EEB131 - Introduction to Artificial Intelligence and Information Technology

AI Frameworks and Libraries for Deep

Learning Applications
Isack Farady, Ph.D.
Outline
1. Overview of AI Frameworks and Libraries
2. Popular Deep Learning Frameworks
3. Specialized Libraries for Deep Learning
4. Tools and Ecosystem
5. Setting Up a Deep Learning Environment
6. Hands-On Example
7. Homework
1. Overview of AI Frameworks and Libraries

• Key features of Framework: Ease of use, scalability,

HW compatibility, community support

image
Factors to Consider
• High-Level APIs vs. Low-Level Control:
• Frameworks like Keras offer a high-level API that
simplifies model building with minimal code.
• PyTorch and TensorFlow provide more flexibility but
may require more effort to customize models.
• Syntax and Readability:
• PyTorch follows a Pythonic approach, making it easier
for developers already familiar with Python.
• TensorFlow 1.x was considered complex, but TensorFlow
2.x improved usability with eager execution and
integrated Keras.
Factors to Consider
High-Level API Example (Keras - TensorFlow)

The model architecture is defined with a

simple, readable structure.
Factors to Consider
High-Level API Example (Keras - TensorFlow)

• The entire deep learning pipeline (data loading, model

definition, compilation, and training) is written in just a
few lines.
• Instead of manually handling weight initialization,
activation functions, and forward propagation, Keras
automates these processes.
• There is no need to manually define computational
graphs, unlike low-level TensorFlow implementations.
2. Popular Deep Learning Frameworks

image
TensorFlow
• Developed by Google Brain and released as an
open-source machine learning framework in 2015.
• Designed for scalability, enabling large-scale deep
learning training and deployment across CPUs,
GPUs, and TPUs.
• Primarily used for both research and enterprise
applications
• https://www.tensorflow.org/tutorials
Key Features of TF
• TensorBoard for Visualization
• Provides an interactive dashboard to monitor training progress, loss curves, and
model architectures.
• Supports real-time tracking of experiments.
Key Features of TF
• Keras API Integration (check previous ppt)
• TensorFlow includes Keras as a high-level API,
simplifying model creation and training.
• Allows quick prototyping with minimal code.
• TensorFlow Lite for Mobile and Edge Devices
• Optimized for mobile AI applications (Android, iOS,
Raspberry Pi).
• Supports model quantization to reduce computational
load.
Use Cases and Applications
• Image and Speech Recognition:
• Used in Google Photos for image tagging and object detection.
• Employed in speech-to-text applications (e.g., Google Assistant).
• Healthcare AI Applications:
• Used in medical imaging analysis (e.g., detecting cancer in CT
scans).
• Applied in drug discovery and patient diagnostics.
• Cloud-Based AI Solutions (GCP):
• TensorFlow is deeply integrated with Google Cloud AI services.
https://www.tensorflow.org/tfx/tutorials/transform/data_preprocessing_w
ith_cloud
• Supports auto-scaling and distributed training in cloud
environments.
Pytorch
• Developed by Facebook AI Research (FAIR) and
released as an open-source deep learning
framework in 2016.
• Designed for flexibility and research, making it the
preferred choice for academic studies.
• Uses dynamic computation graphs, allowing real-
time modifications during training.
• https://pytorch.org/tutorials/beginner/basics/intro
.html
Key Features of PyTorch
• TorchScript for Model Deployment
• Supports Pythonic debugging using standard
debugging tools.
• Easy Integration with Hugging Face and OpenAI
Tools
• Works seamlessly with Hugging Face Transformers for
NLP.
• Used in OpenAI’s GPT models and DALL·E.
Keras
• Originally developed as a standalone framework,
later integrated into TensorFlow.
• Serves as a high-level deep learning API, simplifying
model creation and training.
• Ideal for beginners, educators, and rapid
prototyping.
Key Features of Keras
• User-Friendly API for Quick Prototyping
• Models can be built in a few lines of code.
• Abstracts low-level implementation details.
• Pre-Built Models Available
• Offers ready-to-use pre-trained models via Keras
Applications.
• Supports transfer learning for quick model development.
• https://keras.io/
Keras
https://github.com/PacktPublishing/Deep-
Learning-with-TensorFlow-and-Keras-3rd-
edition
Keras

https://github.com/PacktPublishing/Deep-Learning-with-TensorFlow-and-Keras-
3rd-edition
4. Specialized Libraries for Deep Learning
Specialized libraries for Deep Learning – Useful for
first-time users:
• Hugging Face Transformers (NLP)
• OpenCV (Computer Vision)
• Fast.ai (Simplified Deep Learning)
• Albumentations (Data Augmentation)
• Matplotlib & TensorBoard (Model Visualization)
Hugging Face Transformers (NLP)
• Hugging Face is an open-source library that provides
tools and pre-trained models for NLP tasks.
• State-of-the-Art Models: Access to the latest
advancements in NLP (e.g., GPT, BERT, RoBERTa).
• Community and Resources: Hugging Face has a large
community, tutorials, and documentation to help
students get started.
• Support for multiple frameworks (PyTorch, TensorFlow,
JAX)
• https://huggingface.co/
Benefits of Hugging Face (for Students)
• Hands-On Learning: Students can experiment with
real-world NLP tasks.
• No Need for Expensive Hardware: Pre-trained
models can be fine-tuned on smaller datasets using
free resources like Google Colab.
• Community Support: Hugging Face has an active
community and forums where students can ask
questions and share projects.
• Open Source: All tools and models are free to use.
OpenCV (Computer Vision)
• Provides image processing functions for object
detection, face recognition, and video analysis.
• Compatible with deep learning models
(TensorFlow/PyTorch).
• https://docs.opencv.org/4.x/d9/df8/tutorial_root.h
tml

HW:
Find and Run any OpenCV code for a simple image
processing algorithm!
Fast.ai (Simplified Deep Learning)
• https://www.fast.ai/
• Built on PyTorch and simplifies deep learning model
training.
• Provides easy-to-use high-level APIs for fast
experimentation.
• It’s great for: Students who want quick model
training without deep technical details.
Useful Model Visualization Libraries

• Matplotlib & TensorBoard

• https://matplotlib.org/stable/index.html
• https://www.tensorflow.org/tensorboard