Project Background:
Extreme Classification refers to the task of dealing with multi-class and multi-label classification problems where the number of classes or labels is extremely large, often ranging from thousands to millions. The project revolves around addressing the unique challenges posed by such vast label spaces, including computational complexity, data sparsity, and scalability issues. Traditional classification algorithms struggle to handle these large-scale scenarios efficiently, leading to performance degradation and prohibitive training times. This project aims to advance the state-of-the-art in Extreme Classification by developing novel algorithms, optimization strategies, and evaluation metrics tailored specifically for these complex scenarios. By leveraging techniques such as label embedding, sparse learning, and distributed computing, the project seeks to enable accurate and scalable classification across diverse domains such as text categorization, recommendation systems, and image tagging. The ultimate goal is to facilitate the practical deployment of Extreme Classification solutions in real-world applications, driving advancements in machine learning and artificial intelligence research.
Problem Statement
- Dealing with a large number of classes or labels leads to sparse data, making it challenging to train accurate classifiers.
- Traditional classification algorithms struggle to scale efficiently with the increasing number of classes, resulting in high computational costs.
- Existing methods may not scale well to handle extreme classification tasks with millions of labels, hindering their applicability in real-world scenarios.
- Standard evaluation metrics may not effectively capture the performance of classifiers in extreme classification settings, necessitating the development of specialized metrics.
Aim and Objectives
- Develop efficient and scalable algorithms for Extreme Classification tasks with large label spaces.
- Address label sparsity challenges through innovative techniques such as label embedding and sparse learning.
- Improve computational efficiency by exploring distributed computing and parallelization strategies.
- Enhance scalability to handle millions of labels while maintaining model performance and training times.
- Develop specialized evaluation metrics tailored for Extreme Classification to accurately assess classifier performance.
- Apply Extreme Classification techniques to diverse domains like text classification, recommendation systems, and multimedia tagging.
Deep Learning Algorithms for Extreme Classification
- Extreme Multi-label Classification (XML-CNN)
-
FastXML
-
PfastreXML
-
Multi-label Neural Embedding (MNE)
-
Label-Enhanced Attentive Neural Network (LEAN)
-
Classifier Chains (CC)
-
Deep Extreme Multi-label Learning (DEML)
-
Hierarchical Attentional Memory Network (HAMN)
-
DeepXML
-
Extreme Text Classifier (XTC)
Datasets for Extreme Classification
- Amazon Product Dataset
-
Wikipedia Large Text Classification Benchmark (Wiki-LSHTC)
-
ImageNet
-
MSLT
-
Reuters-21578
-
EUR-Lex
-
DBPedia
-
Yelp Dataset
-
Stack Overflow Tag Prediction Dataset
-
MS COCO (Common Objects in Context)
Software Tools and Technologies
Operating System: Ubuntu 18.04 LTS 64bit / Windows 10
Development Tools: Anaconda3, Spyder 5.0, Jupyter Notebook
Language Version: Python 3.9
Python Libraries:
1. Python ML Libraries:
- Scikit-Learn
- Numpy
- Pandas
- Matplotlib
- Seaborn
- Docker
- MLflow
2. Deep Learning Frameworks:
- Keras
- TensorFlow
- PyTorch
