SOCS Graduate Seminar Series Seminar Schedule

The design of RNA sequences folding into predefined secondary structures is a milestone for many synthetic biology and gene therapy studies. Most of the current software uses similar local search strategies (i.e. a random seed is progressively adapted to acquire the desired folding properties) and more importantly do not allow the user to control explicitly the nucleotide distribution such as the GC-content in their sequences. However, the latter is an important criterion for large-scale applications as it could presumably be used to design sequences with better transcription rates and/or structural plasticity. We introduce a novel algorithm to design RNA sequences folding into target secondary structures with a predefined nucleotide distribution. It uses a global sampling approach and weighted sampling techniques. We show that our approach is fast (i.e. running time comparable or better than local search methods), seed-less (we remove the bias of the seed in local search heuristics), and successfully generates high-quality sequences (i.e. thermodynamically stable) for any GC- content. To complete this study, we develop an hybrid method combining our global sampling approach with local search strategies. Remarkably, our glocal methodology overcomes both local and global approaches for sampling sequences with a specific GC content and target structure.

For those of us who have an empirical component in our research, there are ample of opportunities to present experimental results in one graphical form or another. In this tutorial, I will present some examples of graphs I have produced for a paper. Bring your laptop and you can produce some graphs too.

I will be in MC103 at 11h30 to assist anyone who wants to set up R in their laptop. The SOCS machines also have R installed if you want to VNC in.

If you know R already, please contact me if you can help. It'd be great if we have a few people at 11h30 to help others with the installation and during the tutorial.

Over the past few years, Twitter has emerged as an increasingly influential platform for real-time information distribution and discovery. However, it has been taking a bit of dark turn into adult content, which could significantly hinder the further popularity of Twitter and/or cause serious legal issues. Existing techniques for adult content detection are ill-suited for detecting adult accounts in Twitter.

To tackle this problem, we propose iterative social based classification (ISC)}, an effective solution for adult account detection for online social networks. ISC consisting of three key components: (1) collective interest, a novel social link based feature for effective discrimination of adult accounts; (2) a tag based label propagation algorithm which explores the tagged content embedded in tweets to further boost the detection accuracy; and (3) a social based linear classifier that integrates social consistency based on the collective interest feature and maximal tag scores computed from tag based label propagation. Evaluations using large-scale real-world Twitter data demonstrate that our ISC solution significantly outperforms existing methods in detecting adult accounts. It is able to identify adult accounts accurately among $1.07$ million accounts using only $100$ labeled adult accounts.

The dominant role of social networking in the web is turning human relations into conduits of information flow. This means that the way information spreads on the web is determined to a large extent by human decisions. Consequently, information security, confidentiality and integrity of shared data, lies on the quality of the collective decisions made by the users. Recently, many access control schemes have been proposed to control unauthorized propagation and modification of information in online systems; however, there is still a need for mechanisms to evaluate the risk of information leakage and unauthorized modifications within online systems. First, the thesis focuses on the confidentiality of information in online social networks. A novel community-centric confidentiality control mechanism for information flow management on the social web is presented. A Monte Carlo based algorithm is developed to determine the potential spread of a shared data object and to inform the user of the risk of information leakage associated with different sharing decisions she can make in a social network. The scheme also provides a facility to reduce information flowing to a specific user (i.e., black listing a specific user). Second the thesis focuses on the integrity of artifacts in crowdsourcing systems. A new approach for managing the integrity of contents created in crowdsourcing repositories named Social Integrity Management (SIM) is presented. SIM integrates two conflicting approaches to manage integrity in crowdsourcing systems: owner-centric and owner-free schemes. The ownership bottleneck is relaxed by including co-ownerships and having multiple versions. Finally, the thesis presents a thorough analysis of the Stack Exchange sites as an example of widely used crowdsourcing question answering systems. The dump datasets are used to analyze various user behaviors in crowdsourcing question answering systems by considering the effect of tagging, user reputation and user feedback. Observed characteristics from the studies are used in the modeling and evaluation of social integrity management.

The heart is composed of elongated muscle cells that are grouped together to form cardiac muscle fibers, known as myofibers. In healthy hearts, they are arranged in an efficient manner to allow the pumping of blood to the whole body. Characterizing the geometry and variability of myofibers is central to our understanding of normal heart function. Originating with pioneering work by Streeter more than 40 years ago, research has mostly focused on large-scale descriptions of their arrangement. Recent advances in medical imaging, with the development of diffusion magnetic resonance imaging (dMRI), have allowed to study myofibers non-invasively, in three-dimensions, and at a high resolution. This contrasts with the traditional histological methods that require tedious heart dissections and damage the cardiac issue. dMRI has also revealed important local features of myofibers across many species, which are difficult to explain using global models of cardiac fiber architecture. Consequently, research has been increasingly shifting away from the shortcomings and inconsistencies found in global models, moving instead to local models of myofiber geometry. During this talk I will show how we can gain valuable insight into how myofibers are arranged, by analyzing how they bundle and curve together using a mathematical framework derived from differential geometry.

Transcriptional regulatory networks, the biochemical systems controlling the transcription of genes into RNA in response to activating or repressing inputs from transcription factor molecules, are able to robustly retain their functionality against a wide array of environmental perturbations and evolutionary mutations. However, what injects such robustness in these systems remains largely unexplained. Previous studies have principally focused on identifying topological features of these networks that are unlikely to occur in random networks and then determining the impact of these features on robustness and other dynamical aspects of the system. While this approach has yielded significant insights into the design principles of robust biological systems, a comprehensive analysis of how these topological features act in conjunction with the parameters of the system has not been conducted yet. In this ongoing project, we first analyze how the activating and repressing connections are distributed within the E Coli transcriptional network to identify some features which deviate from the expected behaviour under random connectivity and then, through generation of random ensembles of networks and simulating their dynamics using a standard discrete time boolean network dynamics model, determine the robustness induction effect of these features. Our result thus far indicates that both the first and second order aspects of this connectivity pattern exert impacts on the robustness of the network, suggesting a synergy between the topological features and parametric features defining the system towards attaining robustness.

Non-player characters that act as companions for players are a common feature of modern games. Designing a companion that reacts appropriately to the player's experience, however, is not a trivial task, and even current, triple-A titles tend to provide companions that are either static in behaviour or evince only superficial connection to player activity. To address this issue I am going to present an adaptive companion that analyses the player's in-game experience and behaves accordingly. We evaluate our adaptive companion in different, non-trivial scenarios, as well as compare our proposed model to a straightforward approach to adaptivity based on Dynamic Difficulty Adjustment. The data collected demonstrates that the adaptive companion has more influence over the player's experience and that there exists an orthogonality between our companion adaptivity and the more traditional combat/health scaling approaches to difficulty adjustment. Using adaptive companions is a step forward in offering meaningful and engaging games to players.

MATLAB is the most popular language among scientists for solving numerically intensive problems but it does not allow scientists to take full advantage of today’s supercomputers which they have access to. We try to solve this problem by developing a MATLAB to X10 translator.

X10 is a programming language being developed by IBM research to provide a programming model that provides scalability and productivity for new generation supercomputing architectures. X10 is based on four basic principles of asynchrony, locality, atomicity, and order developed on a type-safe, class-based, object-oriented foundation.

MiX10 is a part of the McLab project at SABLE lab that provides a static backend with X10 as the target language for our MATLAB compiler. Two major challenges in translating MATLAB to X10 are 1) Mapping programming constructs of a matrix-based, dynamic and “wild” language to those of a statically typed, object-oriented, parallel programming language and 2) Providing support for a huge number of MATLAB builtin methods that are a major reason behind its success as a scientific programming language. MiX10 builds upon McLab’s Tamer analysis framework and aims for performance and readability of generated X10 code.