Efficient human computation
|Event Date:||March 31, 2014|
|Speaker Affiliation:||University of Maryland|
|Type:||Prospective ECE Faculty Member
|Contact Name:||Professor Sam Midkiff
|Contact Phone:||(765) 494-3440
|Open To:||ACCEPTABLE FOR ECE 694A
Computers are no longer just tools for data processing. They also serve to coordinate human communication and activities. Human computation (or crowdsourcing) uses these complementary roles to enable solutions to data-intensive problems that combine the capabilities of humans and machines. One such problem is data-driven decisions where the decision-maker already has a clear set of criteria in mind, but lacks the requisite data about each of the alternatives (e.g., specifications, ratings, etc.). Gathering the information can be very time consuming. Crowdsourcing platforms such as Mechanical Turk have made it possible to delegate such tasks to multiple on-demand workers. However, efficiency becomes a chief concern because workers do tasks independently and thus lack the global view needed to decide what information to look at next. I developed AskSheet, a system that takes an arbitrary decision spreadsheet as input, and coordinates crowd workers to supply the inputs. AskSheet leverages the syntactic structure of the spreadsheet formulas to prioritize the decision inputs by value of information. It can calculate a precise value for the cell(s) containing the decision result without needing to acquire every input.
In this talk, I will present AskSheet and show how machine computation can coordinate humans to accomplish complex jobs efficiently. I will also discuss how this pattern can be applied to other types of tasks, and the steps to making human computation a regular part of how people solve data-intensive problems.
Alex Quinn is a Ph.D. candidate at the University of Maryland in the Human Computer Interaction Lab (HCIL), advised by Professor Ben Bederson. His research is about human-in-the-loop solutions to data-intensive problems. His work has resulted in several full papers in top-tier forums, including one about human computation, which has already become the most cited overview of the topic—over 280 citations since 2011—and is widely assigned in university courses. Alex has also contributed to several other areas of human-computer interaction, including mobile storytelling, tabletop interfaces, information visualization, and readability. Some of his research projects have led to long-term public deployments in the Smithsonian and in schools worldwide.