Machine Learning and Knowledge Discovery in Databases. Research Track : European Conference, ECML PKDD 2024, Vilnius, Lithuania, September 9–13, 2024, Proceedings, Part I / edited by Albert Bifet, Jesse Davis, Tomas Krilavičius, Meelis Kull, Eirini Ntoutsi, Indrė Žliobaitė.

Format:

Book

Author/Creator:

Bifet, Albert.

Contributor:

Davis, Jesse.

Krilavičius, Tomas.

Kull, Meelis.

Ntoutsi, Eirini.

Žliobaitė, Indrė.

Series:

Lecture Notes in Artificial Intelligence, 2945-9141 ; 14941

Language:

English

Subjects (All):

Artificial intelligence.

Computer engineering.

Computer networks.

Computers.

Image processing--Digital techniques.

Image processing.

Computer vision.

Software engineering.

Artificial Intelligence.

Computer Engineering and Networks.

Computing Milieux.

Computer Imaging, Vision, Pattern Recognition and Graphics.

Software Engineering.

Local Subjects:

Artificial Intelligence.

Computer Engineering and Networks.

Computing Milieux.

Computer Imaging, Vision, Pattern Recognition and Graphics.

Software Engineering.

Physical Description:

1 online resource (514 pages)

Edition:

1st ed. 2024.

Place of Publication:

Cham : Springer Nature Switzerland : Imprint: Springer, 2024.

Summary:

This multi-volume set, LNAI 14941 to LNAI 14950, constitutes the refereed proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2024, held in Vilnius, Lithuania, in September 2024. The papers presented in these proceedings are from the following three conference tracks: - Research Track: The 202 full papers presented here, from this track, were carefully reviewed and selected from 826 submissions. These papers are present in the following volumes: Part I, II, III, IV, V, VI, VII, VIII. Demo Track: The 14 papers presented here, from this track, were selected from 30 submissions. These papers are present in the following volume: Part VIII. Applied Data Science Track: The 56 full papers presented here, from this track, were carefully reviewed and selected from 224 submissions. These papers are present in the following volumes: Part IX and Part X.

Contents:

Intro

Preface

Organization

Invited Talks Abstracts

The Dynamics of Memorization and Unlearning

The Emerging Science of Benchmarks

Enhancing User Experience with AI-Powered Search and Recommendations at Spotify

How to Utilize (and Generate) Player Tracking Data in Sport

Resource-Aware Machine Learning-A User-Oriented Approach

Contents - Part I

Research Track

Adaptive Sparsity Level During Training for Efficient Time Series Forecasting with Transformers

1 Introduction

2 Background

2.1 Sparse Neural Networks

2.2 Time Series Forecasting

2.3 Problem Formulation and Notations

3 Analyzing Sparsity Effect in Transformers for Time Series Forecasting

4 Proposed Methodology: PALS

5 Experiments and Results

5.1 Experimental Settings

5.2 Results

6 Discussion

6.1 Performance Comparison with Pruning and Sparse Training Algorithms

6.2 Hyperparameter Sensitivity

7 Conclusions

References

RumorMixer: Exploring Echo Chamber Effect and Platform Heterogeneity for Rumor Detection

2 Related Works

3 Methodology

3.1 Overview

3.2 Echo Chamber Extraction and Representation Learning

3.3 Neural Architecture Search for Platform Heterogeneity

4 Experiments

4.1 Experimental Setting

4.2 Performance Comparison (RQ1)

4.3 Ablation Study (RQ2)

4.4 Parameter Analysis (RQ3)

4.5 Early Rumor Detection (RQ4)

5 Conclusion

Diversified Ensemble of Independent Sub-networks for Robust Self-supervised Representation Learning

2 Related Work

3 Method

3.1 Robust Self-supervised Learning via Independent Sub-networks

3.2 Empirical Analysis of Diversity

3.3 Computational Cost and Efficiency Analysis

4 Experimental Setup

5 Results and Discussion

6 Ablation Study

7 Conclusion

References.

Modular Debiasing of Latent User Representations in Prototype-Based Recommender Systems

4 Experiment Setup

5 Results and Analysis

6 Conclusion and Future Directions

A Mathematics Framework of Artificial Shifted Population Risk and Its Further Understanding Related to Consistency Regularization

3.1 Revisiting Data Augmentation with Empirical Risk

3.2 The Augmented Neighborhood

3.3 The Artificial Shifted Population Risk

3.4 Understanding the Decomposition of Shifted Population Risk

4 Experiment

4.1 Experiment Implementation

4.2 Experimental Results

5 Conclusion and Discussion

Attention-Driven Dropout: A Simple Method to Improve Self-supervised Contrastive Sentence Embeddings

2 Background and Related Work

3.1 Attention Rollout Aggregation

3.2 Static Dropout Rate

3.3 Dynamic Dropout Rate

4.1 Datasets and Tasks

4.2 Training Procedure

5 Result and Discussion

5.1 Ablation Study

6 Conclusion

AEMLO: AutoEncoder-Guided Multi-label Oversampling

1.1 Research Goal

1.2 Motivation

1.3 Summary

2.1 Multi-label Classification

2.2 Multi-label Imbalance Learning

2.3 Deep Sampling Method

3 Multi-label AutoEncoder Oversampling

3.1 Method Description and Overview

3.2 Loss Function

3.3 Generate Instances and Post-processing

4 Experiments and Analysis

4.1 Datasets

4.2 Experiment Setup

4.3 Experimental Analysis

4.4 Parameter Analysis

4.5 Sampling Time

MANTRA: Temporal Betweenness Centrality Approximation Through Sampling

3 Preliminaries.

4 MANTRA: Temporal Betweenness Centrality Approxi-mation Through Sampling

4.1 Temporal Betweenness Estimator

4.2 Sample Complexity Bounds

4.3 Fast Approximation of the Characteristic Quantities

4.4 The MANTRA Framework

5 Experimental Evaluation

5.1 Experimental Setting

5.2 Networks

5.3 Experimental Results

6 Conclusions

Dimensionality-Induced Information Loss of Outliers in Deep Neural Networks

2 Problem Setting and Related Work

2.1 Stable Rank of the Matrix

2.2 Feature-Based Detection

2.3 Projection-Based Detection

2.4 Similarity of DNN Representations

2.5 Noise Sensitivity in the DNN

3 Results

3.1 Overview of the Experiments and a Possible Picture

3.2 Observation of Dimensionality via Stable Ranks

3.3 Transition of OOD Detection Performance

3.4 Block Structure of CKA

3.5 Instability of OOD Samples to Noise Injection

3.6 Dataset Bias-Induced Imbalanced Inference

3.7 Quantitative Comparison of OOD Detection Performance

4 Discussion

5 Summary and Conclusion

Towards Open-World Cross-Domain Sequential Recommendation: A Model-Agnostic Contrastive Denoising Approach

2 Methodology

2.1 Problem Formulation

2.2 Embedding Encoder

2.3 Denoising Interest-Aware Network

2.4 Fusion Gate Unit

2.5 Model Training

2.6 Inductive Representation Generator

3 Experiments

3.1 Datasets

3.2 Experiment Setting

3.3 Performance Comparisons (RQ1)

3.4 Ablation Study (RQ2)

3.5 Online Evaluation (RQ3)

3.6 Model Analyses (RQ4)

3.7 Parameter Sensitivity (RQ5)

4 Related Work

MixerFlow: MLP-Mixer Meets Normalising Flows

3 Preleminaries

4 MixerFlow Architecture and Its Components

5 Experiments.

5.1 Density Estimation on 3232 Datasets

5.2 Density Estimation on 6464 Datasets

5.3 Enhancing MAF with the MixerFlow

5.4 Datasets with Specific Permutations

5.5 Hybrid Modelling

5.6 Integration of Powerful Architecture

6 Conclusion and Limitations

7 Future Work and Broader Impact

Handling Delayed Feedback in Distributed Online Optimization: A Projection-Free Approach

1.1 Our Contribution

1.2 Related Work

2 Projection-Free Algorithms Under Delayed Feedback

2.1 Preliminaries

2.2 Centralized Algorithm

2.3 Distributed Algorithm

3 Numerical Experiments

4 Concluding Remarks

Secure Dataset Condensation for Privacy-Preserving and Efficient Vertical Federated Learning

2.1 Vertical Federated Learning

2.2 Privacy Protection in VFL

2.3 Dataset Size Reduction in FL

3 Preliminaries

3.1 Problem Formulation

3.2 Dataset Condensation

3.3 Secure Aggregation

3.4 Differential Privacy

4 Proposed Approach

4.1 Overview

4.2 Class-Wise Secure Aggregation

4.3 VFDC Algorithm

4.4 Privacy Analysis

5 Experimental Study

5.1 Experimental Setup

5.2 Visualization of Condensed Dataset

5.3 Performance Comparison

5.4 Efficiency Improvement

5.5 Impact of Hyperparameters

Neighborhood Component Feature Selection for Multiple Instance Learning Paradigm

2 Methods

2.1 The Lazy Learning Approach for Multiple Instance Learning Setting

2.2 Neighborhood Component Feature Selection for Single Instance Learning Setting

2.3 Our Proposal: Neighborhood Component Feature Selection for the Multiple Instance Learning Setting

3 Datasets

3.1 Musk Dataset

3.2 DEAP Dataset

4 Experimental Procedure

5 Experimental Results.

5.1 Musk Dataset

5.2 DEAP Dataset

5.3 Comparison with State-of-the-Art

5.4 Statistical Significance

5.5 Computational Complexity

MESS: Coarse-Grained Modular Two-Way Dialogue Entity Linking Framework

2.1 Mention-to-Entities

2.2 Transferred EL

3 Our MESS Framework

3.1 M2E Module

3.2 E2M Module

3.3 SS Module

3.4 Dialogue Module

4.1 Setting

4.2 Results

4.3 Ablation Studies

Session Target Pair: User Intent Perceiving Networks for Session-Based Recommendation

3.1 Problem Statement

3.2 Session-Level Intent Representation Module

3.3 Target-Level Intent Representation Module

3.4 Intent Alignment Mechanism Module

3.5 Prediction and Training

4.1 Experiment Setups

4.2 Overall Performance

4.3 Model Analysis and Discussion

Hierarchical Fine-Grained Visual Classification Leveraging Consistent Hierarchical Knowledge

2.1 Fine-Grained Visual Classification

2.2 Hierarchical Multi-granularity Classification

2.3 Graph Representation Learning

3 Approach

3.1 Problem Setting

3.2 Multi-granularity Graph Convolutional Neural Network

3.3 Hierarchy-Aware Conditional Supervised Learning

3.4 Loss Function

3.5 Tree-Structured Granularity Consistency Rate

4.2 Experimental Settings

4.3 Compared Methods

4.4 Ablation Study

4.5 Comparison with State-of-the-Art Method

4.6 Qualitative Analysis

Backdoor Attacks with Input-Unique Triggers in NLP

3.2 NURA: Input-Unique Backdoor Attack.

3.3 Model Training.

Notes:

Description based on publisher supplied metadata and other sources.

ISBN:

3-031-70341-3

OCLC:

1457699979