🤖

Deep Learning Projects and Concepts Overview

May 18, 2025

CMPE 258: Deep Learning Lecture Notes

Instructor Information

Spring 2025 Project Key Components

Option 1: Deep Learning Models

  • Goal: Build a full pipeline of a Deep Learning application with model training.
  • Key Steps:
    • Question/Problem Formulation: Propose your own application and problem.
    • Data Acquisition and Processing: Identify suitable datasets for training and evaluation.
    • State-of-the-Art Models: Assess and compare state-of-the-art open-source models.
    • Model Training and Evaluation: Modify model architecture, tune parameters, and evaluate.
    • Inference and Optimization: Create an end-to-end application for real-time tests and optimize inference.

Option 2: LLMs and AI Agent

  • Goal: Develop AI applications using LLMs (Large Language Models).
  • Key Steps:
    • Question/Problem Formulation: Propose your own AI application.
    • Data Acquisition: Gather few-shot examples for evaluation.
    • LLM Models Comparison: Compare state-of-the-art LLM models.
    • Feature Implementation: Implement parameter-efficient fine-tuning or AI agent tools.
    • AI App Development: Conduct comprehensive testing and UI evaluation.

Default Projects

  • AI Object Detection for Smart City
  • AI Agent for Airplane Pilot: Develop an advanced airplane pilot assistance system using LLM agents for tasks like radio command translation, manual assistance, and flight route planning.

Common Datasets for ML Projects

  • Image Classification: ImageNet, CIFAR-10/100, MNIST, FashionMNIST
  • Object Detection: COCO, Kitti, Waymo, Argo, nuScenes
  • NLP: IMDB Reviews, SQuAD, WMT19
  • Speech Recognition: LibriSpeech, Mozilla Common Voice
  • Time Series Analysis: UCI Repository
  • Video Analysis: Kinetics
  • Medical Imaging: MURA, Chest X-Ray

Project Requirements

  • Team Requirements: 1-3 members, additional documentation for more members.
  • Milestones: Proposal, presentation, final report.
  • Assessment: Based on problem importance, solution novelty, technical excellence, complexity, creativity, code clarity, presentation, and documentation.
  • Innovation Encouragement: Avoid basic tutorials, focus on substantial modifications and open-source alternatives.

Computing Resources

  • Option 1: Personal machine
  • Option 2: Cloud resources (Google Colab, AWS, Azure)
  • Option 3: SJSU CoE HPC (NVIDIA P100 GPU)
  • Note: No guarantee of resource availability; students cover personal cloud costs.

Deep Learning Concepts

  • Model Building: Similar to training a dog, involves learning from data rather than explicit programming.

AI History Highlights

  • Birth of AI (1943-1956): Introduction of artificial neurons, Turing Test, high-level languages.
  • AI Winters: Periods of reduced interest and funding.
  • Current AI Spring: Advances in translation, image recognition, game-playing systems.
  • Deep Learning Milestones: Introduction of "Deep Learning," success of AlexNet, rise of autonomous vehicles.

Influential Figures in Deep Learning

  • Geoffrey Hinton: Pioneered neural networks, developed AlexNet.
  • Yoshua Bengio and Yann LeCun: Major contributors to AI breakthroughs.

Important AI Contributions

  • AlexNet's Success: Won ImageNet challenge, leading to Google acquiring DNNResearch.
  • OpenAI and Influence: Founded by Ilya Sutskever, furthered AI research and development.

This lecture provides a comprehensive overview of the deep learning landscape, project guidelines for students, and historical context of AI development.