Towards rapid interactive machine learning: evaluating tradeoffs of classification without ...
summary
A web-based tool called CHISSL implements a representation-free O(n) classifier built from a hierarchical clustering parent-pointer array to incrementally relabel data in milliseconds for hundreds of thousands of instances. It shows users a tip-of-the-iceberg recommendations column plus a borderlines column of likely misclassifications; computational results tie or beat semi‑supervised baselines in one‑shot and 100‑label cases and run tens to thousands of times faster, while lab studies show improved or comparable user labeling accuracy.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 1:34Active Learning vs visual labeling
- 3:00CHISSL tool demonstration
- 5:52Background on classification
- 7:18Representation-free algorithm
- 10:27Computational evaluation
- 11:50User study design
- 13:54Results and user feedback
- 15:40Conclusion and future work
- 16:45Q&A session
related talks
17mFrom messy to meaningful data: LLM-powered classification in R
12mIncreasing the Speed and Accuracy of Data Labeling Through an AI Assisted Interface
01h 29mCreate ML framework - introduction to no-code ML
56mDehumanizing Agents: Why Explainability is Crucial in the LLM Era
05mModel LineUpper: Supporting Interactive Model Comparison at Multiple Levels for AutoML
25mProduct Management and AI
37mAutomated Machine Learning: Building a Pipeline
35mML: Beyond confusion matrix, model performance analysis