Machine Learning Methods in Visualisation for Big Data

Permanent URI for this community

https://diglib.eg.org/handle/10.2312/2632402

Browse

Now showing 1 - 11 of 11

Controllably Sparse Perturbations of Robust Classifiers for Explaining Predictions and Probing Learned Concepts
(The Eurographics Association, 2021) Roberts, Jay; Tsiligkaridis, Theodoros; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
Explaining the predictions of a deep neural network (DNN) in image classification is an active area of research. Many methods focus on localizing pixels, or groups of pixels, which maximize a relevance metric for the prediction. Others aim at creating local "proxy" explainers which aim to account for an individual prediction of a model. We aim to explore "why" a model made a prediction by perturbing inputs to robust classifiers and interpreting the semantically meaningful results. For such an explanation to be useful for humans it is desirable for it to be sparse; however, generating sparse perturbations can computationally expensive and infeasible on high resolution data. Here we introduce controllably sparse explanations that can be efficiently generated on higher resolution data to provide improved counter-factual explanations. Further we use these controllably sparse explanations to probe what the robust classifier has learned. These explanations could provide insight for model developers as well as assist in detecting dataset bias.
Interactive Dense Pixel Visualizations for Time Series and Model Attribution Explanations
(The Eurographics Association, 2023) Schlegel, Udo; Keim, Daniel; Archambault, Daniel; Nabney, Ian; Peltonen, Jaakko
The field of Explainable Artificial Intelligence (XAI) for Deep Neural Network models develops significantly, offering numerous techniques to extract explanations from models. However, evaluating explanations is often not trivial, and differences in applied metrics can be subtle, especially with non-intelligible data. Thus, there is a need for visualizations tailored to explore explanations for domains with such data, e.g., time series. We propose DAVOTS, an interactive visual analytics approach to explore raw time series data, activations of neural networks, and attributions in a dense-pixel visualization to gain insights into the data, models' decisions, and explanations. To further support users in exploring large datasets, we apply clustering approaches to the visualized data domains to highlight groups and present ordering strategies for individual and combined data exploration to facilitate finding patterns. We visualize a CNN trained on the FordA dataset to demonstrate the approach.
Interpreting Black-Box Semantic Segmentation Models in Remote Sensing Applications
(The Eurographics Association, 2019) Janik, Adrianna; Sankaran, Kris; Ortiz, Anthony; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
In the interpretability literature, attention is focused on understanding black-box classifiers, but many problems ranging from medicine through agriculture and crisis response in humanitarian aid are tackled by semantic segmentation models. The absence of interpretability for these canonical problems in computer vision motivates this study. In this study we present a usercentric approach that blends techniques from interpretability, representation learning, and interactive visualization. It allows to visualize and link latent representation to real data instances as well as qualitatively assess strength of predictions. We have applied our method to a deep learning model for semantic segmentation, U-Net, in a remote sensing application of building detection. This application is of high interest for humanitarian crisis response teams that rely on satellite images analysis. Preliminary results shows utility in understanding semantic segmentation models, demo presenting the idea is available online.
ModelSpeX: Model Specification Using Explainable Artificial Intelligence Methods
(The Eurographics Association, 2020) Schlegel, Udo; Cakmak, Eren; Keim, Daniel A.; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
Explainable artificial intelligence (XAI) methods aim to reveal the non-transparent decision-making mechanisms of black-box models. The evaluation of insight generated by such XAI methods remains challenging as the applied techniques depend on many factors (e.g., parameters and human interpretation). We propose ModelSpeX, a visual analytics workflow to interactively extract human-centered rule-sets to generate model specifications from black-box models (e.g., neural networks). The workflow enables to reason about the underlying problem, to extract decision rule sets, and to evaluate the suitability of the model for a particular task. An exemplary usage scenario walks an analyst trough the steps of the workflow to show the applicability.
On KDE-based Brushing in Scatterplots and how it Compares to CNN-based Brushing
(The Eurographics Association, 2019) Fan, Chaoran; Hauser, Helwig; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
In this paper, we investigate to which degree the human should be involved into the model design and how good the empirical model can be with more careful design. To find out, we extended our previously published Mahalanobis brush (the best current empirical model in terms of accuracy for brushing points in a scatterplot) by further incorporating the data distribution information that is captured by the kernel density estimation (KDE). Based on this work, we then include a short discussion between the empirical model, designed in detail by an expert and the deep learning-based model that is learned from user data directly.
Revealing Multimodality in Ensemble Weather Prediction
(The Eurographics Association, 2021) Galmiche, Natacha; Hauser, Helwig; Spengler, Thomas; Spensberger, Clemens; Brun, Morten; Blaser, Nello; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
Ensemble methods are widely used to simulate complex non-linear systems and to estimate forecast uncertainty. However, visualizing and analyzing ensemble data is challenging, in particular when multimodality arises, i.e., distinct likely outcomes. We propose a graph-based approach that explores multimodality in univariate ensemble data from weather prediction. Our solution utilizes clustering and a novel concept of life span associated with each cluster. We applied our method to historical predictions of extreme weather events and illustrate that our method aids the understanding of the respective ensemble forecasts.
Saliency Clouds: Visual Analysis of Point Cloud-oriented Deep Neural Networks in DeepRL for Particle Physics
(The Eurographics Association, 2022) Mulawade, Raju Ningappa; Garth, Christoph; Wiebel, Alexander; Archambault, Daniel; Nabney, Ian; Peltonen, Jaakko
We develop and describe saliency clouds, that is, visualization methods employing explainable AI methods to analyze and interpret deep reinforcement learning (DeepRL) agents working on point cloud-based data. The agent in our application case is tasked to track particles in high energy physics and is still under development. The point clouds contain properties of particle hits on layers of a detector as the input to reconstruct the trajectories of the particles. Through visualization of the influence of different points, their possible connections in an implicit graph, and other features on the decisions of the policy network of the DeepRL agent, we aim to explain the decision making of the agent in tracking particles and thus support its development. In particular, we adapt gradient-based saliency mapping methods to work on these point clouds. We show how the properties of the methods, which were developed for image data, translate to the structurally different point cloud data. Finally, we present visual representations of saliency clouds supporting visual analysis and interpretation of the RL agent's policy network.
ViNNPruner: Visual Interactive Pruning for Deep Learning
(The Eurographics Association, 2022) Schlegel, Udo; Schiegg, Samuel; Keim, Daniel A.; Archambault, Daniel; Nabney, Ian; Peltonen, Jaakko
Neural networks grow vastly in size to tackle more sophisticated tasks. In many cases, such large networks are not deployable on particular hardware and need to be reduced in size. Pruning techniques help to shrink deep neural networks to smaller sizes by only decreasing their performance as little as possible. However, such pruning algorithms are often hard to understand by applying them and do not include domain knowledge which can potentially be bad for user goals. We propose ViNNPruner, a visual interactive pruning application that implements state-of-the-art pruning algorithms and the option for users to do manual pruning based on their knowledge. We show how the application facilitates gaining insights into automatic pruning algorithms and semi-automatically pruning oversized networks to make them more efficient using interactive visualizations.
Visual Analysis of Multivariate Urban Traffic Data Resorting to Local Principal Curves
(The Eurographics Association, 2019) Silva, Carla; d'Orey, Pedro; Aguiar, Ana; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
Traffic congestion causes major economic, environmental and social problems in modern cities. We present an interactive visualization tool to assist domain experts on the identification and analysis of traffic patterns at a city scale making use of multivariate empirical urban data and fundamental diagrams. The proposed method combines visualization techniques with an improved local principle curves method to model traffic dynamics and facilitate comparison of traffic patterns - resorting to the fitted curve with a confidence interval - between different road segments and for different external conditions. We demonstrate the proposed technique in an illustrative real-world case study in the city of Porto, Portugal.
Visual Analysis of the Impact of Neural Network Hyper-Parameters
(The Eurographics Association, 2020) Jönsson, Daniel; Eilertsen, Gabriel; Shi, Hezi; Zheng, Jianmin; Ynnerman, Anders; Unger, Jonas; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
We present an analysis of the impact of hyper-parameters for an ensemble of neural networks using tailored visualization techniques to understand the complicated relationship between hyper-parameters and model performance. The high-dimensional error surface spanned by the wide range of hyper-parameters used to specify and optimize neural networks is difficult to characterize - it is non-convex and discontinuous, and there could be complex local dependencies between hyper-parameters. To explore these dependencies, we make use of a large number of sampled relations between hyper-parameters and end performance, retrieved from thousands of individually trained convolutional neural network classifiers. We use a structured selection of visualization techniques to analyze the impact of different combinations of hyper-parameters. The results reveal how complicated dependencies between hyper-parameters influence the end performance, demonstrating how the complete picture painted by considering a large number of trainings simultaneously can aid in understanding the impact of hyper-parameter combinations.
Visual Interpretation of DNN-based Acoustic Models using Deep Autoencoders
(The Eurographics Association, 2020) Grósz, Tamás; Kurimo, Mikko; Archambault, Daniel and Nabney, Ian and Peltonen, Jaakko
In the past few years, Deep Neural Networks (DNN) have become the state-of-the-art solution in several areas, including automatic speech recognition (ASR), unfortunately, they are generally viewed as black boxes. Recently, this started to change as researchers have dedicated much effort into interpreting their behavior. In this work, we concentrate on visual interpretation by depicting the hidden activation vectors of the DNN, and propose the usage of deep Autoencoders (DAE) to transform these hidden representations for inspection. We use multiple metrics to compare our approach with other, widely-used algorithms and the results show that our approach is quite competitive. The main advantage of using Autoencoders over the existing ones is that after the training phase, it applies a fixed transformation that can be used to visualize any hidden activation vector without any further optimization, which is not true for the other methods.

Browse

Browsing Machine Learning Methods in Visualisation for Big Data by Subject "Computing methodologies"

Results Per Page

Sort Options