Showing papers by "Vladimir Jojic published in 2010"

Accelerated dual decomposition for MAP inference

Abstract: Approximate MAP inference in graphical models is an important and challenging problem for many domains including computer vision, computational biology and natural language understanding. Current state-of-the-art approaches employ convex relaxations of these problems as surrogate objectives, but only provide weak running time guarantees. In this paper, we develop an approximate inference algorithm that is both efficient and has strong theoretical guarantees. Specifically, our algorithm is guaranteed to converge to an e-accurate solution of the convex relaxation in O (1/e) time. We demonstrate our approach on synthetic and real-world problems and show that it outperforms current state-of-the-art techniques.

Genovo: de novo assembly for metagenomes

Abstract: Next-generation sequencing technologies produce a large number of noisy reads from the DNA in a sample Metagenomics and population sequencing aim to recover the genomic sequences of the species in the sample, which could be of high diversity Methods geared towards single sequence reconstruction are not sensitive enough when applied in this setting We introduce a generative probabilistic model of read generation from environmental samples and present Genovo, a novel de novo sequence assembler that discovers likely sequence reconstructions under the model A Chinese restaurant process prior accounts for the unknown number of genomes in the sample Inference is made by applying a series of hill-climbing steps iteratively until convergence We compare the performance of Genovo to three other short read assembly programs across one synthetic dataset and eight metagenomic datasets created using the 454 platform, the largest of which has 311k reads Genovo's reconstructions cover more bases and recover more genes than the other methods, and yield a higher assembly score.