Project Background:
The Shallow Broad Neural Network (SBN) represents an innovative approach to neural network architecture designed to address specific challenges in machine learning applications. Unlike deep neural networks that consist of many layers, SBNs are characterized by a shallow architecture with a broad width, meaning they have a large number of neurons in each layer but fewer layers overall. This design choice is motivated by the need for models that can efficiently process vast amounts of data while maintaining computational efficiency and interpretability. SBNs are particularly suited for tasks where feature extraction is not as complex or hierarchical, such as in certain types of image classification, text classification, and regression tasks.
This project work aims to explore the capabilities of SBNs in various domains and applications. By leveraging the broad width of the network, SBNs can capture complex relationships in data without the computational overhead of deep architectures, making them suitable for resource-constrained environments or real-time processing scenarios. Additionally, the interpretability of SBNs is a significant advantage, as the models decisions can be more easily understood and explained compared to deeper architectures.
Problem Statement
- Developing neural networks that can process large datasets efficiently with fewer layers.
-
Addressing computational resource constraints by designing shallow but broad architectures.
- Enhancing model interpretability for easier understanding of decision-making processes.
- Balancing between model performance and computational efficiency in machine learning tasks.
- Identifying domains and applications where shallow broad neural networks excel compared to deep architectures.
Aim and Objectives
- Develop a Shallow Broad Neural Network (SBN) for efficient and interpretable machine learning.
- Design an SBN architecture with a broad width and fewer layers for efficient data processing.
-
Optimize training algorithms to achieve competitive performance while maintaining computational efficiency.
- Enhance model interpretability by incorporating explainable features and decision-making processes.
- Conduct comparative studies to evaluate the trade-offs between SBNs and deep architectures.
- Explore applications in resource-constrained environments where SBNs offer computational advantages.
- Develop tools and frameworks for implementing and deploying SBNs in real-world scenarios.
-
Collaborate with domain experts to validate the effectiveness of SBNs in specific machine learning tasks.
Contributions to Shallow Broad Neural Network
- Developed an efficient and interpretable SBN architecture.
Optimized training algorithms to achieve competitive performance with fewer layers.
- Enhanced model interpretability through explainable features and decision processes.
- Explored applications in resource-constrained environments, showcasing SBNs computational advantages.
Deep Learning Algorithms for Shallow Broad Neural Network
- Feedforward Neural Networks (FNNs)
-
Radial Basis Function Networks (RBFNs)
-
Extreme Learning Machines (ELMs)
-
Generalized Regression Neural Networks (GRNNs)
-
Cascade Correlation Neural Networks (CCNNs)
-
Probabilistic Neural Networks (PNNs)
-
Associative Neural Networks (ANNs)
-
K-Means Clustering Neural Networks
-
Self-Organizing Maps (SOMs)
-
Adaptive Resonance Theory Neural Networks (ARTNNs)
Datasets for Shallow Broad Neural Network
- MNIST dataset
-
CIFAR-10 (Canadian Institute for Advanced Research-10) dataset
-
Fashion-MNIST dataset
-
IMDB (Internet Movie Database) movie reviews dataset
-
Wine quality dataset
-
Boston housing dataset
-
Iris dataset
-
Bank marketing dataset
-
Breast cancer Wisconsin (diagnostic) dataset
-
Human activity recognition dataset
Performance Metrics
- Accuracy
-
Precision
-
Recall
-
F1 Score
-
Mean Squared Error (MSE)
-
Mean Absolute Error (MAE)
-
Root Mean Squared Error (RMSE)
-
Area Under the Receiver Operating Characteristic Curve (AUC-ROC)
-
Area Under the Precision-Recall Curve (AUC-PR)
-
Mean Average Precision (MAP)
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
