Welcome to my collection of deep learning projects! 🌟
This repository showcases various neural network architectures and their applications in image processing, sequence prediction, multi-output modeling, and more.
The repository is organized into the following directories:
- ANN/ : Implementations of Artificial Neural Networks.
- CNN/ : Convolutional Neural Networks for image-related tasks.
- RNN/ : Recurrent Neural Networks for sequence data.
- *.ipynb files : Jupyter Notebooks demonstrating various models and experiments and projects
- Explore basic to advanced ANN architectures.
- Tasks include classification and regression.
- Example: Predicting house prices, digit recognition.
- CNNs for image classification, object detection, and feature extraction.
- Understand convolution, pooling, and feature extraction.
- Example: CIFAR-10, MNIST classification.
- RNNs and variants (SimpleRNN, GRU, LSTM) for sequence prediction.
- Applications: Time series forecasting, NLP, sentiment analysis.
- Models predicting more than one target simultaneously.
- Example: Age & gender prediction from images.
Transformers revolutionized NLP and now dominate many AI tasks.
They replace RNNs by using self-attention, allowing models to learn relationships between tokens in parallel.
- Self-Attention: Each token attends to all others
- Multi-Head Attention: Multiple attention heads capture richer features
- Positional Encoding: Adds sequence order information
- Encoder / Decoder Architecture:
- Encoder → representation tasks
- Decoder → autoregressive generation
Transformers handle long-range dependencies extremely well and enable massive parallelism.
They are used in NLP, vision (ViT), speech, and multimodal models.
GPT is a decoder-only transformer trained via next-token prediction.
- Autoregressive (left-to-right generation)
- Causal masking
- Excellent at text generation, conversation, reasoning, summarization
- Scales extremely well with large datasets and model sizes
BERT is an encoder-only transformer trained using masked language modeling (MLM).
- Deep bidirectional understanding
- Great for classification, QA, NER, embeddings
- Learns context from both left & right of a token
- Often fine-tuned for downstream tasks
To run the projects locally:
- Clone the repository:
git clone https://github.com/chandan11248/deep-learning.git
• cd deep-learning
-
Install dependencies:
• pip install -r requirements.txt -
Navigate to the desired project directory and run the corresponding Jupyter Notebook:
jupyter notebook project.ipynb
• datacamp
• documentation
• youtube
Feel free to explore the concepts, contribute, or reach out if you have any questions or suggestions!