This page contains the index for the Technical Reports Series of the Computer Science Department of the University of Otago.
All documents are in pdf format.
The documents have not been subjected to peer review and the department is not responsible for the contents of the papers. Some papers may be removed completely later, replaced by more up-to-date versions in due time, or may be removed for copyright reasons after publication in journals, in which case they will keep their position in the technical reports series but only as a pointer to that journal.
Submissions page (for Computer Science Staff only)
The archive is maintained by Robert Pollock.
Technical Reports and Theses Index
2019
OUCS-2019-03
Matthew Jenkin, Steven Mills, David Eyers
Department of Computer Science, University of Otago
Git Badges: We’re not ready to commit yet
Abstract: Mastering the use of version-control systems (VCS) is a crucial skill for efficiently working with software. Undergraduate students should thus learn how to use a VCS early in their university studies, if they have not already done so. However, an introduction to the use of VCSs is not a good topic for academic teaching, when the VCS is just being used as a tool. Micro-credentials or badges show promise as a mechanism for lifting teaching of skills such as use of VCSs out of the flow of the academic curriculum, while supporting students appropriately. This paper presents our preparatory work to form a lesson on the git VCS, as part of a larger study into the use of a form of micro-credentials for skills-based teaching. We survey a number of popular, online resources for git education, contrasting their pedagogical approaches. We note key considerations for providing students a grounding in git skills that is efficient for their learning, at multiple levels of detail.
OUCS-2019-02
Anthony Robins
Department of Computer Science, University of Otago
Novice programmers and introductory programming
Abstract: One of the central topics in computing education research (CEdR) is the exploration of how a person learns their first programming language, also described in terms such as understanding "novice programmers", introductory programming, teaching and learning in "CS1" (a first course in computer science), and so on. This chapter explores key issues and surveys some of the important research in this domain.
2018
OUCS-2018-04
Lech Szymanski, Chris Gorman, Alistair Knott, Brendan McCane
Department of Computer Science, University of Otago
Martin Takac
Comenius University, Bratislava, Slovakia
On Learning Object Properties in Convolutional Neural Networks via an Inhibition of Return (IOR) Mechanism
Abstract:This is a report of an investigation into suitability of the visual features produced by a Convolutional Neural Network (CNN) for recognition of objects and their properties via an Inhibition of Return (IOR) mechanism. We train a linear model on visual features using Conditional Principal Component Analysis (CPCA) rule with a simple IOR to produce a sequence of activity in response to a single input stimulus. After training on a synthetic/localist representation of the visual features, the model recognises objects and pays attention to it most unusual properties. However, the same IOR mechanism does not reliably prioritise recognition of the object type nor its most unusual attribute after training on CNN features derived from real images. Our main conclusion is that the distributed representation of the CNN features is not suitable for the proposed IOR.
OUCS-2018-03
Xiaogang Yan, Alistair Knott, Steven Mills
Department of Computer Science, University of Otago
A neural network model for learning to represent 3D objects via tactile exploration: technical appendix
Abstract:
This technical report is to present a neural network model for learning to represent 3D objects via tactile exploration, which complements a paper named "a neural network model for learning to represent 3D objects via tactile exploration" and published in the conference CogSci 2018. This report includes the details of the proposed model.
OUCS-2018-02
Anthony Robins
Department of Computer Science, University of Otago
Outcomes in introductory programming
Abstract:
This technical report is best thought of as an Appendix to my chapter on introductory programming in The Cambridge Handbook of Computing Education Research (Robins, 2018). It is intended to evolve into further discussion of the issue of outcomes in introductory programming courses (and for novice programmers in general).
OUCS-2018-01
Hayim Dar and Alistair Knott
Department of Computer Science, University of Otago
Martin Takac
Centre for Cognitive Science, Comenius University, Bratislava, Slovak Republic
Learning and representing the spatial properties of objects via tactile exploration
Abstract:
Neural representations of physical space must be learned through physical exploration, since only experience of this kind can give meaning to the visual markers of three dimensional structure. Mammalian brains represent navigable environmental space via the activity of specialised cells, most notably place, grid and boundary cells. Moreover, there are similar neural responses to hypothetically navigable object surfaces as to actually navigable environments, amongst other parallels. We therefore propose a dead-reckoning model of object-centered spatial learning, as an analogue to environmental spatial learning in hippocampal and related cortices. Based upon the trajectory-abstraction model of Takac and Knott (2017), our model successfully generates canonical, single location place cells, as well as a variety of more complex spatial receptive elds, from which locations on object surfaces may be reconstructed. We propose that tactile exploration is the natural analogue to environment navigation, and predict that hippocampus-like spatial representations should be found in parietal tactile pathways as well. Looking ahead, our proposal suggests an elegant duality in the representation of objects and spaces, upon which may be built a powerful and rich representation of the compound and fractal spatial structures in which animals live.
2017
OUCS-2017-07
Richard O'Keefe
Department of Computer Science, University of Otago
Complex Arithmetic is Complex
Abstract:
Floating-point arithmetic is tricky. This report examines the apparently simple case of basic complex arithmetic in order to illustrate this.
OUCS-2017-06
Richard O'Keefe
Department of Computer Science, University of Otago
Logic Programming Modules as Possible Worlds
Abstract:
Existing module systems for logic programming, such as ISO Prolog part 2 or Mercury are basically syntactic in nature. They are intended to be compatible with the usual semantics of logic programming. Taking a semantic view leads to the idea of modules as possible worlds, and cross- module calling as a modal operator. This opens up additional possibilities, including some useful for software engineering.
OUCS-2017-05
Richard O'Keefe
Department of Computer Science, University of Otago
Child Modules for Erlang and Prolog
Abstract:
Prolog and Erlang have similar module systems, where modules in a flat namespace are both the sole form of encapsulation and the units of code loading and replacement. They also support the inclusion of text files as a way of sharing declarations and private functions between modules. This note proposes a replacement for text inclusion inspired by child modules in Ada.
OUCS-2017-04
Alistair Knott
Department of Computer Science, University of Otago
Sensorimotor cognition and natural language syntax Part II
Abstract:
This book is a continuation of the idea I developed in my earlier book, 'Sensorimotor Cognition and Natural Language Syntax' (Knott, 2010). In that book, I suggested that the syntactic structure of a sentence reporting a concrete episode could be interpreted as a description of sensorimotor processing. I expressed this idea using the syntactic framework of Minimalism (Chomsky, 1995), in which every sentence has two syntactic representations: a phonetic form (PF) and an underlying logical form (LF). My proposal was that the LF of a sentence S reporting a concrete episode E can be characterised as a description of the sensorimotor processes involved in actually experiencing the episode E. In the earlier book, I focussed on a single syntactic construction (a transitive clause) when presenting and motivating this proposal. Obviously I must consider a wider range of constructions. In the current book I examine how the original proposal extends to other syntactic constructions.
OUCS-2017-03
Alistair Knott, Lech Szymanski, Brendan McCane
Department of Computer Science, University of Otago
Martin Takac
Centre for Cognitive Science, Comenius University, Slovakia
A model of object property representations: visual object classification, working memory and the syntax of predication
OUCS-2017-02
Alistair Knott
Department of Computer Science, University of Otago
An extended model of deictic routines, supporting a wider-coverage SM interpretation of syntax
OUCS-2017-01
Alistair Knott
Department of Computer Science, University of Otago
Road map for an embodied model of language and cognition
2016
OUCS-2016-04
Alistair Knott
Department of Computer Science, University of Otago
Martin Takac
Centre for Cognitive Science, Comenius University, Slovakia
A simulationist model of episode representations in working memory: syntactic interpretation, nested episodes and storage requirements
Abstract:
This report supplements Takac and Knott's model of working memory (WM) for episodes and individuals. In Section 1 we introduce our interpretation of syntactic heads in relation to this WM model in more detail. In Section 2 we present some ideas about how the WM model can be extended to handle nested episodes. In Section 3 we discuss the storage requirements of the model, and assess whether it can be extended to represent a realistic number of episodes and individuals.
OUCS-2016-03
Richard O'Keefe
Department of Computer Science, University of Otago
How to compute a mean?
Abstract:
Computing the mean of a sequence of numbers is easy; computing means of other kinds of measurements, and for other kinds of container, is harder. This note presents some algorithms and benchmarks.
OUCS-2016-02
Lahiru Ariyasinghe, Zhiyi Huang, David Eyers
Department of Computer Science, University of Otago
The Impact of IP Network Impairments on Optimal Playback Buffer Size in Video Streaming
Abstract:
A key challenge for online video streaming services is how to deliver their data over networks that suffer packet losses and delays while maintaining a good Quality of Experience (QoE). Metrics such as start-up delay, the count of the number of times that re-buffering occurs and re-buffering delays provide useful indicators to the streaming services to measure the impact of IP network impairments (e.g. packet loss and delay) on overall video stream quality. Playback buffering is one of the key application-level techniques that can mitigate the impact of network impairments and protect the video quality by sensibly balancing the effect of the above metrics. However for the mitigation to be effective, while maximising user QoE, setting an optimal playback buffer size is vital. In this paper, a comprehensive analysis is performed to investigate the impact of packet loss and link delay on optimal playback buffer sizing, with respect to video files that contain different amounts of motion. Experimental results indicate that the buffer size that delivers an optimal start-up delay with acceptable levels of playback disruption remains reasonably constant, when changing the degree of motion in the video, and in response to minor variation of link delay and small amounts of packet loss. This is in contrast to the buffer size that ensures no interruptions to playback, which, as expected, rises steadily.
2014
OUCS-2014-04
Glenn Blanchette, Brendan McCane, Willem Labuschagne and Anthony Robins
Department of Computer Science, University of Otago
Representing Symbolic Logic in an Artifical Neural Network
Abstract:
We report detailed experimental results for the paper 'Representing Symbolic Logic in an Artificial Neural Network, Part I: the Static Case'.
OUCS-2014-02
Lech Szymanski and David Eyers
Department of Computer Science, University of Otago
Practical use of SELinux for enhancing the security of web applications
Note: If readers wish to run through examples, the report is also available here. This has links to Virtual Machine files so that people working through the tutorial can choose to start off at any chapter.
Abstract:
The Security-Enhanced Linux (SELinux) module has been available in the mainline kernel for many years (since 2003), and is included as part of a growing number of popular Linux distributions. However adopting its new and powerful security capabilities has been daunting to many. On forums, the advice regarding many SELinux related issues is simply, 'Disable SELinux'. The earlier sections of this document should provide enough practical SELinux context to avoid the need for such blunt solutions. The later sections will demonstrate how to write your own SELinux policy to enhance the security of a web-based content-management system. This tutorial came about as a result of an investigation into the viability of using SELinux to secure multi-tier web systems that process sensitive data, as inspired by the SafeWeb project.
OUCS-2014-01
Jeremy Lee-Hand and Alistair Knott
Department of Computer Science, University of Otago
A neural network model of causative actions
Abstract:
In this paper we present a neural network model of motor learning and motor control that learns a class of actions termed causative actions. A causative action is an action that brings about a speci ed e ect or movement in a target object: for instance the action of causing a lever to bend, or of causing a door to open. The network comprises a layered set of motor learning circuits that associate motor actions with their sensory effects, as proposed by Hommel et al. [1]. The circuit for learning simple manual actions is trained by sensory representations in the haptic modality that function as rewards, as previously proposed by Oztop and Arbib [2]. We propose that the circuit responsible for learning causative actions makes similar use of sensory representations as reward signals. The key novel idea is that the sensory representations in this case come not from the tactile system, but from a high-level perceptual module that registers arbitrary movements taking place in external objects in the world.
2013
OUCS-2013-13
Adeel Javed, Haibo Zhang, and Zhiyi Huang
Department of Computer Science, University of Otago
Jeremiah Deng
Department of Information Science, University of Otago
BWS: Beacon-driven Wake-up Scheme for Train Localization using Wireless Sensor Networks
Abstract:
Real-time train localization using wireless sensor networks (WSNs) offers huge benefits in terms of cost reduction and safety enhancement in railway environments. A challenging problem in WSN-based train localization is how to guarantee timely communication between the anchor sensors deployed along the track and the gateway deployed on the train with minimum energy consumption. This paper presents an energyefficient scheme for timely communication between the gateway and the anchor sensors, in which each anchor sensor runs an asynchronous duty-cycling protocol to conserve energy and wakes up only when it goes into the communication range of the gateway on the train. A beacon-driven wake-up scheme is designed and we establish the upper bound on the amount of time that an anchor sensor can sleep in one duty cycle to guarantee timely wake up once a train approaches. We also give a thorough theoretical analysis for the energy efficiency of our scheme and give the optimal amount of time that an anchor sensor should sleep in terms of minimizing the total energy consumption at each anchor sensor. We evaluate the performance of our scheme through simulations. Simulation results show that our scheme can timely wake up anchors sensors at a very low cost on energy consumption.
OUCS-2013-12
Paul McCarthy, Lubica Benuskova
Department of Computer Science, University of Otago
Elizabeth Franz
Department of Psychology, University of Otago
Functional network analysis of aging and Alzheimer's Disease: Results
Abstract:
This report, and the associated data files, present the results of a functional network analysis conducted upon a fMRI data set, with the aim of identifying differences, in various functional network properties, between a group of healthy young individuals, a group of healthy aged individuals, and a group of individuals diagnosed with probable to mild Alzheimer's Disease. This report is intended as a complement to the PhD thesis Functional network analysis of aging and Alzheimer's Disease, by Paul McCarthy, submitted for the degree of Doctor of Philosophy, at the University of Otago, Dunedin, New Zealand.
These data files may be used for research purposes. Please cite this report if you do so.
Supporting data files:
connectivity_analysis.tbz2 (3.8MB)
global_analysis.tbz2 (535KB)
network_data_awith.tbz2 (473MB)
network_data_awout.tbz2 (551MB)
network_data_young.tbz2 (576MB)
networks_awith.tbz2 (2.2GB)
networks_awout.tbz2 (2.5GB)
networks_young.tbz2 (2.5GB)
regional_analysis.tbz2 (7.3MB)
voxelwise_analysis.tbz2 (492MB)
If you want these data files, please email Robert Pollock - rpollock@cs.otago.ac.nz - for a download link.
OUCS-2013-11
Jeremy Lee-Hand and Alistair Knott
Department of Computer Science, University of Otago
Training and testing of a neural network model of motor control
Abstract:
This paper is an appendix to Lee-Hand and Knott (in submission) - A neural network model of causative motor actions and causative alternation. It describes the training and testing of the network model of motor control presented in that paper.
OUCS-2013-10
Mira Guise, Alistair Knott, Lubica Benuskova
Department of Computer Science, University of Otago
Response Fingerprinting: a probabilistic method for evaluating the network response to stimuli
Abstract:
Spiking neural networks that have variable connection delays have the interesting property that they are sensitive to both spatial and temporal patterns of input. Each neuron in the network receives input from spatially-distributed input neurons whose precise firing times interact with connection delays to determine whether the neuron can exceed the firing threshold and produce an output. The network is therefore most responsive to spatio-temporal stimuli whose temporal components match the delays in the connected structure of the network. Izhikevich (2006a) has shown that certain strongly connected groups of neurons known as polychronous neural groups (or PNGs) exist in large numbers within the network structure. The activation of these neural groups is stimulus-specific and produces unique firing signatures that are detectable in the firing data. Previous methods for detecting PNG activation have relied on a template matching technique that assumes a deterministic response to each stimulus presentation (Izhikevich, 2006a; Martinez and Paugam-Moisy, 2009; Guise et al., 2013a). Here we present an alternative probabilistic view of the stimulus response and demonstrate the application of this new detection method.
OUCS-2013-09
Note: updated 13.9.13
Hayden Walles, Anthony Robins and Alistair Knott
Department of Computer Science, University of Otago
A neural network model of visual attention and object classification: technical details
Abstract:
This document provides technical details of our neural network model of visual object classification and attention.
OUCS-2013-08
Mira Guise, Alistair Knott, Lubica Benuskova
Department of Computer Science, University of Otago
Consistency of polychronous neural group activation supports a role as an underlying mechanism for representation and memory: detailed methods and results
Abstract:
Izhikevich (2006a) has proposed that certain strongly connected groups of neurons known as polychronous neural groups (or PNGs) might provide the neural basis for representation and memory. Polychronous groups exist in large numbers within the connection graph of a spiking neural network, providing a large repertoire of structures that can potentially match an external stimulus (Izhikevich, 2006a; Izhikevich et al., 2004). In this paper we examine some of the requirements of a representational system and test the idea of PNGs as the underlying mechanism against one of these requirements, the requirement for consistency in the neural response to stimuli. The results provide preliminary evidence for consistency of PNG activation in response to known stimuli, although these results are limited by problems with the current methods for detecting PNG activation.
OUCS-2013-07
Not available: to be published
Yawen Chen, Jason Mair, Zhiyi Huang, David Eyers
Department of Computer Science, University of Otago
A State-based Energy/Performance Model for A Parallel Application on Multicore Computers
Abstract:
In this paper, we design a state-based energy/ performance model for a given parallel application on multicore computer systems. Given the application properties of parallel degree and computation intensity, and the system energy features, we derive the optimal number of cores and the optimal frequencies for the application to achieve the minimum energy consumption. By estimating and quantifying the energy cost for each component in different states, the impact of energy costs for serial/parallel portions and computation/memory portions are revealed when different number of cores and CPU frequencies are assigned. Our proposed model provides an insight on estimating the energy/performance impact of an application on multicore computers, and provides the guidance for achieving tradeoffs between performance and energy consumption.
OUCS-2013-06
Not available: to be published
Yawen Chen, Haibo Zhang
Department of Computer Science, University of Otago
Ke Chen, Huaxi Gu
State Key Laboratory of Integrated Service Networks, Xidian University, Xi'an, China
TWC-based approach for Improving Communication Reliability in Optical Network-on-chip
Abstract:
Optical Network-on-Chip (ONoC) architectures are emerging as a new paradigm to interconnect a large number of processing cores at chip level, thereby enabling to meet the pressing demands for extremely high bandwidth and low power consumption. Some existing ONoC architectures are implemented in an electronic-controlled way in a two-layer 3D chip based on TSV (Through-Silicon-Via). However, on-chip thermal effect is an inherent deficiency and chip temperature can fluctuate spatially, which can affect the operation of silicon nanophotonic devices. This leads to significant influence on the reliability of communications. To minimize the thermal impacts and improve reliability, we propose a novel approach by adding a tunable wavelength converter (TWC) in the router for mesh-based ONoC. Simulation results show that our proposed approach can increase the signal-to-noise ratio (SNR) effectively.
OUCS-2013-05
Not available: to be published
Yawen Chen, Haibo Zhang
Department of Computer Science, University of Otago
Liu Bai, Huaxi Gu
Xidian University, China
A Hierarchical Hybrid Optical-Electronic Clos Architecture for Network-on-Chip
Abstract:
With more and more processor cores integrated on a chip, Networks-on-chip (NoC) is emerging as a candidate architecture for multiprocessor systems-on-chip (MPSoC). Traditional metallic interconnects have become the bottleneck of NoC due to the limited bandwidth, long delay, and high power consumption. Optical Network-on-Chip (ONoC) can decrease interconnect delay and provide higher bandwidth with lower power consumption. In this paper, we propose a Clos-based Hierarchical Optical-Electronic NoC, called CHONoC, which can take advantage of both optical routers and interconnects in a hierarchical manner. CHONoC employs novel designs including two different topologies in optical layer and electric layers, communication mechanism with high path diversity, and two optical symmetric routers with strictly non-blocking property. Simulation results show that CHONoC can achieve small latency and high throughput especially for high local traffic compared with Mesh-based, Cmesh-based and Clos(8,8,8)-based ONoCs.
OUCS-2013-04
Not available: to be published
Yawen Chen, Haibo Zhang
Department of Computer Science, University of Otago
Zheng Chen, Huaxi Gu
State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an, China
Source- and Destination-based Wavelength Assignment Approach in Optical Network-on-Chip: Design and Performance
Abstract:
Photonic interconnect can provide ultra-high bandwidth with low power consumption, which makes Optical Network-on-Chips (ONoCs) a promising alternative for the next generation high performance chip multiprocessors. To make better use of the optical interconnect, Wavelength Division Multiplexing (WDM) is commonly used to provide high bandwidth and decrease the contention possibility. In this paper, we propose two wavelength assignment methods for Mesh ONoCs: source-based wavelength assignment (SW) and destination-based wavelength assignment (DW). These two low complexity approaches allow us to use a limited number of wavelengths according to the source and destination addresses. We evaluate their performance in simulations, and results show that allocating wavelengths according to the source node outperforms destination-based assignment in terms of latency and throughput.
OUCS-2013-thesis01
Kai-Cheung Leung
Department of Computer Science, University of Otago
PhD Thesis: View-Oriented Parallel Programming and its Performance Evaluation on Multicore Architectures
Abstract:
Shared-memory multicore architectures have become pervasive, and there is a pressing need for parallel programming models to facilitate both performance and convenience. However, most existing shared-memory programming models are tedious for programming and are prone to errors such as data race, which are difficult to debug.
To solve this problem, this thesis proposes a data race prevention scheme in the View-Oriented Parallel Programming (VOPP) paradigm. VOPP was proposed for distributed shared memory systems. It is adapted to shared-memory multicore architectures in this thesis. VOPP is a shared-memory data-centric parallel programming model, which uses views to bundle mutual exclusion with data access. In VOPP, programmers partition the shared memory into “views”, which are non-overlapping sets of shared data objects. The data race prevention scheme proposed for VOPP can prevent data race through the memory protection mechanism while keeping the extra overhead low.
To improve the programmability of VOPP, this thesis proposes an automatic view access management scheme where a view is automatically acquired upon its first access, and automatically released when no longer needed, thus relieving programmers from arranging locks to protect critical sections.
To further improve performance and programmability, this thesis proposes the View-Oriented Transactional Memory (VOTM) system, which uses Restricted Admission Control (RAC) to manage the number of processes holding each view according to its contention. In VOTM, RAC can restrict the number of processes holding the view when its contention is high, and in extreme cases, RAC can fall back to the locking mode, in order to avoid abort overheads of transactions. On the other hand, RAC allows unlimited concurrent access to other low-contention views to maximize concurrency, just as in transactional memory. Therefore, VOTM has the merits of both the locking mechanism and the transactional memory (TM) and integrated them nicely through RAC.
This thesis has also provided a theoretical analysis for RAC to investigate factors that indicate performance gain by restricting admission to a view, including disproportionately large portion of time spent in aborted transactions due to high contention and excessive TM mechanism overheads.
Experimental results demonstrate that in many cases, RAC correctly responds to these situations by restricting admission to a view, thus improves the performance. Apart from the improvements of programmability in VOPP, this thesis has done extensive experiments on two multicore architectures, a 16-core machine and a 64-core machine. Experimental results demonstrate that VOPP can provide a data race free environment with low overheads on multicore architectures, and VOTM outperforms both traditional transactional memory models and lock-based models in most benchmark applications.
OUCS-2013-03
Note: This report was updated 31.10.2014
Mira Guise, Alistair Knott, Lubica Benuskova
Department of Computer Science, University of Otago
Spinula: software for simulation and analysis of spiking network models
Abstract:
Spinula is a software package for the simulation and analysis of spiking neural network models. It consists of a core library that provides the simulation environment, and additional libraries that support analysis and visualization of the resulting data. This report examines each of these libraries and provides a detailed description of the included services and examples of how they might be used.
OUCS-2013-02
Mira Guise, Alistair Knott, Lubica Benuskova
Department of Computer Science, University of Otago
Experiments on the effect of synaptic disruption on polychronous group formation: detailed methods and results
Abstract:
Izhikevich (2006a) claims that groups of strongly connected neurons known as polychronous neural groups (PNGs) could provide the basis for memory in the brain. Polychronous groups exist as spatio-temporal patterns of connection lengths and weights; when activated by a matching input pattern, they are able to produce a causal chain of firing events that is not synchronous but is nevertheless precisely timed and consistently reproducible.
Izhikevich et al. (2004) have previously shown that synaptic disruption produces a dramatic decrease in PNG counts, leading them to the conclusion that PNG formation depends on activity-dependent changes in synaptic plasticity. However, despite the disruption some polychronous groups remain in the network, suggesting that polychronous groups can sometimes be formed by chance arrangements of synaptic weights in the network. In the following report we reproduce and extend the work of Izhikevich et al. (2004) by delving further into the nature of PNG formation: firstly, we examine the effects of synaptic disruption on the groups that remain; we then examine the considerable inter-network variation in PNG counts over the time-course of network maturation and categorize the resulting pro.les into discrete classes of behavior.
The first experiment found a highly signifcant decrease in PNG count following synaptic disruption (paired t(19) = 9:0; p < 0:001), reproducing the previous report (Izhikevich et al., 2004). However, we also found a highly significant decrease in both size (paired t(19) = 10:7; p < 0:001) and temporal length (paired t(19) = 10:8; p < 0:001) in the remaining polychronous groups. A further experiment employing sampling at multiple time-points found two previously unreported phenomena: firstly, there is a consistent small peak in the PNG counts immediately following initialization; secondly, comparison of the temporal profiles of multiple networks over the course of maturation shows two broad classes of network behavior: either cyclic, or an initial burst of productivity followed by a slow decline. These results are discussed in the context of a model of PNG formation involving an activity- dependent interaction between supported and adapted groups which is at the heart of PNG formation.
OUCS-2013-01
Martin Takac
Centre for Cognitive Science, Comenius University, Slovakia
Alistair Knott
Department of Computer Science, University of Otago
A neural network model of working memory for episodes
Abstract:
We present a neural network model of the storage of episode representations in working memory (WM). Our key idea is that episodes are encoded in WM as prepared sensorimotor routines: i.e. as prepared sequences of attentional and motor operations. Our network reproduces several experimental findings about the representation of prepared sequences in prefrontal cortex. Interpreted as a model of WM episode representations, it has useful applications in an account of long-term memory for episodes and in accounts of sentence processing.
2012
Now published
Alistair Knott
Department of Computer Science, University of Otago
Sensorimotor Cognition and Natural Language Syntax
Abstract:
(17/9/2012 - Report has been removed as the book was published in October 2012. See the publisher page.)
This book is about the interface between natural language and the sensorimotor system. It is obvious that there is an interface between language and sensorimotor cognition, because we can talk about what we see and do. The main proposal in the book is that the interface is more direct than is commonly assumed.
2011
OUCS-2011-04
Richard O'Keefe
Department of Computer Science, University of Otago
Specifying Exact Scaled Decimal Arithmetic
Abstract:
The ANSI Smalltalk standard includes a ScaledDecimal class for decimal fixed point arithmetic, but the specification is so vague that implementations vary greatly. The Language Independent Arithmetic standard has nothing to say about this data type. This article presents one reasonable specification, treating these numbers as exact.
OUCS-2011-03
Martin Takac, Lubica Benuskova and Alistair Knott
Department of Computer Science, University of Otago
A connectionist model of language acquisition and sentence generation: Technical appendix
Abstract:
In this report we present technical details of a neural network model of sentence generation, including details of the artificial languages it was trained on, its training regime, and of the performance of the trained network.
OUCS-2011-02
Hayden Walles, Anthony Robins and Alistair Knott
Department of Computer Science, University of Otago
Performance of a convolutional classifier network on the MNIST handwritten digit database
Abstract:
This report describes an experiment in which the convolutional neural network of Walles et al. (2008) is trained and tested on the MNIST database of handwritten digits (LeCun et al, 1999).
OUCS-2011-01
Martin Takac, Lubica Benuskova, Alistair Knott
Department of Computer Science, University of Otago
Mapping sensorimotor sequences to word sequences: A connectionist model of language acquisition and sentence generation
Abstract:
In this article we present a neural network model of sentence generation. The main technical novelty in the model is in its semantic representations: the messages which form the input to the network are structured as sequences, which are delivered to the network one at a time. Rather than learning to linearise a static semantic representation as a sequence of words, our network rehearses a sequence of semantic signals, and learns to generate words from selected signals. Our use of sequences to encode semantic representations has several benefits, both conceptual and technical. Conceptually, the use of rehearsed sequences of semantic signals connects to work in embodied cognition, which posits that the structure of semantic representations has its origin in the serial structure of sensorimotor processing. It also connects to nativist models of language development: we argue that some of the linguistic universals proposed within Chomskyan models of syntax can be interpreted as reflections of sensorimotor processing. Technically, the use of sequentially structured semantic representations permits a novel answer to the question of how a neural network can learn genuinely abstract syntactic rules (a vexed question in connectionist models of language). Equally importantly, it supports a way of using abstract syntactic rules in combination with rules about surface patterns in language. In summary, sequentially structured semantic representations allow a neural network model which combines elements from nativist, empiricist and embodied theories of language in a novel way.
