Yuxin Chen

iLike: integrating visual and textual features for vertical search

Content-based image search on the Internet is a challenging problem, mostly due to the semantic gap between low-level visual features and high-level content, as well as the excessive computation brought by huge amount of images and high dimensional features. In this paper, we present iLike, a new approach to truly combine textual features from web pages, and visual features from image content for better image search in a vertical search engine. We tackle the first problem by trying to capture the meaning of each text term in the visual feature space, and re-weight visual features according to their significance to the query content. Our experimental results in product search for apparels and accessories demonstrate the effectiveness of iLike and its capability of bridging semantic gaps between visual features and abstract concepts.

Privacy preserving group linkage

The problem of privacy preserving record linkage is to find the intersection of records from two parties, while not revealing any private records to each other. Recently, group linkage has been introduced to measure the similarity of groups of records [19]. When we extend the traditional privacy preserving record linkage methods to group linkage measurement, group membership privacy becomes vulnerable - record identity could be discovered from unlinked groups. In this paper, we introduce threshold privacy preserving group linkage (TPPGL) schemes, in which both parties only learn whether or not the groups are linked. Therefore, our approach is secure under group membership inference attacks. In experiments, we show that using the proposed TPPGL schemes, group membership privacy is well protected against inference attacks with a reasonable overhead.

Learning Shape Analysis

We present a data-driven verification framework to automatically prove memory safety of heap-manipulating programs. Our core contribution is a novel statistical machine learning technique that maps observed program states to (possibly disjunctive) separation logic formulas describing the invariant shape of (possibly nested) data structures at relevant program locations. We then attempt to verify these predictions using a program verifier, where counterexamples to a predicted invariant are used as additional input to the shape predictor in a refinement loop. We have implemented our techniques in Locust, an extension of the GRASShopper verification tool. Locust is able to automatically prove memory safety of implementations of classical heap-manipulating programs such as insertionsort, quicksort and traversals of nested data structures.

IPKB: a digital library for invertebrate paleontology

In this paper, we present the Invertebrate Paleontology Knowledgebase (IPKB), an effort to digitize and share the Treatise on Invertebrate Paleontology. The Treatise is the most authoritative compilation of invertebrate fossil records. Unfortunately, the PDF version is simply a clone of paper publications and the content is in no way organized to facilitate search and knowledge discovery. We extracted texts and images from the Treatise, stored them in a database, and built a system for efficient browsing and searching. For image processing in particular, we segmented fossil photos from figures, recognized the embedded labels, and linked the images to the corresponding data entries. The detailed information of each genus, including fossil images, is delivered to users through a web access module. Some external applications (e.g. Google Earth) are acquired through web services APIs to improve user experience. Given the rich information in the Treatise, analyzing, modeling and understanding paleontological data are significant in many areas, such as: understanding evolution; understanding climate change; finding fossil fuels, etc. IPKB builds a general framework that aims to facilitate knowledge discovery activities in invertebrate paleontology, and provides a solid foundation for future explorations. In this article, we report our initial accomplishments. The specific techniques we employed in the project, such as those involved in text parsing, image-label association and meta data extraction, can be insightful and serve as examples for other researchers.

Near-optimal Bayesian active learning with correlated and noisy tests

We consider the Bayesian active learning and experimental design problem, where the goal is to learn the value of some unknown target variable through a sequence of informative, noisy tests. In contrast to prior work, we focus on the challenging, yet practically relevant setting where test outcomes can be conditionally dependent given the hidden target variable. Under such assumptions, common heuristics, such as greedily performing tests that maximize the reduction in uncertainty of the target, often perform poorly. In this paper, we propose ECED, a novel, computationally efficient active learning algorithm, and prove strong theoretical guarantees that hold with correlated, noisy tests. Rather than directly optimizing the prediction error, at each step, ECED picks the test that maximizes the gain in a surrogate objective, which takes into account the dependencies between tests. Our analysis relies on an information-theoretic auxiliary function to track the progress of ECED, and utilizes adaptive submodularity to attain the near-optimal bound. We demonstrate strong empirical performance of ECED on two problem instances, including a Bayesian experimental design task intended to distinguish among economic theories of how people make risky decisions, and an active preference learning task via pairwise comparisons.

Cephalometric Landmark Tracing Using Deformable Templates

Automatic detection and identification of landmarks in cephalometry is of great significance to orthognathic surgery and clinic applications. Motivated by the increasing demands of computerized cephalometric analysis, we present a tree-shaped deformable template which detects the landmark points of a grayscale cephalometric x-ray image. After normalization, a group of randomly selected images are used to train the geometric prior, and a dynamic programming algorithm enhanced by down sampling is employed to find the optimal landmark configuration. The proposed algorithm demonstrates promising detection results as well as time efficiency on both soft and hard contours. This leads to a significant improvement over the state-of-art diagnostic tools in the area of cephalometric diagnosis.