Some Applications in Human Behavior Modeling
summary
Models verbal fluency as a random walk observed only at first visits, uses an absorbing random walk construction to compute hit probabilities and builds a nonconvex MLE; experiments on star, grid and ring graphs compare three estimators (ours beats naïve RW and first-two baseline) but consistency and rates remain open. Formalizes machine teaching via the teaching dimension, shows two examples: two points teach a 1D threshold, D+1 points teach a D‑dim Gaussian, runs human categorization experiments that improve generalization, and brute‑force finds size‑3 seed sets on a 4×4 grid.
AI-written summary · about
chapters (19)
- 0:00Intro
- 2:37Censored random walk
- 5:47The estimation problem
- 5:57Each observed step is an absorbing random walk
- 9:19Maximum Likelihood
- 11:09Other estimators
- 12:38Star graph
- 14:012D Grid
- 14:39Ring graph
- 17:01Learning a noiseless 1D threshold classifier
- 20:25Active learning (sequential experimental design)
- 24:22Comparing the three
- 28:16Example 2: Teaching a Gaussian distribution
- 29:04An Optimization View of Machine Teaching
- 32:50Example 3. Works on Humans
- 34:46New task: teaching humans how to label a graph
- 38:07Example: A = local global consistency Zhou et al. 2004
- 39:43Optimal seed set 2: S = 3
- 40:09Questions
related talks
37mActive Teaching, Human Learning
43mToward Human-Centered Machine Learning
26mHow Machine Learning Can Help
01h 04mStatistical Physics and Computation in High Dimension
30mProbabilistic modelling with Scala
41mMachine Learning Models in a Dynamic Environment
01h 03mGeneralization from the behavioral perspective
18mThe Importance of Modeling Data Collection