Skip to content

Latest commit

 

History

History
77 lines (66 loc) · 3.71 KB

internal_ideas.md

File metadata and controls

77 lines (66 loc) · 3.71 KB

ViennaUp AI for Sustainability Hackathon '23

Challenge Automated Recycling

  • Input
    • images from a custom hardware scanner
    • some data samples are already provided (via AWS)
    • additional data can be collected with the available hardware
    • seems only to be camera images (no fancy modalities like uv-scanner etc.)
  • AI model
    • self-learning solution that balances performance vs. time to deploy.
  • Business aspects
    • What business model patterns might be applied? Please develop a promising business model.
    • Not clear if it is aimed to be a B2B or B2C solution
  • User Experience
    • Please design your solution and interface.
    • image driven UI, in python or javascript?
    • UI for end-user (B2C) or expert user (B2B)?

Comments on the specification

  • I think they want us to design a system that can be deployed to customers and customers can easily perform a fine-tuning step
    • e.g.: deployed to MA48 and they can provide sample images and the system quickly learns how to classify certain trash objects and gives rich feedback how to system performs
  • the two key components seem to be the self-learning capability as well as communicating the models performance to the user (visualizing metrics, fail-cases, problem classes etc.)

Milestones during hackathon

  • understanding the problem

    • in the first hours we focus on this
    • we need more information to understand the problem and clearify some a assumptions we made (e.g. B2B vs. B2C)
    • achieved when: We know what the challenge is and what the want from us

  • finding a suitable model & training process

    • the more we know how the self-learning aspect could look like the sooner we can work on UX, prototype and business plan
    • achieved when: We approximately know how our self-learning system could work & how we want the user to interact with it

  • user communication & visualization idea

    • we have to think of how we want to visualize
    • achieved when: we know how to visualize validation performance and inference to the user?

  • prototype

    • we need a working prototype that shows how to interact with our system
      • e.g. in order to retrain the model and get results how the model performs
    • achieved when: we have integrated the model with a working UI that we can live demonstrate during our pitch

  • coming up with a suitable business pattern

    • after we know if B2B or B2C and how the solution should approximately look like we can think about business plans
    • achieved when: we have a idea how to produce revenue with our solution, provided some case studies as well as calculated possible market size etc.

  • pitch presentation

    • need to clearify if live working prototype is required or not
    • achieved when: we pitched that thing

responsibilities among the team

  • research model & self-learning process
    • Lisa, Reza, Matthias
  • AWS infrastructure & deployment
    • Reza
  • metrics & validation
    • Herwig, Reza, Matthias
  • visualization of metrics
    • Florian P., Herwig
  • user interface (& UX)
    • Florian P., Herwig, Florian K., Tai
  • end-to-end integration
    • Matthias, Florian P.
  • business plan
    • Tai, Lisa
  • pitch planning
    • Lisa, Reza, Florian K., (Tai)
  • pitch presentation
    • Lisa, Florian K.

collection of ideas

random thoughts that are worth to be considered in future

  • gamification to improve the UX
    • e.g. supermarket deploys the hardware and offers discount for heavy users (who recycle a lot the correct way?)
  • Saliency Map (and other XAI techniques for UX & visualization?)
  • use LoRA for self-learning/fine-tuning (--> more stable, less catastrophic forgetting)