UROC

Human Computation (a.k.a. crowdsourcing) is the idea of combining human and machine intelligence on the Web to help tackle difficult problems that algorithms cannot yet solved. Examples of crowdsourcing includes ReCaptcha which helps digitize books, commercial crowdsourcing platforms such as Amazon Mechanical Turk, and systems for solving scientific puzzles such as FoldIt and Zooniverse.

A major problem in crowdsourcing is incentives -- how can we design user interfaces and reward systems to incentivize participation, especially when people are volunteers, participating out of their own intrinsic motivation? X-with-a-Purpose systems engage participants to perform tasks as a by-product of another activity they are intrinsically motivated to do. Games with a Purpose is a prime example, where game players generate useful annotations as a by-product of playing a casual game. Research goes into making these games both fun and spammer-prove - two important but often conflicitng objectives. In this mini-research project, we will design, prototype and test a variety of incentive mechanisms for volunteer-based crowdsourcing.

DNA and proteins are important molecules that determine or affect many things about an organism. These include properties that are easy to observe, such as the gender and color; and properties that are harder to observe, such as the strength, IQ, and health. The new abilities to measure the hard-to-observe properties, such as an underdeveloped disease, have much greater impacts to improve the living conditions of the mankind. Modern biology and health research rely on advanced instruments to measure the DNA and proteins molecules in a biological system, and produce huge amount of data. Bioinformatics tries to manage, visualize and analyze these large datasets, and computes the answer to the biologists questions. In this mini-research project, a particular measurement, the mass spectrometry, is examined. Mass spectrometry measures the mass (molecular weights) of the peptides in a carefully prepared biological sample. A visualization software tool will be developed to visualize the signals in the data, and associate the signal in the data to peptides in a protein database.

The goal of complexity theory is to determine the fundamental limitations of computers and algorithms or, in other words, to establish the minimal resources (in terms of time, disk space, network communication, etc.) that are required to solve various computational problems. One area in complexity theory that has led to surprising results and baffling open problems is circuit complexity. A circuit is a collection of AND, OR, and NOT gates that take in some Boolean (0 or 1) values and returns a single Boolean value as output. What are the functions that require large circuits to compute? This deceptively simple question has important ramifications---a complete answer to the question could give a proof of the famous P vs. NP problem---but it has also proved to be incredibly elusive. In this mini-project, we will explore some of the open questions in circuit complexity that follow from this central question. We will prove and conjecture, and in the process catch a glimpse of how research in theoretical computer science is conducted.

Dirty data is everywhere. Entry errors, integration from multiple sources, and imperfection of automatic data acquisition techniques are examples of the causes of low-quality data. In the Big Data era, extracting the value of the massive amounts of collected data is crucial in fuelling meaningful analytics and effective business intelligence. Large body of research is in the area of detecting data anomalies and data inconsistencies, and even more approaches are proposed to automatically repair the data. However, it is unclear how to verify the correctness or soundness of dirty databases without reaching out to the main digital sources where this data was born. In this project, we try to use available corpora from text, social media, news and other external sources (mostly unstructured) to support the facts (or errors) in the collected data sets. The project explores using techniques from information retrieval, natural language processing, statistics and probability theory, with a judicious involvment of users and experts to verify and repair low-quality data sets.