June 2010

• Wednesday, 23rd June 2010, 1-2pm (Room 135/26)  *
  Host: Marsha Minchenko
  Title: Using generating functions to count closed walks

  Abstract: The goal here is to present an alternative approach to counting all closed walks in connected regular graphs. Thus, given such a graph, G, we look at extending a set of base walks to the set of all closed walks of G. These chosen base walks and the types of extensions will be described. But more importantly, we explain the use of generating functions to give all closed walks that are extensions of a given closed walk; and explain why this helps reach the desired goal. But are we really counting all closed walks? And is each walk being counted only once?

  
  
• Thursday, 10th June 2010, 1-2pm (Room 115/63)  *
  Host: Fabian Bohnert, Yanir Seroussi
  Title: UMAP Talks

  Talk 1:
  Host: Fabian Bohnert
  Title: A User- and Item-Aware Weighting Scheme for Combining Predictive User Models
  
  Abstract: Hybridising user models can improve predictive accuracy. However, research on linearly combining predictive user models (e.g., used in recommender systems) has often made the implicit assumption that the individual models perform uniformly across the user and item space, using static model weights when computing a weighted average of the predictions of the individual models. This talk proposes a weighting scheme which combines user- and item-specific weight vectors to compute user- and item-aware model weights. The proposed hybridisation approach adaptively estimates online the model parameters that are specific to a target user as information about this user becomes available. Hence, it is particularly well-suited for domains where little or no information regarding the target user's preferences or interests is available at the time of offline model training. The proposed weighting scheme is evaluated by applying it to a real-world scenario from the museum domain. Our results show that in our domain, our hybridisation approach attains a higher predictive accuracy than the individual component models. Additionally, our approach outperforms a non-adaptive hybrid model that uses static model weights.

  Talk 2:
  Host: Yanir Seroussi
  Title: Collaborative Inference of Sentiments from Texts
  
  Abstract: Sentiment analysis deals with inferring people's sentiments and opinions from texts. An important aspect of sentiment analysis is polarity classification, which consists of inferring a document's polarity -- the overall sentiment conveyed by the text -- in the form of a numerical rating. In contrast to existing approaches to polarity classification, we propose to take the authors of the documents into account. Specifically, we present a nearest-neighbour collaborative approach that utilises novel models of user similarity. Our evaluation shows that our approach improves on state-of-the-art performance, and yields insights regarding datasets for which such an improvement is achievable.

May 2010

• Tuesday, 25th May 2010, 1-2pm (Room 115/63)  *
  Host: Daniel Chen-fu Chen
  Title: Multiattribute Decision Support for Eco-Product Form Design
  
  Eco-product design decision-making is often based on vaguely structured information because of the technical and natural attributes as well as inherently conflicting design objectives to be considered. To address this issue, this paper presents a design decision support system to translate consumers’ perception or feeling of an eco-product into design elements, and help product designers understand consumers’ perception. An experimental study on office chairs is conducted, which has identified 7 design elements and 27 representative office chairs as experimental samples for developing neural network (NN) models. NN models are applied to illustrate consumers’ perception on designing eco-product with design element combinations. Fifteen visually perceivable eco-product value (EPV) attributes of eco-products are identified and categorized into aesthetic, functional, and environmental dimensions. A multiattribute decision making (MADM) model is used to generate an eco-product value (EPV) for each eco-product, on which eco-product form design preference ranking is based. It is demonstrated that the MADM model is applicable for environmental issues which are increasing in complexity. This study provides new insights in supporting eco-product form design decision making.
• Tuesday, 11th May 2010, 1-2pm (Room 115/63)  *
  Host: Imran Rao (Uni. Melbourne)
  Title: Asynchronous Aggregation through Gossiping in Large Scale Distributed Systems
  
  Abstract: The goal of my research is to investigate asynchronous gossip-based protocols for computing aggregates in large scale dynamic systems. Aggregation protocols provide global statistical information (e.g. AVG, COUNT, MIN, MAX, etc.) of local attributes of individual nodes in the system. This aggregated information is often useful for other complex applications including network monitoring, load balancing, identifying hot spots, reaching consensus in distributed control and decision applications etc. A typical gossip-based aggregation protocol is described in terms of periodic rounds of a given time interval. In each round, a node exchanges the local information with one of its neighbors chosen at random. Both nodes update their local estimates based on this exchanged information. These local estimates asymptotically converge to a value if the network topology is connected. For an accurate estimate of the true system aggregate, these aggregation protocols crucially rely on a useful property termed as mass conservation. In an AVG aggregation protocol, for example, mass conservation means that at the completion of a round, the sum of estimated averages at all the nodes is always the correct sum of the attributes of all the nodes.
  
  In large scale distributed and dynamic systems, mass is not trivial to conserve. Therefore, to conserve the system mass, existing gossip-based aggregation protocols make either one or all of the following assumptions: (1) underlying system is robust to link and node failures or handles these failures in a transparent manner (2) message communication is reliable and (3) time delays present in the system are globally synchronized among all nodes. For example, it is often assumed that aggregation rounds proceed in a lock-step mode among all nodes. This means, the length of an aggregation round should be large enough so that all nodes in the system complete their r-th round before any node initiates its (r+1) round.
  
  In this paper, we present a new gossip-based asynchronous aggregation protocol. Our protocol preserves the system mass in the presence of asynchronous round lengths, message communication delays and arbitrary node arrival time. Moreover, our protocol neither blocks the execution nor skips any received message -- consequently, it achieves faster convergence. Further, our proposed protocol is also adaptive to system churn. 

April 2010

• Monday, 12th April 2010, 1-2pm (Room 115/63)  *
  Host: SK Rokonuzzaman
  Title: A Cross-layer Approach for QoS Topology Control in Wireless Ad Hoc Networks
  
  Abstract: Wireless ad hoc networks using omni-directional antennas do not scale well due to interference between nearby nodes. Maintaining the QoS of the communications in this type of network is a difficult task. Using multiple narrow beam directional antennas alleviates this problem at the expense of connectivity. Multi-beam Smart Antennas allow the network topology to be adjusted dynamically by adjusting the beamwidth and beam directions to minimize interference and to maximize the number of possible concurrent network communications. This in turn helps to maintain the QoS of the communications. QoS routing has long been used to meet the user requirements by finding appropriate paths to the destinations. We extend this concept to create an Adaptive QoS Topology Control (AQTC) System using Smart Antennas. We use a cross-layer approach to control the topology dynamically where the topology control layer sits between the MAC and the routing protocol. The performance of our protocol has been evaluated using extensive simulations. Simulation results show that different topologies for a set of communications perform differently. AQTC always forms a topology to facilitate the current communications and improves the network throughput and end-to-end delay.
• Thursday, 29th April 2010, 1-2pm (Room 115/63)
  Special seminar session with the "veteran" PhD students panel  *
  Panel speakers: comprised of PhD students who submitted/about to submit their thesis
  

  As part of the CRIS postgrad lunch seminar series, we hosted a panel of PhD students who have submitted their thesis recently or are submitting soon. It was in an open discussion format, with some questions asked by the host and audience (e.g. 'what advice would you give to students who recently commenced their studies?'). Our list of PhD speakers included:

  • Boo Yee Ling
  • Chris Mears
  • Fabian Bohnert
  • Iris Yan
  • Kerri Morgan
  • Michael Wybrow
  • Minh Duc Cao
  • Subrata Chakraborty

March 2010

• Tuesday, 30th March 2010, 1-2pm (115/63)  *
  Host: Regina Bohnert (Friedrich Miescher Laboratory of the Max Planck Society)
  Title: Quantitative Transcriptome Analysis
  
  Abstract: This talk focuses on techniques to quantitatively describe transcriptomes measured with current high-throughput sequencing technologies (RNA-Seq). We have developed and applied machine learning-based methods to the various levels of the RNA-Seq data analysis:
  
  • a) We extended the alignment method QPALMA and combined it with the GenomeMapper short read aligner to align both spliced and unspliced reads with high accuracy, while taking advantage of each read's quality information and computational splice site predictions.
  • b) We advanced methods to de novo predict transcripts based on RNA-Seq data. In particular, we extended the gene finding system mGene by benefit from read alignments to more accurately predict gene structures.
  • c) Moreover, we developed a method, called rQuant, that simultaneously estimates biases inherent in library preparation, sequencing, and read mapping and determines the abundances of given transcripts.
  Our tool inventory allows to study transcriptomes under different conditions or in mutants, leading to a considerably improved understanding of the underlying mechanisms of gene expression and processing.

• Thursday, 18th March 2010, 1-2pm (CEMA Media Room)  *
  Host: Bart Veldstra and Ben Porter
  Title: Generative geometry in Blender
  
  Abstract: This talk presents some recent work into integrating a generative geometry system into the open source 3d modelling package Blender. The generative system, called SDS, is based on biological development and has applications in computer graphics, specifically the automated animation of complex organic shapes that are difficult to model directly. SDS simulates individual cells in a physical environment, with cell division, movement and growth controlled by morphogenetic chemical simulation. It produces a very specific kind of 3d form, and will only be useful to a broad audience if it complements existing tools and methods. Bart Veldstra recently integrated SDS into the popular open source 3d modelling package Blender. The SDS/Blender plugin allows an artist to use Blenders modelling environment to interact with SDS, from setting up the initial environment and simulation parameters, through to importing the results into Blenders animation system.

• Wednesday, 3rd March 2010, 1-2pm (115/63)  *
  Host: Yanir Seroussi
  Title: Utilising User Texts to Improve Recommendations
  
  Abstract: Recommender systems traditionally rely on numeric ratings to represent user opinions, and thus are limited by the single-dimensional nature of such ratings. Recent years have seen an abundance of user-generated texts available online, and advances in natural language processing allow us to better understand users by analysing the texts they write. Specifically, sentiment analysis enables inference of people's sentiments and opinions from texts, while authorship profiling investigates authors' characteristics. We propose to use these techniques to build text-based user models, and incorporate these models into state-of-the-art recommender systems to generate recommendations that are based on a more profound understanding of the users than rating-based recommendations. Our preliminary results suggest that this is a promising direction.

February 2010

• Wednesday, 17th February 2009, 1-2pm (CEMA Media Room)  *
  Host: Marc Cheong
  Title: Wisdom of the Twitterverse: Leveraging the knowledge, latent demography, and user habits within the Twitter ecosystem
  
  Abstract: This presentation is a part of Marc's PhD confirmation. In it, he'll detail the background information and current work involved in studying the Twitter 'ecosystem'. He proposes that Twitter could be segmented into two discrete domains: the user domain and the message domain; and how existing research does not emphasize the interdependence and interrelation between the two domains. He also details several publications and research contributions of his PhD candidature to date, and other potentially important research areas arising from this line of research.

• Welcome back everyone!
  Wednesday, 10th February 2009, 1-2pm (115/63)  *
  Title: TED talks: "Ideas worth spreading"
  
  Please come to the welcome back lunch for all IT postgraduate students, old and new, for the new decade!
  With lunch provided by the Clayton School of IT and CRIS, the seminar series is a great way to catch up with your fellow students and hear about the latest research.
  To start off the year we'll be watching some TED lectures!