Machine Learning and Knowledge Discovery in Databases. Applied Data Science Track European Conference, ECML PKDD 2021, Bilbao, Spain, September 13-17, 2021, Proceedings, Part V

• Main Authors: Dong, Yuxiao, Kourtellis, Nicolas, Hammer, Barbara, Lozano, Jose A
• Format: eBook
• Language: English
• Published: Cham Springer International Publishing AG 2021; Springer International Publishing
• Edition: 1
• Series: Lecture Notes in Computer Science
• Subjects: Data mining-Congresses; Machine learning-Congresses
• ISBN: 9783030865160; 3030865169

The multi-volume set LNAI 12975 until 12979 constitutes the refereed proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2021, which was held during September 13-17, 2021. The conference was originally planned to take place in Bilbao, Spain, but...

Table of Contents:

• Balancing Speed and Accuracy in Neural-Enhanced Phonetic Name Matching -- 1 Introduction -- 1.1 Challenges -- 2 Related Work -- 3 Phonetic Name Matching Systems -- 3.1 Neural Name Transliteration -- 3.2 Neural Name Matching -- 4 Experimental Results -- 4.1 Training and Hyperparameters -- 4.2 Results -- 5 Conclusion and Future Work -- References -- Robust Learning for Text Classification with Multi-source Noise Simulation and Hard Example Mining -- 1 Introduction -- 2 Related Work -- 2.1 Noise Reduction -- 2.2 Adversarial Training -- 2.3 Training with Noisy Data -- 3 Problem -- 3.1 Notation -- 3.2 Text Classification -- 3.3 A Practical Scenario -- 3.4 OCR Noise Simulation -- 3.5 Robust Training -- 4 Approach -- 4.1 OCR Noise Simulation -- 4.2 Noise Invariance Representation -- 4.3 Hard Example Mining -- 4.4 The Overall Framework -- 5 Experiment -- 5.1 Dataset -- 5.2 Implementation -- 5.3 Results -- 6 Analysis -- 6.1 Naive Training with a Single Noise Simulation Method -- 6.2 The Impact of Different Noise Level -- 6.3 The Impact of Hard Example Mining -- 6.4 The Impact of Stability Loss -- 7 Conclusion -- References -- Topic-to-Essay Generation with Comprehensive Knowledge Enhancement -- 1 Introduction -- 2 Related Work -- 3 Methodology -- 3.1 Task Formulation -- 3.2 Model Description -- 3.3 Training and Inference -- 4 Experiments -- 4.1 Datasets -- 4.2 Settings -- 4.3 Baselines -- 4.4 Evaluation Metrics -- 4.5 Experimental Results -- 4.6 Validity of Knowledge Transfer -- 4.7 Case Study -- 5 Conclusion -- References -- Analyzing Research Trends in Inorganic Materials Literature Using NLP -- 1 Introduction -- 2 Related Work -- 3 Corpus Preparation -- 3.1 Definition of Types -- 3.2 Collecting Literature -- 3.3 Annotation -- 4 Approach -- 4.1 Sequence Labeling Architecture -- 4.2 Numeric Normalization -- 5 Results -- 5.1 Inter-Annotator Agreement
• 2 Problem Definition -- 3 Related Work -- 4 Residual Graph-Level Graph Convolutional Networks -- 5 Datasets -- 6 Workflow Similarity -- 7 Structural Performance Prediction -- 8 Component Refinement and Suggestion -- 9 Conclusion -- References -- ConCAD: Contrastive Learning-Based Cross Attention for Sleep Apnea Detection -- 1 Introduction -- 2 Related Work -- 2.1 Sleep Apnea Detection -- 2.2 Attention-Based Feature Fusion -- 2.3 Contrastive Learning -- 3 Methodology -- 3.1 Expert Feature Extraction and Data Augmentation -- 3.2 Feature Extractor -- 3.3 Cross Attention -- 3.4 Contrastive Learning. -- 4 Experiments and Results -- 4.1 Datasets -- 4.2 Compared Methods -- 4.3 Experiment Setup -- 4.4 Results and Discussions -- 5 Conclusions and Future Work -- References -- Machine Learning Based Simulations and Knowledge Discovery -- DeepPE: Emulating Parameterization in Numerical Weather Forecast Model Through Bidirectional Network -- 1 Introduction -- 2 Related Work -- 3 Methods -- 3.1 Problem Definition -- 3.2 Deep Parameterization Emulator -- 3.3 Transfer Scheme -- 3.4 Training -- 4 Experiments -- 4.1 Datasets -- 4.2 Experimental Setup -- 5 Results -- 5.1 DeepPE Performance Analysis -- 5.2 Transfer Analysis -- 6 Conclusion -- References -- Effects of Boundary Conditions in Fully Convolutional Networks for Learning Spatio-Temporal Dynamics -- 1 Introduction -- 2 Method -- 2.1 Learning an Auto-Regressive Model -- 2.2 Neural Network Convolutional Architecture -- 2.3 Boundary Condition Treatment -- 2.4 Loss Function -- 3 Applications: Time-Evolving PDEs -- 3.1 Acoustic Propagation of Gaussian Pulses -- 3.2 Diffusion of Temperature Spots -- 3.3 Datasets Generation and Parameters -- 4 Results -- 5 Conclusion -- References -- Physics Knowledge Discovery via Neural Differential Equation Embedding -- 1 Introduction -- 2 Phase-Field Model -- 3 Problem Statement
• Intro -- Preface -- Organization -- Contents - Part V -- Automating Machine Learning, Optimization, and Feature Engineering -- PuzzleShuffle: Undesirable Feature Learning for Semantic Shift Detection -- 1 Introduction -- 2 Related Work -- 2.1 Out-of-Distribution Detection -- 2.2 Data Augmentation -- 2.3 Uncertainty Calibration -- 3 Preliminaries -- 3.1 The Effects by Perturbation -- 3.2 Adversarial Undesirable Feature Learning -- 4 Proposed Method -- 4.1 PuzzleShuffle Augmentation -- 4.2 Adaptive Label Smoothing -- 4.3 Motivation -- 5 Experiments -- 5.1 Experimental Settings -- 5.2 Compared Methods -- 5.3 Results -- 5.4 Analysis -- 6 Conclusion -- References -- Enabling Machine Learning on the Edge Using SRAM Conserving Efficient Neural Networks Execution Approach -- 1 Introduction -- 2 Background and Related Work -- 2.1 Deep Model Compression -- 2.2 Executing Neural Networks on Microcontrollers -- 3 Efficient Neural Network Execution Approach Design -- 3.1 Tensor Memory Mapping (TMM) Method Design -- 3.2 Loading Fewer Tensors and Tensors Re-usage -- 3.3 Finding the Cheapest NN Graph Execution Sequence -- 3.4 Core Algorithm -- 4 Experimental Evaluation -- 4.1 SRAM Usage -- 4.2 Model Performance -- 4.3 Inference Time and Energy Consumption -- 5 Conclusion -- References -- AutoML Meets Time Series Regression Design and Analysis of the AutoSeries Challenge -- 1 Introduction -- 2 Challenge Setting -- 2.1 Phases -- 2.2 Protocol -- 2.3 Datasets -- 2.4 Metrics -- 2.5 Platform, Hardware and Limitations -- 2.6 Baseline -- 2.7 Results -- 3 Post Challenge Experiments -- 3.1 Reproducibility -- 3.2 Overfitting and Generalisation -- 3.3 Comparison to Open Source AutoML Solutions -- 3.4 Impact of Time Budget -- 3.5 Dataset Difficulty -- 4 Conclusion and Future Work -- References -- Methods for Automatic Machine-Learning Workflow Analysis -- 1 Introduction
• MMNet: Multi-granularity Multi-mode Network for Item-Level Share Rate Prediction -- 1 Introduction -- 2 Related Works -- 3 Preliminary -- 4 Methodology -- 4.1 Overall Framework -- 4.2 Fine-Granularity Module -- 4.3 Coarse-Granularity Module -- 4.4 Meta-info Modeling Module -- 4.5 Optimization Objectives -- 5 Online Deployment -- 6 Experiments -- 6.1 Datasets -- 6.2 Baselines and Experimental Settings -- 6.3 Offline Item-Level Share Rate Prediction -- 6.4 Online A/B Tests -- 6.5 Ablation Studies -- 6.6 Parameter Analyses -- 7 Conclusion and Future Work -- References -- The Joy of Dressing Is an Art: Outfit Generation Using Self-attention Bi-LSTM -- 1 Introduction -- 2 Related Work -- 3 Methodology -- 3.1 Bayesian Personalized Ranking (MF) Embedding -- 3.2 Training Dataset Generation -- 3.3 Bi-LSTM -- 3.4 Self-attention Bi-LSTM -- 3.5 Generation of New Outfits -- 4 Results -- 5 Conclusion -- References -- On Inferring a Meaningful Similarity Metric for Customer Behaviour -- 1 Introduction -- 2 Problem Definition -- 3 SIMPRIM Framework -- 3.1 Journey Log to Journey Profiles -- 3.2 Measuring Similarity -- 3.3 Dimensionality Reduction -- 3.4 Co-learning of Metric Weights and Journey Clustering -- 3.5 Evaluation -- 4 Experimental Evaluation -- 4.1 Customer Service Process at Anonycomm -- 4.2 BPIC 2012 Real Dataset -- 5 Related Work -- 6 Conclusion -- References -- Quantifying Explanations of Neural Networks in E-Commerce Based on LRP -- 1 Introduction -- 2 Preliminaries -- 3 Formal Model of an Online Shop -- 4 Explanation Approach -- 4.1 Explanation via Layer-Wise Relevance Propagation -- 4.2 Input Analysis with Leave-One-Out Method -- 4.3 Explanation Quantity Measures -- 5 Evaluation -- 5.1 Evaluation Setting -- 5.2 Evaluation Data Set -- 5.3 Evaluation Results -- 6 Conclusion -- References -- Natural Language Processing
• 4 Neural Differential Equation Embedding -- 5 Related Work -- 6 Experiments -- 7 Conclusion -- References -- A Bayesian Convolutional Neural Network for Robust Galaxy Ellipticity Regression -- 1 Introduction -- 2 Estimating Galaxy Ellipticity from Images -- 3 A Method to Assess Uncertainty in Ellipticity Estimation -- 3.1 Estimation of Noise Related Uncertainty -- 3.2 Estimation of Blend Related Uncertainty -- 3.3 Training Protocol -- 4 Experiments -- 4.1 Estimation of Uncertainty Related to Noise -- 4.2 Estimation of Uncertainty Related to Blending -- 5 Conclusion -- References -- Precise Weather Parameter Predictions for Target Regions via Neural Networks -- 1 Introduction -- 2 Related Work -- 3 Pertinent Background -- 4 Learning-Based Modelets for Weather Forecasting -- 4.1 Micro Model -- 4.2 Micro-Macro Model -- 5 Experiment -- 5.1 Setting -- 5.2 Overall Performance -- 5.3 Comparing to Other Methods -- 5.4 Ablation Study -- 5.5 Abnormal Weather Forecasting -- 6 Conclusion -- References -- Action Set Based Policy Optimization for Safe Power Grid Management -- 1 Introduction -- 2 Related Work -- 3 Preliminary -- 3.1 Power Grid Management -- 3.2 Search-Based Planning -- 4 Methodology -- 4.1 Search with the Action Set -- 4.2 Policy Optimization -- 4.3 Discussion on Action Set Size -- 4.4 Algorithm Summary -- 5 Experiments -- 5.1 Experiment Setup -- 5.2 Implementation -- 5.3 Competition -- 6 Conclusion -- A Grid2Op Environment -- References -- Conditional Neural Relational Inference for Interacting Systems -- 1 Introduction -- 2 Related Work -- 3 The Conditional Neural Inference Model -- 3.1 Encoding, Establishing the Body-Part Interactions -- 3.2 Decoding, Establishing the Dynamics -- 3.3 Conditional Generation -- 4 Experiments -- 4.1 Experimental Setup -- 4.2 Results -- 5 Conclusion -- References -- Recommender Systems and Behavior Modeling
• 5.2 Comparing Language Models