Supervisor Meeting - September 19th

[Sevhena]

September 19th @ 2:30pm

Attendance

• [x] Sevhena
• [x] Ayushi
• [x] Mya
• [x] Nivetha
• [x] Tanveer

Agenda

• Get more clarity into the scope of the project (tech used, methods, supervisor knowledge)
• Address some housekeeping issues (set up communication channel, set up repository, weekly meetings)

Questions to Ask

• When can we meet you? Online/in person
• Do you recommend building the library in Python or Rust or Golang?
• Technological conversions are the word, start with python measure the refactor code. Factor in refactoring.
• What language/framework would you recommend for reinforcement learning?
• What Machine Learning System are we using?
• How can we move the repo? Get auth?
• Resources for specific refactoring's? Papers, textbooks...
• How many built-in libraries do you want us to use and how much do you want us to do from scratch?

[Sevhena]

Meeting Minutes

Energy Efficient Library

• Custom code
• Neural network OR machine learning
• Raw calculations vs AI

Other Info

• In Europe till oct14th
• Meeting Thursday 25th 10-11 (one off)
• After that (30th) Monday 2:30pm (weekly)
• Can just of the refactoring library or also add a dashboard or somehow indicate to the user how much more efficient the code became after the refactoring and give suggestions
• Language to use is up to us but don’t use an exotic language. Python will be the easiest (magical language)
• Lots of companies use python
• Start with python and measure how much energy it consumes to parse and refactor and then measure how much energy we are saving after the refactor for the source and compare.
• Stable baseline or TensorFlow: DONT use it
• Pytorch vs TensorFlow shows that there's a shift from TF to Pytorch and its easier to work with.
• Highly recommended PyTorch (industry rec)
• Use RL to learn patterns or use it to learn partially from human
• Factor in uncertainty of humans
• Human input RI but wait till next term if u want to use it
• Could talk to his student about her framework to use (use next term)
  • Decent api
• Have a focused view of the goal
  • First milestone : show we have something (could be libraries) but show that it saves energy
  • Next iteration u can move away from pre-built libraries
  • Optimize algorithms and yes, it'll consume less energy but it's all up to us.
  • Shifting from raw calculations to more ai might actually save energy

[Sevhena]

Tasks to Complete

• Make discord chat with him
• Send him calendar meeting invite for Thursday and the rest of Mondays