Helsinki Institute for Information Technology

About: Helsinki Institute for Information Technology is a facility organization based out in Espoo, Finland. It is known for research contribution in the topics: Population & Bayesian network. The organization has 630 authors who have published 1962 publications receiving 63426 citations.

Polya-gamma augmentations for factor models

Abstract: Bayesian inference for latent factor models, such as principal component and canonical correlation analysis, is easy for Gaussian likelihoods with conjugate priors using both Gibbs sampling and mean-eld variational approximation. For other likelihood potentials one needs to either resort to more complex sampling schemes or to specifying dedicated forms for variational lower bounds. Recently, however, it was shown that for specic likelihoods related to the logistic function it is possible to augment the joint density with auxiliary variables following a P olya-Gamma distribution, leading to closed-form updates for binary and over-dispersed count models. In this paper we describe how Gibbs sampling and mean-eld variational approximation for various latent factor models can be implemented for these cases, presenting easy-to-implement and ecient inference schemas.

Fully Dynamic de Bruijn Graphs

Abstract: We present a space- and time-efficient fully dynamic implementation de Bruijn graphs, which can also support fixed-length jumbled pattern matching.

DiMaS: distributing multimedia on peer-to-peer file sharing networks

Abstract: This demonstration presents the Digital Content Distribution Management System (DiMaS). DiMaS proves as a concept that it is possible to make a system for multimedia producing communities to publish their work on highly popular P2P networks, and importantly, the system enables producers to insert content metadata, to manage intellectual property and usage rights, and to charge for the consumption. All this can be done without introducing another new content or metadata file format and a dedicated client application to read the format.

Explaining a Weighted DAG with Few Paths for Solving Genome-Guided Multi-Assembly

Abstract: RNA-Seq technology offers new high-throughput ways for transcript identification and quantification based on short reads, and has recently attracted great interest. This is achieved by constructing a weighted DAG whose vertices stand for exons, and whose arcs stand for split alignments of the RNA-Seq reads to the exons. The task consists of finding a number of paths, together with their expression levels, which optimally explain the weights of the graph under various fitting functions, such as least sum of squared residuals. In (Tomescu et al. BMC Bioinformatics, 2013) we studied this genome-guided multi-assembly problem when the number of allowed solution paths was linear in the number of arcs. In this paper, we further refine this problem by asking for a bounded number $k$ of solution paths, which is the setting of most practical interest. We formulate this problem in very broad terms, and show that for many choices of the fitting function it becomes NP-hard. Nevertheless, we identify a natural graph parameter of a DAG $G$ , which we call arc-width and denote $\langle G\rangle$ , and give a dynamic programming algorithm running in time $O(W^k\langle G\rangle ^k(\langle G\rangle + k)n)$ , where $n$ is the number of vertices and $W$ is the maximum weight of $G$ . This implies that the problem is fixed-parameter tractable (FPT) in the parameters $W$ , $\langle G\rangle$ , and $k$ . We also show that the arc-width of DAGs constructed from simulated and real RNA-Seq reads is small in practice. Finally, we study the approximability of this problem, and, in particular, give a fully polynomial-time approximation scheme (FPTAS) for the case when the fitting function penalizes the maximum ratio between the weights of the arcs and their predicted coverage.

Partial order MCMC for structure discovery in Bayesian networks

Abstract: We present a new Markov chain Monte Carlo method for estimating posterior probabilities of structural features in Bayesian networks. The method draws samples from the posterior distribution of partial orders on the nodes; for each sampled partial order, the conditional probabilities of interest are computed exactly. We give both analytical and empirical results that suggest the superiority of the new method compared to previous methods, which sample either directed acyclic graphs or linear orders on the nodes.