Proceedings of the Northern Lights Deep Learning Workshop <p>Deep learning is an emerging subfield in machine learning that has in recent years achieved state-of-the-art performance in image classification, object detection, segmentation, time series prediction and speech recognition to name a few. This workshop will gather researchers both on a national and international level to exchange ideas, encourage collaborations and present cutting-edge research.</p> Septentrio Academic Publishing en-US Proceedings of the Northern Lights Deep Learning Workshop 2703-6928 Consistent and accurate estimation of stellar parameters from HARPS-N Spectroscopy using Deep Learning <p>Consistent and accurate estimation of stellar parameters is of great importance for information retrieval in astrophysical research. The parameters span a wide range from effective temperature to rotational velocity. We propose to estimate the stellar parameters directly from spectral signals coming from the HARPS-N spectrograph pipeline before any spectrum-processing steps are applied to extract the 1D spectrum. We propose an attention-based model to estimate the stellar parameters, which estimate both mean and uncertainty of the stellar parameters through estimation of the parameters of a Gaussian distribution. The estimated distributions create a basis to generate data-driven Gaussian confidence intervals for the estimated stellar parameters. We show that residual networks and attention-based models can estimate the stellar parameters with high accuracy for low Signal-to-noise ratio (SNR) compared to previous methods. With an observation of the Sun from the HARPS-N spectrograph, we show that the models can estimate stellar parameters from real observational data.</p> Frederik Boe Hüttel Line Katrine Harder Clemmensen Copyright (c) 2021 Frederik Boe Hüttel, Line Katrine Harder Clemmensen 2021-04-19 2021-04-19 2 10.7557/18.5693 Reducing Objective Function Mismatch in Deep Clustering with the Unsupervised Companion Objective <p>Preservation of local similarity structure is a key challenge in deep clustering. Many recent deep clustering methods therefore use autoencoders to help guide the model's neural network towards an embedding which is more reflective of the input space geometry. However, recent work has shown that autoencoder-based deep clustering models can suffer from objective function mismatch (OFM). In order to improve the preservation of local similarity structure, while simultaneously having a low OFM, we develop a new auxiliary objective function for deep clustering. Our Unsupervised Companion Objective (UCO) encourages a consistent clustering structure at intermediate layers in the network -- helping the network learn an embedding which is more reflective of the similarity structure in the input space. Since a clustering-based auxiliary objective has the same goal as the main clustering objective, it is less prone to introduce objective function mismatch between itself and the main objective. Our experiments show that attaching the UCO to a deep clustering model improves the performance of the model, and exhibits a lower OFM, compared to an analogous autoencoder-based model.</p> Daniel J. Trosten Robert Jenssen Michael C. Kampffmeyer Copyright (c) 2021 Daniel J. Trosten, Robert Jenssen, Michael C. Kampffmeyer 2021-04-19 2021-04-19 2 10.7557/18.5709 Robust Deep Interpretable Features for Binary Image Classifi cation <p>The problem of interpretability for binary image classification is considered through the lens of kernel two-sample tests and generative modeling. A feature extraction framework coined Deep Interpretable Features (DIF) is developed, which is used in combination with IntroVAE, a generative model capable of high-resolution image synthesis. Experimental results on a variety of datasets, including COVID-19 chest x-rays demonstrate the benefits of combining deep generative models with the ideas from kernel-based hypothesis testing in moving towards more robust interpretable deep generative models.</p> Robert Hu Dino Sejdinovic Copyright (c) 2021 Robert Hu, Dino Sejdinovic 2021-04-19 2021-04-19 2 10.7557/18.5708 Seafloor Pipeline Detection With Deep Learning <p>This paper presents fast, accurate, and automatic methods for detecting seafloor pipelines in multibeam echo sounder data with deep learning. The proposed methods take inspiration from the highly successful ResNet and YOLO deep learning models and tailor them to the idiosyncrasies of the seafloor pipeline detection task.</p> <p>We use the area between lines and Hausdorff line distance functions to accurately evaluate how well methods can localize (pipe)lines. The same functions also show promise as loss functions compared to standard mean squared error, which does not include the regression variables' geometrical interpretation.</p> <p>The model outperforms the highest likelihood baseline by more than 35% on a region-wise F1-score classification evaluation while being more than eight times more accurate than the baseline in locating pipelines. It is efficient, operating at over eighteen 32-ping image segments per second, which is far beyond real-time requirements.</p> Vemund Sigmundson Schøyen Narada Dilp Warakagoda Øivind Midtgaard Copyright (c) 2021 Vemund Sigmundson Schøyen, Narada Dilp Warakagoda, Øivind Midtgaard 2021-04-19 2021-04-19 2 10.7557/18.5699 A Tomographic Reconstruction Method using Coordinate-based Neural Network with Spatial Regularization <p>Tomographic reconstruction is concerned with computing the cross-sections of an object from a finite number of projections. Many conventional methods represent the cross-sections as images on a regular grid. In this paper, we study a recent coordinate-based neural network for tomographic reconstruction, where the network inputs a spatial coordinate and outputs the attenuation coefficient on the coordinate. This coordinate-based network allows the continuous representation of an object. Based on this network, we propose a spatial regularization term, to obtain a high-quality reconstruction. Experimental results on synthetic data show that the regularization term improves the reconstruction quality significantly, compared to the baseline. We also provide an ablation study for different architecture configurations and hyper-parameters.</p> Jakeoung Koo Elise Otterlei Brenne Anders Bjorholm Dahl Vedrana Andersen Dahl Copyright (c) 2021 Jakeoung Koo, Elise Otterlei Brenne, Anders Bjorholm Dahl, Vedrana Andersen Dahl 2021-04-19 2021-04-19 2 10.7557/18.5676