Key Sequence

Notation

New Concepts

• sorting functions
• assembly
  • Registers

Important Results / Claims

Questions

Interesting Factoids

Core Themes of CS107

how and why of 107:

• how is program data represented in the hardware
• how does the heap work and how is it implemented
• how does a computer know how run code
• how does an executable map onto computer systems
• why is my program doing one thing when I expect it to do something else

“why is this broken system behaving the way it does?”

Core Goals of CS107

fluency

• pointers and memory, and how to make use of them
• an executable’s address space + runtime behavior

competency

• the translation of C to and from assembly
• implement programs with limits of computer arithmetic
• identify bottlenecks and improve runtime performance
• navigate Unix
• ethical frameworks to design and implement software

exposure

computer architecture

New Concepts

• Unix
• C
  • printf
  • bool
• integer

Important Results / Claims

• principles of C
• C limitations

Key Sequence

Notation

New Concepts

• Mencius Philosophy
• types of harm
  • protected group

Important Results / Claims

• Procedural vs. Distributive Fairness

Questions

Interesting Factoids

• logistic regression is a linear classifier
• Naive Bayes is a linear classier: there is literally no interaction between input features

Diffusion Models

We can consider a model between random noise and trees.

For every step, we sample Gaussian noise and add it to the image. The original approach adds Gaussian to the pixels, and nowadays people replace the pixel.

Usually, there is a few thousand steps of noising.

Why is it that we can’t have a one-step policy from noise to pictures? Because of a physics result that says the stability of diffusion becomes intractable at too large steps.