Lars Linsen

Hennes Rave, Katharina Kronenberg, Hannes Gödde et al.

Hyperspectral bioimaging techniques such as infrared (IR) microscopy and laser ablation-inductively coupled plasma-mass spectrometry (LA-ICP-MS) produce high-dimensional, spatially resolved datasets that require sophisticated analysis to reveal chemically and anatomically meaningful structures. Existing software solutions are typically modality-specific and cover only parts of the analytical workflow, forcing researchers to transfer data across multiple tools and manually reconcile results. We present MIA (Multiscale Image Analysis), a modality-agnostic visual analysis environment that integrates the full exploratory workflow -- from spectral preprocessing and dimensionality reduction to interactive segmentation and spectral similarity analysis -- within a single, tightly coupled interface. MIA supports hierarchical and landmark-based embeddings to handle datasets of varying scale and complexity, interactive and automatic segmentation with a shared state across all linked views, and multimodal analysis of co-registered datasets from different instruments. We demonstrate the effectiveness of MIA through three use cases drawn from real analytical chemistry workflows: (1) the recovery of biologically meaningful tissue compartments through derivative preprocessing and hierarchical embedding, (2) pigment identification via spectral similarity search with spatial overview, and (3) multimodal tissue characterization combining molecular IR and elemental LA-ICP-MS data. Qualitative feedback from domain expert collaborators confirms that MIA reduces the need for tool-switching and supports analytical insights that are difficult to obtain with existing software.

Reyhaneh Sabbagh Gol, Dimitar Valkov, Lars Linsen

Hand kinematics can be measured in Human-Computer Interaction (HCI) with the intention to predict the user's intention in a reach-to-grasp action. Using multiple hand sensors, multivariate time series data are being captured. Given a number of possible actions on a number of objects, the goal is to classify the multivariate time series data, where the class shall be predicted as early as possible. Many machine-learning methods have been developed for such classification tasks, where different approaches produce favorable solutions on different data sets. We, therefore, employ an ensemble approach that includes and weights different approaches. To provide a trustworthy classification production, we present the XMTC tool that incorporates coordinated multiple-view visualizations to analyze the predictions. Temporal accuracy plots, confusion matrix heatmaps, temporal confidence heatmaps, and partial dependence plots allow for the identification of the best trade-off between early prediction and prediction quality, the detection and analysis of challenging classification conditions, and the investigation of the prediction evolution in an overview and detail manner. We employ XMTC to real-world HCI data in multiple scenarios and show that good classification predictions can be achieved early on with our classifier as well as which conditions are easy to distinguish, which multivariate time series measurements impose challenges, and which features have most impact.

Valerie Müller, Christian Sieg, Lars Linsen

Topic modeling is a state-of-the-art technique for analyzing text corpora. It uses a statistical model, most commonly Latent Dirichlet Allocation (LDA), to discover abstract topics that occur in the document collection. However, the LDA-based topic modeling procedure is based on a randomly selected initial configuration as well as a number of parameter values than need to be chosen. This induces uncertainties on the topic modeling results, and visualization methods should convey these uncertainties during the analysis process. We propose a visual uncertainty-aware topic modeling analysis. We capture the uncertainty by computing topic modeling ensembles and propose measures for estimating topic modeling uncertainty from the ensemble. Then, we propose to enhance state-of-the-art topic modeling visualization methods to convey the uncertainty in the topic modeling process. We visualize the entire ensemble of topic modeling results at different levels for topic and document analysis. We apply our visualization methods to a text corpus to document the impact of uncertainty on the analysis.

Karim Huesmann, Luis Garcia Rodriguez, Lars Linsen et al.

Overfitting is one of the fundamental challenges when training convolutional neural networks and is usually identified by a diverging training and test loss. The underlying dynamics of how the flow of activations induce overfitting is however poorly understood. In this study we introduce a perplexity-based sparsity definition to derive and visualise layer-wise activation measures. These novel explainable AI strategies reveal a surprising relationship between activation sparsity and overfitting, namely an increase in sparsity in the feature extraction layers shortly before the test loss starts rising. This tendency is preserved across network architectures and reguralisation strategies so that our measures can be used as a reliable indicator for overfitting while decoupling the network's generalisation capabilities from its loss-based definition. Moreover, our differentiable sparsity formulation can be used to explicitly penalise the emergence of sparsity during training so that the impact of reduced sparsity on overfitting can be studied in real-time. Applying this penalty and analysing activation sparsity for well known regularisers and in common network architectures supports the hypothesis that reduced activation sparsity can effectively improve the generalisation and classification performance. In line with other recent work on this topic, our methods reveal novel insights into the contradicting concepts of activation sparsity and network capacity by demonstrating that dense activations can enable discriminative feature learning while efficiently exploiting the capacity of deep models without suffering from overfitting, even when trained excessively.

Karim Huesmann, Soeren Klemm, Lars Linsen et al.

Overfitting is one of the most common problems when training deep neural networks on comparatively small datasets. Here, we demonstrate that neural network activation sparsity is a reliable indicator for overfitting which we utilize to propose novel targeted sparsity visualization and regularization strategies. Based on these strategies we are able to understand and counteract overfitting caused by activation sparsity and filter correlation in a targeted layer-by-layer manner. Our results demonstrate that targeted sparsity regularization can efficiently be used to regularize well-known datasets and architectures with a significant increase in image classification performance while outperforming both dropout and batch normalization. Ultimately, our study reveals novel insights into the contradicting concepts of activation sparsity and network capacity by demonstrating that targeted sparsity regularization enables salient and discriminative feature learning while exploiting the full capacity of deep models without suffering from overfitting, even when trained excessively.