Learning Nonlinear Visual Processing from Natural Images

The paradigm of computational vision hypothesizes that any visual function -- such as the recognition of your grandparent -- can be replicated by computational processing of the visual input. What are these computations that the brain performs? What should or could they be? Working on the latter question, this dissertation takes the statistical approach, where the suitable computations are attempted to be learned from the natural visual data itself. In particular, we empirically study the computational processing that emerges from the statistical properties of the visual world and the constraints and objectives specified for the learning process. This thesis consists of an introduction and 7 peer-reviewed publications, where the purpose of the introduction is to illustrate the area of study to a reader who is not familiar with computational vision research. In the scope of the introduction, we will briefly overview the primary challenges to visual processing, as well as recall some of the current opinions on visual processing in the early visual systems of animals. Next, we describe the methodology we have used in our research, and discuss the presented results. We have included some additional remarks, speculations and conclusions to this discussion that were not featured in the original publications. We present the following results in the publications of this thesis. First, we empirically demonstrate that luminance and contrast are strongly dependent in natural images, contradicting previous theories suggesting that luminance and contrast were processed separately in natural systems due to their independence in the visual data. Second, we show that simple cell -like receptive fields of the primary visual cortex can be learned in the nonlinear contrast domain by maximization of independence. Further, we provide first-time reports of the emergence of conjunctive (corner-detecting) and subtractive (opponent orientation) processing due to nonlinear projection pursuit with simple objective functions related to sparseness and response energy optimization. Then, we show that attempting to extract independent components of nonlinear histogram statistics of a biologically plausible representation leads to projection directions that appear to differentiate between visual contexts. Such processing might be applicable for priming, \ie the selection and tuning of later visual processing. We continue by showing that a different kind of thresholded low-frequency priming can be learned and used to make object detection faster with little loss in accuracy. Finally, we show that in a computational object detection setting, nonlinearly gain-controlled visual features of medium complexity can be acquired sequentially as images are encountered and discarded. We present two online algorithms to perform this feature selection, and propose the idea that for artificial systems, some processing mechanisms could be selectable from the environment without optimizing the mechanisms themselves. In summary, this thesis explores learning visual processing on several levels. The learning can be understood as interplay of input data, model structures, learning objectives, and estimation algorithms. The presented work adds to the growing body of evidence showing that statistical methods can be used to acquire intuitively meaningful visual processing mechanisms. The work also presents some predictions and ideas regarding biological visual processing.

Computational Identification of Recessive Mutations in Cancers using High Throughput SNP-arrays

This thesis presents a highly sensitive genome wide search method for recessive mutations. The method is suitable for distantly related samples that are divided into phenotype positives and negatives. High throughput genotype arrays are used to identify and compare homozygous regions between the cohorts. The method is demonstrated by comparing colorectal cancer patients against unaffected references. The objective is to find homozygous regions and alleles that are more common in cancer patients. We have designed and implemented software tools to automate the data analysis from genotypes to lists of candidate genes and to their properties. The programs have been designed in respect to a pipeline architecture that allows their integration to other programs such as biological databases and copy number analysis tools. The integration of the tools is crucial as the genome wide analysis of the cohort differences produces many candidate regions not related to the studied phenotype. CohortComparator is a genotype comparison tool that detects homozygous regions and compares their loci and allele constitutions between two sets of samples. The data is visualised in chromosome specific graphs illustrating the homozygous regions and alleles of each sample. The genomic regions that may harbour recessive mutations are emphasised with different colours and a scoring scheme is given for these regions. The detection of homozygous regions, cohort comparisons and result annotations are all subjected to presumptions many of which have been parameterized in our programs. The effect of these parameters and the suitable scope of the methods have been evaluated. Samples with different resolutions can be balanced with the genotype estimates of their haplotypes and they can be used within the same study.

Computation of confined swirling flow using RNG and k-epsilon turbulence models

In this work we numerically model isothermal turbulent swirling flow in a cylindrical burner. Three versions of the RNG k-epsilon model are assessed against performance of the standard k-epsilon model. Sensitivity of numerical predictions to grid refinement, differing convective differencing schemes and choice of (unknown) inlet dissipation rate, were closely scrutinised to ensure accuracy. Particular attention is paid to modelling the inlet conditions to within the range of uncertainty of the experimental data, as model predictions proved to be significantly sensitive to relatively small changes in upstream flow conditions. We also examine the characteristics of the swirl--induced recirculation zone predicted by the models over an extended range of inlet conditions. Our main findings are: - (i) the standard k-epsilon model performed best compared with experiment; - (ii) no one inlet specification can simultaneously optimize the performance of the models considered; - (iii) the RNG models predict both single-cell and double-cell IRZ characteristics, the latter both with and without additional internal stagnation points. The first finding indicates that the examined RNG modifications to the standard k-e model do not result in an improved eddy viscosity based model for the prediction of swirl flows. The second finding suggests that tuning established models for optimal performance in swirl flows a priori is not straightforward. The third finding indicates that the RNG based models exhibit a greater variety of structural behaviour, despite being of the same level of complexity as the standard k-e model. The plausibility of the predicted IRZ features are discussed in terms of known vortex breakdown phenomena.

Computational analyses of the catalytic and heparin-binding sites and their interactions with glycosaminoglycans in glycoside hydrolase family 79 endo-d-glucuronidase (heparanase)

Mammalian heparanase is an endo-β-glucuronidase associated with cell invasion in cancer metastasis, angiogenesis and inflammation. Heparanase cleaves heparan sulfate proteoglycans in the extracellular matrix and basement membrane, releasing heparin/heparan sulfate oligosaccharides of appreciable size. This in turn causes the release of growth factors, which accelerate tumor growth and metastasis. Heparanase has two glycosaminoglycan-binding domains; however, no three-dimensional structure information is available for human heparanase that can provide insights into how the two domains interact to degrade heparin fragments. We have constructed a new homology model of heparanase that takes into account the most recent structural and bioinformatics data available. Heparin analogs and glycosaminoglycan mimetics were computationally docked into the active site with energetically stable ring conformations and their interaction energies were compared. The resulting docked structures were used to propose a model for substrates and conformer selectivity based on the dimensions of the active site. The docking of substrates and inhibitors indicates the existence of a large binding site extending at least two saccharide units beyond the cleavage site (toward the nonreducing end) and at least three saccharides toward the reducing end (toward heparin-binding site 2). The docking of substrates suggests that heparanase recognizes the N-sulfated and O-sulfated glucosamines at subsite +1 and glucuronic acid at the cleavage site, whereas in the absence of 6-O-sulfation in glucosamine, glucuronic acid is docked at subsite +2. These findings will help us to focus on the rational design of heparanase-inhibiting molecules for anticancer drug development by targeting the two heparin/heparan sulfate recognition domains.

Computational methods for the prediction of the structure and interactions of coiled-coil peptides

The past several years have seen significant advances in the development of computational methods for the prediction of the structure and interactions of coiled-coil peptides. These methods are generally based on pairwise correlations of amino acids, helical propensity, thermal melts and the energetics of sidechain interactions, as well as statistical patterns based on Hidden Markov Model (HMM) and Support Vector Machine (SVM) techniques. These methods are complemented by a number of public databases that contain sequences, motifs, domains and other details of coiled-coil structures identified by various algorithms. Some of these computational methods have been developed to make predictions of coiled-coil structure on the basis of sequence information; however, structural predictions of the oligomerisation state of these peptides still remains largely an open question due to the dynamic behaviour of these molecules. This review focuses on existing in silico methods for the prediction of coiled-coil peptides of functional importance using sequence and/or three-dimensional structural data.

Complexity of collaborative work in residential aged care facilities: An analysis of information exchange for medication management

Information exchange (IE) is a critical component of the complex collaborative medication process in residential aged care facilities (RACFs). Designing information and communication technology (ICT) to support complex processes requires a profound understanding of the IE that underpins their execution. There is little existing research that investigates the complexity of IE in RACFs and its impact on ICT design. The aim of this study was thus to undertake an in-depth exploration of the IE process involved in medication management to identify its implications for the design of ICT. The study was undertaken at a large metropolitan facility in NSW, Australia. A total of three focus groups, eleven interviews and two observation sessions were conducted between July to August 2010. Process modelling was undertaken by translating the qualitative data via in-depth iterative inductive analysis. The findings highlight the complexity and collaborative nature of IE in RACF medication management. These models emphasize the need to: a) deal with temporal complexity; b) rely on an interdependent set of coordinative artefacts; and c) use synchronous communication channels for coordination. Taken together these are crucial aspects of the IE process in RACF medication management that need to be catered for when designing ICT in this critical area. This study provides important new evidence of the advantages of viewing process as a part of a system rather than as segregated tasks as a means of identifying the latent requirements for ICT design and that is able to support complex collaborative processes like medication management in RACFs. © 2012 IEEE.

Negotiation schemes for multi-agent cooperative search

Because of limited sensor and communication ranges, designing efficient mechanisms for cooperative tasks is difficult. In this article, several negotiation schemes for multiple agents performing a cooperative task are presented. The negotiation schemes provide suboptimal solutions, but have attractive features of fast decision-making, and scalability to large number of agents without increasing the complexity of the algorithm. A software agent architecture of the decision-making process is also presented. The effect of the magnitude of information flow during the negotiation process is studied by using different models of the negotiation scheme. The performance of the various negotiation schemes, using different information structures, is studied based on the uncertainty reduction achieved for a specified number of search steps. The negotiation schemes perform comparable to that of optimal strategy in terms of uncertainty reduction and also require very low computational time, similar to 7 per cent to that of optimal strategy. Finally, analysis on computational and communication requirement for the negotiation schemes is carried out.

Situated interpretation in computational creativity

This paper describes, formalizes and implements an approach to computational creativity based on situated interpretation. The paper introduces the notions of framing and reframing of conceptual spaces based on empirical studies as the driver for this research. It uses concepts from situated cognition, and situated interpretation in particular, to be the basis of a formal model of the movement between conceptual spaces. This model is implemented using rules within interacting neural networks. This implementation demonstrates behaviour similar to that observed in studies of human designers.

Two highly unsymmetrical tetradentate (N3O) Schiff base copper(II) complexes: Template synthesis, structural characterization, magnetic and computational studies

Two new copper(II) complexes, [Cu-2(L-1)(2)](ClO4)(2) (1) and [Cu(L-2)(ClO4)] (2), of the highly unsymmetrical tetradentate (N3O) Schiff base ligands HL1 and HL2 (where HL1 = N-(2-hydroxyacetophenone)-bis-3-aminopropylamine and HL2 = N-(salicyldehydine)-bis-3-aminopropylamine) have been synthesised using a template method. Their single crystal X-ray structures show that in complex 1 two independent copper(II) centers are doubly bridged through sphenoxo-O atoms (O1A and O1B) of the two ligands and each copper atom is five-coordinated with a distorted square pyramidal geometry. The asymmetric unit of complex 2 consists of two crystallographically independe N-(salicylidene) bis(aminopropyl)amine-copper(II) molecules, A and B, with similar square pyramidal geometries. Cryomagnetic susceptibility measurements (5-300 K) on complex 1 reveal a distinct antiferromagnetic interaction with J=-23.6 cm(-1), which is substantiated by a DFT calculation (J=-27.6 cm(-1)) using the B3LYP functional. Complex 1, immobilized over highly ordered hexagonal mesoporous silica, shows moderate catalytic activity for the epoxidation of cyclohexene and styrene in the presence of TBHP as an oxidant.

Computational systems approach for drug target discovery

Importance of the field: The shift in focus from ligand based design approaches to target based discovery over the last two to three decades has been a major milestone in drug discovery research. Currently, it is witnessing another major paradigm shift by leaning towards the holistic systems based approaches rather the reductionist single molecule based methods. The effect of this new trend is likely to be felt strongly in terms of new strategies for therapeutic intervention, new targets individually and in combinations, and design of specific and safer drugs. Computational modeling and simulation form important constituents of new-age biology because they are essential to comprehend the large-scale data generated by high-throughput experiments and to generate hypotheses, which are typically iterated with experimental validation. Areas covered in this review: This review focuses on the repertoire of systems-level computational approaches currently available for target identification. The review starts with a discussion on levels of abstraction of biological systems and describes different modeling methodologies that are available for this purpose. The review then focuses on how such modeling and simulations can be applied for drug target discovery. Finally, it discusses methods for studying other important issues such as understanding targetability, identifying target combinations and predicting drug resistance, and considering them during the target identification stage itself. What the reader will gain: The reader will get an account of the various approaches for target discovery and the need for systems approaches, followed by an overview of the different modeling and simulation approaches that have been developed. An idea of the promise and limitations of the various approaches and perspectives for future development will also be obtained. Take home message: Systems thinking has now come of age enabling a `bird's eye view' of the biological systems under study, at the same time allowing us to `zoom in', where necessary, for a detailed description of individual components. A number of different methods available for computational modeling and simulation of biological systems can be used effectively for drug target discovery.

Low ML-Decoding Complexity, Large Coding Gain, Full-Rate, Full-Diversity STBCs for 2 x 2 and 4 x 2 MIMO Systems

This paper deals with low maximum-likelihood (ML)-decoding complexity, full-rate and full-diversity space-time block codes (STBCs), which also offer large coding gain, for the 2 transmit antenna, 2 receive antenna (2 x 2) and the 4 transmit antenna, 2 receive antenna (4 x 2) MIMO systems. Presently, the best known STBC for the 2 2 system is the Golden code and that for the 4 x 2 system is the DjABBA code. Following the approach by Biglieri, Hong, and Viterbo, a new STBC is presented in this paper for the 2 x 2 system. This code matches the Golden code in performance and ML-decoding complexity for square QAM constellations while it has lower ML-decoding complexity with the same performance for non-rectangular QAM constellations. This code is also shown to be information-lossless and diversity-multiplexing gain (DMG) tradeoff optimal. This design procedure is then extended to the 4 x 2 system and a code, which outperforms the DjABBA code for QAM constellations with lower ML-decoding complexity, is presented. So far, the Golden code has been reported to have an ML-decoding complexity of the order of for square QAM of size. In this paper, a scheme that reduces its ML-decoding complexity to M-2 root M is presented.

High-Rate Space-Time Coded Large-MIMO Systems: Low-Complexity Detection and Channel Estimation

In this paper, we present a low-complexity algorithm for detection in high-rate, non-orthogonal space-time block coded (STBC) large-multiple-input multiple-output (MIMO) systems that achieve high spectral efficiencies of the order of tens of bps/Hz. We also present a training-based iterative detection/channel estimation scheme for such large STBC MIMO systems. Our simulation results show that excellent bit error rate and nearness-to-capacity performance are achieved by the proposed multistage likelihood ascent search (M-LAS) detector in conjunction with the proposed iterative detection/channel estimation scheme at low complexities. The fact that we could show such good results for large STBCs like 16 X 16 and 32 X 32 STBCs from Cyclic Division Algebras (CDA) operating at spectral efficiencies in excess of 20 bps/Hz (even after accounting for the overheads meant for pilot based training for channel estimation and turbo coding) establishes the effectiveness of the proposed detector and channel estimator. We decode perfect codes of large dimensions using the proposed detector. With the feasibility of such a low-complexity detection/channel estimation scheme, large-MIMO systems with tens of antennas operating at several tens of bps/Hz spectral efficiencies can become practical, enabling interesting high data rate wireless applications.

Room temperature metathesis of aryl isocyanates and aromatic aldehydes catalyzed by group(IV) metal alkoxides: An experimental and computational study

Aromatic aldehydes and aryl isocyanates do not react at room temperature. However, we have shown for the first time that in the presence of catalytic amounts of group(IV) n-butoxide, they undergo metathesis at room temperature to produce imines with the extrusion of carbon dioxide. The mechanism of action has been investigated by a study of stoichiometric reactions. The insertion of aryl isocyanates into the metal n-butoxide occurs very rapidly. Reaction of the insertion product with the aldehyde is responsible for the metathesis. Among the n-butoxides of group(IV) metals, Ti((OBu)-Bu-n)(4) (8aTi) was found to be more efficient than Zr((OBu)-Bu-n)(4) (8aZr) and Hf((OBu)-Bu-n)(4) (8aHf) in carrying out metathesis. The surprisingly large difference in the metathetic activity of these alkoxides has been probed computationally using model complexes Ti(OMe)(4) (8bTi), Zr(OMe)(4) (8bZr) and Hf(OMe)(4) (8bHf) at the B3LYP/LANL2DZ level of theory. These studies indicate that the insertion product formed by Zr and Hf are extremely stable compared to that formed by Ti. This makes subsequent reaction of Zr and Hf complexes unfavorable.

A Low-Complexity Near-ML Performance Achieving Algorithm for Large MIMO Detection

In this paper, we present a low-complexity, near maximum-likelihood (ML) performance achieving detector for large MIMO systems having tens of transmit and receive antennas. Such large MIMO systems are of interest because of the high spectral efficiencies possible in such systems. The proposed detection algorithm, termed as multistage likelihood-ascent search (M-LAS) algorithm, is rooted in Hopfield neural networks, and is shown to possess excellent performance as well as complexity attributes. In terms of performance, in a 64 x 64 V-BLAST system with 4-QAM, the proposed algorithm achieves an uncoded BER of 10(-3) at an SNR of just about 1 dB away from AWGN-only SISO performance given by Q(root SNR). In terms of coded BER, with a rate-3/4 turbo code at a spectral efficiency of 96 bps/Hz the algorithm performs close to within about 4.5 dB from theoretical capacity, which is remarkable in terms of both high spectral efficiency as well as nearness to theoretical capacity. Our simulation results show that the above performance is achieved with a complexity of just O(NtNt) per symbol, where N-t and N-tau denote the number of transmit and receive antennas.

Algebraic distributed space-time codes with low ML decoding complexity

"Extended Clifford algebras" are introduced as a means to obtain low ML decoding complexity space-time block codes. Using left regular matrix representations of two specific classes of extended Clifford algebras, two systematic algebraic constructions of full diversity Distributed Space-Time Codes (DSTCs) are provided for any power of two number of relays. The left regular matrix representation has been shown to naturally result in space-time codes meeting the additional constraints required for DSTCs. The DSTCs so constructed have the salient feature of reduced Maximum Likelihood (ML) decoding complexity. In particular, the ML decoding of these codes can be performed by applying the lattice decoder algorithm on a lattice of four times lesser dimension than what is required in general. Moreover these codes have a uniform distribution of power among the relays and in time, thus leading to a low Peak to Average Power Ratio at the relays.