of past life on Mars. Geologists measure and interpret 3D DOMs, create sedimentary logs and combine them in `correlation panels' to map the extents of key geological horizons, and build a stratigraphic model to understand their position in the ancient landscape. Currently, the creation of correlation panels is completely manual and therefore time-consuming, and inflexible. With InCorr we present a visualization solution that encompasses a 3D logging tool and an interactive data-driven correlation panel that evolves with the stratigraphic analysis. For the creation of InCorr we closely cooperated with leading planetary geologists in the form of a design study. We verify our results by recreating an existing correlation analysis with InCorr and validate our correlation panel against a manually created illustration. Further, we conducted a user-study with a wider circle of geologists. Our evaluation shows that InCorr efficiently supports the domain experts in tackling their research questions and that it has the potential to significantly impact how geologists work with digital outcrop representations in general.

cs.GR↗

arXiv2017

Clustering of imbalanced high-dimensional media data

Media content in large repositories usually exhibits multiple groups of strongly varying sizes. Media of potential interest often form notably smaller groups. Such media groups differ so much from the remaining data that it may be worthy to look at them in more detail. In contrast, media with popular content appear in larger groups. Identifying groups of varying sizes is addressed by clustering of imbalanced data. Clustering highly imbalanced media groups is additionally challenged by the high dimensionality of the underlying features. In this paper, we present the Imbalanced Clustering (IClust) algorithm designed to reveal group structures in high-dimensional media data. IClust employs an existing clustering method in order to find an initial set of a large number of potentially highly pure clusters which are then successively merged. The main advantage of IClust is that the number of clusters does not have to be pre-specified and that no specific assumptions about the cluster or data characteristics need to be made. Experiments on real-world media data demonstrate that in comparison to existing methods, IClust is able to better identify media groups, especially groups of small sizes.

stat.ME↗

arXiv2017

Robust and sparse k-means clustering for high-dimensional data

In real-world application scenarios, the identification of groups poses a significant challenge due to possibly occurring outliers and existing noise variables. Therefore, there is a need for a clustering method which is capable of revealing the group structure in data containing both outliers and noise variables without any pre-knowledge. In this paper, we propose a $k$-means-based algorithm incorporating a weighting function which leads to an automatic weight assignment for each observation. In order to cope with noise variables, a lasso-type penalty is used in an objective function adjusted by observation weights. We finally introduce a framework for selecting both the number of clusters and variables based on a modified gap statistic. The conducted experiments on simulated and real-world data demonstrate the advantage of the method to identify groups, outliers, and informative variables simultaneously.

stat.ME↗

arXiv2017

Multigroup discrimination based on weighted local projections

A novel approach for supervised classification analysis for high dimensional and flat data (more variables than observations) is proposed. We use the information of class-membership of observations to determine groups of observations locally describing the group structure. By projecting the data on the subspace spanned by those groups, local projections are defined based on the projection concepts from Ortner et al. (2017a) and Ortner et al. (2017b). For each local projection a local discriminant analysis (LDA) model is computed using the information within the projection space as well as the distance to the projection space. The models provide information about the quality of separation for each class combination. Based on this information, weights are defined for aggregating the LDA-based posterior probabilities of each subspace to a new overall probability. The same weights are used for classifying new observations. In addition to the provided methodology, implemented in the R-package lop, a method of visualizing the connectivity of groups in high-dimensional spaces is proposed on the basis of the posterior probabilities. A thorough evaluation is performed using three different real-world datasets, underlining the strengths of local projection based classification and the provided visualization methodology.

stat.ME↗

arXiv2017

Local projections for high-dimensional outlier detection

In this paper, we propose a novel approach for outlier detection, called local projections, which is based on concepts of Local Outlier Factor (LOF) (Breunig et al., 2000) and RobPCA (Hubert et al., 2005). By using aspects of both methods, our algorithm is robust towards noise variables and is capable of performing outlier detection in multi-group situations. We are further not reliant on a specific underlying data distribution. For each observation of a dataset, we identify a local group of dense nearby observations, which we call a core, based on a modification of the k-nearest neighbours algorithm. By projecting the dataset onto the space spanned by those observations, two aspects are revealed. First, we can analyze the distance from an observation to the center of the core within the projection space in order to provide a measure of quality of description of the observation by the projection. Second, we consider the distance of the observation to the projection space in order to assess the suitability of the core for describing the outlyingness of the observation. These novel interpretations lead to a univariate measure of outlyingness based on aggregations over all local projections, which outperforms LOF and RobPCA as well as other popular methods like PCOut (Filzmoser et al., 2008) and subspace-based outlier detection (Kriegel et al., 2009) in our simulation setups. Experiments in the context of real-word applications employing datasets of various dimensionality demonstrate the advantages of local projections.

stat.ME↗

arXiv2017

Guided projections for analysing the structure of high-dimensional data

A powerful data transformation method named guided projections is proposed creating new possibilities to reveal the group structure of high-dimensional data in the presence of noise variables. Utilising projections onto a space spanned by a selection of a small number of observations allows measuring the similarity of other observations to the selection based on orthogonal and score distances. Observations are iteratively exchanged from the selection creating a non-random sequence of projections which we call guided projections. In contrast to conventional projection pursuit methods, which typically identify a low-dimensional projection revealing some interesting features contained in the data, guided projections generate a series of projections that serve as a basis not just for diagnostic plots but to directly investigate the group structure in data. Based on simulated data we identify the strengths and limitations of guided projections in comparison to commonly employed data transformation methods. We further show the relevance of the transformation by applying it to real-world data sets.

stat.ME↗