This is a follow-up project of Parasocial Consensus Sampling (PCS). The basic idea is to outsource human cognition tasks to many non-experts. By aggregating many non-experts' data, it is almost guaranteed that we can get a very good approximation of "ground truth".

When people interact, their speech prosody, gesture, gaze, posture, and facial expression contribute to establishment of a sense of rapport. Rapport is argued to underlie success in negotiations, psychotherapeutic effectiveness, classroom performance and even susceptibility to hypnosis. The rapport project uses machine vision and prosody analysis to create virtual humans that can detect and respond in real-time to human gestures, facial expressions and emotional cues and create a sense of rapport. These techniques have a demonstrable beneficial impact on human interaction.

Traditionally, virtual humans learn from annotated recordings of face-to-face interaction. However, there are some drawbacks: (1) human behavior contains variability and not all of them are positive examples of the behavior a virtual human attempts to learn. (2) a virtual human attempts to learn a general behavior pattern that it could apply across social situations, yet each example in a face-to-face dataset is intrinsically idiosyncratic, illustrating how one particular individual responded to another. We proposed Parasocial Consensus Sampling (PCS) to leverage human cognitive power to solve the two problems. This framework is based on the theory of Parasocial Interaction(pdf). It allows multiple individuals to experience the same social situation so that their responses can be aggregated into a consensus view of how a typical individual would respond. Read More. PCS has been applied to model:

• Visualize any time-series data (continuous or discrete, offline or online)
• can conveniently navigate through long time-series data
• can label time-series data and navigate through labels
• with customized display settings
• associate time-series data (i.e. human behaviors) with corresponding materials (i.e. video clips)

This game is developed for the iPhone/iPod Touch, and can support up to 4 players (devices) via Wifi or Bluetooth. This prototype was created over a semester in USC's CSCI526 class. We are a 7-person team (3 programmers, 1 producer, 1 game designer, 1 artist, 1 sound designer). demo

• Implemented the GeoBox algorithm for fast geocode query
• Developed this demo using Google App Engine and Google MAP API