Chapter 1: Introduction

• Audio Check
    - Indication of audio working for online participants.
    - TA available for questions and to manage Zoom's waiting room.

• Course Adjustments and Syllabus
    - Expect shifting in grading policies, assignments in the upcoming week.
    - Anticipate some changes in the syllabus; it will stabilize soon.

• Piazza as Main Tool
    - Piazza will be crucial for course communication.
    - This is the instructor's first use of Piazza; assistance expected from students if issues arise.
    - The official course website provides access to all resources and is open to all, without registration.
    - Main resources found via the course website include:
      - Access to Piazza.
      - Submission area on G Drive for assignments.
      - Lecture slides and recordings.
    - Access Requirements:
      - Requires USC single sign-on for accessing files on G Drive.
      - Lecture slides are secured from public access to prevent unauthorized use.

• Attendance and Health Measures
    - Encouragement to attend courses online if unwell, especially considering COVID-19.

• Course Communication
    - All course-related communication should be via Piazza to ensure clarity and organization.
    - For personal matters or emergencies, use the dedicated course email for private communication.
    - Expect that routine questions will be directed to Piazza and not addressed through the email.

• Office Hours
    - Instructor's office hours: Mondays, 10-11 AM.
    - Office location: Cell 244 (access through an elevator).
    - Office hours primarily for project/research questions, not homework.
    - TAs and Their Office Hours:
      - Gautam & Desheng: Wednesdays, 3-4 PM.
      - Tejas & Bingjie: Thursdays, 10-11 AM.

• Course Deliverables
    - Project constituting 45 points, major source of course credit.
    - More detailed project information and rubric provided throughout the semester.
    - Project Team Requirements:
      - Projects to be done in teams of four, members must contribute evenly.
      - At least two mandatory meetings with TAs for project feedback required for credit.

• Enrollment & New Students
    - Welcome extended to 100 new students, suggesting integration into ongoing course discussions.
    - Mention of demographic surveys needed for identifying class preparedness to structure assignments appropriately.   

• Enrollment Procedures
    - New students are encouraged to complete "Quiz Zero" for demographic information credit during the lecture.

Chapter 2: Deep Learning Model

• General Overview
    - Introduction of deep learning as a subset of machine learning; the former used as a tool to solve ML problems.
    - Definition of AI involves systems designed to perform cognitive tasks typically associated with human intelligence.

• Framing AI Applications
    - Approach to framing an AI application as a machine learning problem.
    - Goals of Machine Learning:
      - Optimize the loss function to minimize error through learned function approximation based on input-output relationships.

• Solution Space
    - Models must be framed in terms of their functions to approximate relationships between inputs (x) and outputs (y).   

• Understanding of Deep Learning
    - Emphasis on representational constraints through data representations and model architectures.
    - Function Approximation
      - Utilizes various model architectures constrained by data representation chosen.

• Deep Learning Architecture Constraints
    - Deciding model architectures is crucial as they define limitations based on selected representations.   

• Examples
    - Regression: Continuous output goals.
    - Classification: Discrete output goals involving non-ordered categories.

Chapter 3: Set Of Data

 - Confusion in Feedback
   - Discussion of issues where machine learning amplifies bias present in training data.
   - Analyzation of the implications within supervised models, gradients, and representations.

Chapter 4: Deep Learning Model

• Recap on Deep Learning Structure
    - Neural networks consist of layers connected through parameters (weights, biases) learning through backpropagation, optimized through specific loss functions.

• Losses, Optimizations, and Training Goals
    - The definition of loss functions is vital in formulating the approach to optimizing models.   

• Training Neural Networks
    - Use of Stochastic Gradient Descent (SGD) for optimization over large data sets due to memory constraints.     # Chapter 5: Modern Deep Learning

• Stochastic Gradient Descent
    - Mini-batching increases randomization and variance, preventing overfitting; ensures data gradients remain within an acceptable scale.     # Chapter 6: Use A Data

• Pixel Representations in Image Processing
    - Understanding the importance of pixel values as inputs to model architectures, while ensuring proper management of images being processed.

Chapter 7: Apply Deep Learning

• Input Representations
    - Necessity of appropriately selecting inputs and representations over complex, larger learning tasks going forward.

Chapter 8: Deep Learning Model

• Recap on Deep Learning Concepts
    - Amidst deep learning strategies in the future, understanding the smoothing of model classes per guidelines and considering variations through accessible software will be crucial to students' success moving through learning curve.

Chapter 9: The Right Model

• Effective Model Choices for Tasks
    - Recommended structures for efficient learning and handling of tasks in conjunction with model categories.

Chapter 10: Conclusion

• Final Recap and Reflective Notes
    - Emphasis on both aid student understanding and implementation of ML to enhance their modeling problem-solving capabilities, engaging with optimal training strategies from initial projects towards larger-scale implementations in the future.