Back to in-person
Finally, we go back to in-person! It is so exciting to see our teammates face to face. Also, we became much more efficient than before because in-person meetings can get everyone on board. This week we made a lot of progress. First, we had our ¼ walkaround on Monday. We received a lot of useful feedback from professors. And then, on Tuesday, Wednesday and Thursday, we had in-person meetings to think deeply about this feedback in order to improve our ideas. Finally, we got a clear idea and we presented them to our instructor Ruth and our client from Google – Erin on Friday.
In 1/4 walkarounds
On Monday, we had our ¼ walkaround. We presented our ideas about visualizing different machine learning models and also the design of mini-games to professors. We received so much useful feedback and suggestions. Here is a summary of feedback from the walkaround.
Accept:
- Show the limitation of AI
- Don’t competing with the machine learning algorithm
- Don’t choose IOS as platform, PC is a good choice
- Target user needs to be much more specific
- Focus on model visualization first.
- Game needs to help players learn ML Model.
Keep thinking:
- What data set do you think your audience might want to play with?
- Who is our target audience?
- Keep thinking why when we make decision
- What concept of ML do we want to teach?
- How to find playtester?
- What if AI makes a mistake?
In-person meetings
After receiving the feedback from ¼ walkaround, we finally went back to ETC and started to brainstorm together. We learned different ML algorithms together. We search the proper dataset together and we discuss the ideas together. Everything has become so high-efficient. We came up with tons of ideas. But finally, we narrowed it down to one specific sprint, which is creating a K-means Clustering Visualization to show data as trees, and clusters as islands in the ocean, aggregated in an unsupervised machine learning process. The user can interact with the data and algorithm in real-time. The user can also experience an example related to the real-life application of this model. We want to choose PC Web Browser as our platform and the duration for this sprint is 3 weeks. We also decided our demographic is higher education students. We also came up with some audience-focused questions and team-focused questions from four different perspectives, which are understandability, data visualization, ML, and 3D. And finally, we decided our goal for this project is to document the discoveries using white paper or documentary format video of prototypes.
Our presentation slides:
https://docs.google.com/presentation/d/1gUu1EtQ-ywfF5vByvPfX99yL6wJSddQxWZEX18lqRA0/edit?usp=sharing
Meeting with Client
On Friday, we had our first meeting with Erin from Google. Here is the flow of the meeting.
- Start with self-introduction. We introduced our background and what we hope to learn through this project.
- We presented our ideas to Erin
- Erin introduced this “gift” project
- Erin gave us the feedback about Explainability, Controllability, Paper direction
- Erin said the recommendation system is a good direction. She also gave us a brief introduction of Google Search.
- Erin suggested a new direction about visualizing complex models part by part.
- Schedule a weekly meeting with her on Friday 4 p.m.
We also received so much good feedback from Erin. Here is a summary.
- Explainability : Use interaction, animation to show how a algorithm works
- Controllability: What is the end result of the behavior of a model in a given context which is very important, and how can a person express intent into the function of that end result
- Paper direction: 3d Interactive interfaces for model explainability & controllability
- Suggest a new direction:
Taking a reasonably complex model and making it intuitively interactive enough, that it doesn’t matter that whether they understand whether it works or not, that would be a different approach
Only render the part that I’m looking at, and let people explore through the giant model, and you track where they went so you know which parts have been reviewed. Using GPT-3 as an example
- Research project vs Explainability project (Educating)
Research project: Come up with different 3D visualizations. Conduct in-depth research, write papers that can be submitted to the conferences.
- Erin will provide conferences that our project can be submitted to. Google research team can provide resources.
Explainability Project: Since how ML works is regarded as a “black box” try to find a method to explain how ML works to a general audience.
- Possible method can be creating an animation.