Task-Level Robot Learning

We are investigating how to program robots so that they learn from experience. Our goal is to develop principled methods of learning that can improve a robot's performance of a wide range of dynamic tasks. We have developed task-level learning that successfully improves a robot's performance of two complex tasks, ball-throwing and juggling. With task- level learning, a robot practices a task, monitors its own performance, and uses that experience to adjust its task-level commands. This learning method serves to complement other approaches, such as model calibration, for improving robot performance.

Pi: A Parallel Architecture Interface for Multi-Model Execution

This thesis defines Pi, a parallel architecture interface that separates model and machine issues, allowing them to be addressed independently. This provides greater flexibility for both the model and machine builder. Pi addresses a set of common parallel model requirements including low latency communication, fast task switching, low cost synchronization, efficient storage management, the ability to exploit locality, and efficient support for sequential code. Since Pi provides generic parallel operations, it can efficiently support many parallel programming models including hybrids of existing models. Pi also forms a basis of comparison for architectural components.

Virtual Model Control of a Biped Walking Robot

The transformation from high level task specification to low level motion control is a fundamental issue in sensorimotor control in animals and robots. This thesis develops a control scheme called virtual model control which addresses this issue. Virtual model control is a motion control language which uses simulations of imagined mechanical components to create forces, which are applied through joint torques, thereby creating the illusion that the components are connected to the robot. Due to the intuitive nature of this technique, designing a virtual model controller requires the same skills as designing the mechanism itself. A high level control system can be cascaded with the low level virtual model controller to modulate the parameters of the virtual mechanisms. Discrete commands from the high level controller would then result in fluid motion. An extension of Gardner's Partitioned Actuator Set Control method is developed. This method allows for the specification of constraints on the generalized forces which each serial path of a parallel mechanism can apply. Virtual model control has been applied to a bipedal walking robot. A simple algorithm utilizing a simple set of virtual components has successfully compelled the robot to walk eight consecutive steps.

Model Selection in Summary Evaluation

A difficulty in the design of automated text summarization algorithms is in the objective evaluation. Viewing summarization as a tradeoff between length and information content, we introduce a technique based on a hierarchy of classifiers to rank, through model selection, different summarization methods. This summary evaluation technique allows for broader comparison of summarization methods than the traditional techniques of summary evaluation. We present an empirical study of two simple, albeit widely used, summarization methods that shows the different usages of this automated task-based evaluation system and confirms the results obtained with human-based evaluation methods over smaller corpora.

Categorization in IT and PFC: Model and Experiments

In a recent experiment, Freedman et al. recorded from inferotemporal (IT) and prefrontal cortices (PFC) of monkeys performing a "cat/dog" categorization task (Freedman 2001 and Freedman, Riesenhuber, Poggio, Miller 2001). In this paper we analyze the tuning properties of view-tuned units in our HMAX model of object recognition in cortex (Riesenhuber 1999) using the same paradigm and stimuli as in the experiment. We then compare the simulation results to the monkey inferotemporal neuron population data. We find that view-tuned model IT units that were trained without any explicit category information can show category-related tuning as observed in the experiment. This suggests that the tuning properties of experimental IT neurons might primarily be shaped by bottom-up stimulus-space statistics, with little influence of top-down task-specific information. The population of experimental PFC neurons, on the other hand, shows tuning properties that cannot be explained just by stimulus tuning. These analyses are compatible with a model of object recognition in cortex (Riesenhuber 2000) in which a population of shape-tuned neurons provides a general basis for neurons tuned to different recognition tasks.

Corticosterone plasma concentrations in Carioca High- and Low-Conditioned freezing rats after a fear conditioned task

Our group in the Psychology Department at Pontifícia Universidade Católica do Rio de Janeiro (PUC-Rio) developed a rat genetic model of extreme freezing in response to contextual cues in an experimental chamber previously associated with footshock. One of the lines, Carioca High Freezing (CHF), exhibits an enhanced conditioned freezing response, whereas the other line, Carioca Low Freezing (CLF), shows the opposite response. The present study investigated corticosterone concentration between these two lines of animals and a random (RND) line of rats both under basal conditions and test condition after an emotional challenge using a contextual fear conditioning protocol. Comparisons between basal and test plasma corticosterone concentrations suggested differential basal and fear-induced differences between the two lines. The differences between basal conditions is an important and relevant aspect to be considered in behavioral experiments using or assessing stress and could help to understand variability in naïve populations.

Cap a un model per interpretar i facilitar l'aprendre: estratègies i recursos per a la reflexió-acció en i des de la pràctica

1.- L'enquadrament d'aquest treball de recerca s'ha fet en i des de la praxis. EI que interessa és descobrir i proposar instruments pedagògics d'ajuda, assequibles i contextualitzats, especialment en el terreny de la comunicació i la interacció Educador-Educand. La metodologia que s'ha fet servir és de caire qualitatiu, etnogràfic, en un enfocament basat en la investigació-acció. La visió de la persona és volgudament holística; els sentiments, els significats, I' orientació personal, I' autodirecció. esdevenen elements centrals. La hipòtesi de treball, en la qual es fonamenta la recerca, podria formular-se així: 'En la pràctica, els Educadors, d'una forma més o menys reflexiva, perceben i interpreten el procés d'aprendre dels Educands; hi intervenen, el mediatitzen; n'avaluen la direcció i l'encert". A partir d'aquest supòsit, l'investigador veu convenient donar resposta a tres qüestions centrals: A) Com comprenem i interpretem els Educadors el procés d'aprendre dels Educands?; B) Quin tipus d'intervenció resultarà adient per promoure i/o facilitar l'aprendre?; C) Amb quins instruments i estratègies comptem per ajudar pedagògicament? 2.- Per llegir l'aprendre, l'investigador fa ús d'un model mental, indispensable per ordenar les dades de l'experiència. Convé, per tant, explicitar-lo, fer-ne ciència, coneixement compartit. En aquesta direcció de treball se li plantegen dos tipus d'interrogants: A) Quins són els elements comuns a qualsevol experiència d'aprenentatge?; Quina mena d'activitat o experiència personal desenvolupen tots els Educands; i B) Com es manifesta aquesta experiència? Amb quins indicadors? Quina mena de "text" llegeix l'Educador per orientar la seva intervenció d'ajuda? 3.- L'aprenentatge. en aquest treball, és considerat per l'investigador com la resultant de tres processos personals: el posicionament, l'estratègia i l'avaluació. Quan l'Educador vol compartir amb l'Educand el seu procés idiosincràtic d'aprendre, procura reconstruir amb ell aquestes tres accions bàsiques: A) Com es posiciona: què tem o desitja, què creu, què espera, quines expectatives viu, com es motiva? B) Quines són les seves pautes d'actuació: com treballa, com memoritza, com recupera la informació que té a la memòria, què fa davant d'un problema... ? C) Què busca; què és important per a ella; què pretén... ? 4.- Davant la complexitat dels missatges emesos per l'Educand l'investigador opta per llegir tres tipus de "textos": els productes i els resultats; les conductes; i els missatges parlats. Entre tots tres, pensa, podrà trobar elements i indicadors adequats per fonamentar, sempre hipotèticament, la seva actuació pedagògica. 5.- Procura sobretot detectar i fer existir els èxits, aquells productes i/o resultats que l'Educand troba valuosos, per tal d'ajudar-lo a prendre consciència dels seu repertori personal d'estratègies i capacitats. En aquest àmbit es proposen tres actituds o enfocaments del treball d'ajuda: A) El primer fa referència a la presa de consciència de l'estratègia personal, que s'amaga darrera el producte valorat. B) El segon apunta cap al respecte per l'estratègia que cadascú executa i, per tant, li és familiar. L'Educand la necessita. Es la seva. Es troba en la seva experiència, encara que no necessàriament existeixi en la seva construcció conscient. C) El tercer ,el duu a valorar l'estratègia d'acord amb les finalitats de l'Educand. La seva adequació es legitima pel que es proposa. Partim del supòsit que tota conducta es troba dirigida per un propòsit a vegades difícil de copsar i no sempre recomanable per a l'Educand. 6.- La conducta percebuda de l'Educand és entesa com un missatge, un conjunt d'indicadors de la seva activitat contextualitzada, interna i externa; missatge que, en relació amb altres, com els productes i les verbalitzacions, manifesta fragments dels seus significats, projectes, estratègies, valors. Es fa un èmfasi especial en les conductes "internes", els gestos mentals, l'acció interior, tramesa per mitja de microcomportaments sovint no conscient, i certificada per mitjà de la verbalització del viscut. 7.- Parlar amb l'Educand suposa dues menes d'accions: escoltar i emetre. Escoltem per comprendre; emetem per perfilar la comprensió i també per ajudar. En l'emissió, el missatge pedagògic té dues funcions: a) rellançar i orientar el pensament i l'autoexploració de l'Educand; i b) influir per tal que desenvolupi amb èxit el seu projecte d'aprendre. Interessa d'una manera especial ajudar a integrar en la consciència de l'Educand aquests elements de la seva experiència que poden facilitar-li l'adquisició del coneixement. I entenem que, en aquesta empresa, la paraula i la interacció verbal poden tenir-hi un joc important. Per aquesta raó s'ha considerat necessari oferir un model d'anàlisi de la interacció i els missatges verbals. 8.- "Les persones aprenen sempre, amb recursos, processos i sistemes de valoració idiosincràtics, per fa qual cosa la seva orientació en el context esdevé un referent central en el disseny de l'ajuda pedagògica i en la seva avaluació". Aquesta és la hipòtesi de sortida per dissenyar la intervenció pedagògica. Tothom aprèn, inevitablement; la qüestió és quina cosa està aprenent i de quina manera els seus resultats d'ara són mediatitzats per l'experiència passada i, alhora, condicionen el seu aprendre futur. L'aprenent es posiciona, anticipa el procés d' aprendre, valora la seva. capacitat per desenvolupar-lo amb èxit, es motiva en una determinada direcció, d'acord amb la seva experiència, els seus aprenentatges anteriors. Executa estratègies, mostra un tipus d'intel·ligència, una forma personal de processar la informació. Pretén quelcom. És un sistema obert en relació amb el medi: hi ha uns valors que dirigeixen la seva presa de decisions. Utilitza uns criteris propis, una gamma personal d'opcions conscients. Avalua el que fa, el resultat que obté i la seva capacitat personal. 9.- L'ajuda pedagògica que I' autor proposa s'encamina sobretot a facilitar en l'Educand la descoberta dels seus propis recursos. Es tracta de portar-lo cap a la consideració atenta de la seva pròpia experiència, per amplificar-la i fer-la existir com a recurs conscient . Ha dibuixat i experimentat tres conjunts d'intervenció cadascun enfocat vers un àmbit de l'experiència d'aprendre, el qual col·loca com a prioritari, sense oblidar qualsevol dels altres que pugui ser rellevant, per comprendre o ajudar. A) Intervenció sobre el posicionament. En aquest àmbit enfoca l'estil de motivació que executa l'Educand, mira de corregir-lo, si cal, a partir de l'anàlisi i la comprensió de les seves formes de motivar-se quan ell viu l'èxit. Treballa proposant objectius paradoxals de fracàs gairebé impossible, buscant l'assoliment de petits èxits, potser aparentment insignificants, però estratègics; prescriu l'automatisme, per modificar-lo si l'Educand ho desitja; comprova el procés d'anticipació de I' experiència que l'aprenent es construeix per orientar-se; l'ajuda a contextualitzar anticipació i a fer ús dels seus Ilenguatges interns més eficaços i còmodes;... B) Intervenció sobre les estratègies i processos. En un segon enfocament, no necessàriament posterior al descrit, considera les estratègies de I'Educand, també a partir dels seus encerts i èxits. Mira de portar-lo cap a fa descripció i presa de consciència de les seves maneres de fer mes còmodes i segures, les que lliguen amb les seves preferències cerebrals. Quan viu dificultats, el convida a explorar les excepcions, els moments en els quals les seves realitzacions són satisfactòries. Pretén sobretot modificar les seves creences limitants, posant-lo en conflicte amb els fets de l'experiència. A vegades, caldrà facilitar l'adquisició d'estratègies i procediments nous que l'Educand considera plausibles. Es tracta específicament de fer existir opcions noves d'actuació per tal d'assolir allò que vol i/o necessita. C) Intervenció sobre el sistema de valors de l'Educand. L'Educand viu uns valors, els quals expliquen el seu món intern les conductes que realitza i els resultats que obté. Aquest àmbit és col·locat, en el model, al centre del procés d'aprendre. Hom actua amb propòsits determinats, no necessariament conscients. L'obertura de la persona a l'experiència d'aprendre es dirigeix segons criteris i valors irrenunciables. 10.- L 'Educador procura compartir els objectius de l'Educand i els seus projectes per assolir-los; vol tanmateix que se'n faci coneixedor i director responsable. Per a això li convé preguntar-se per quina mena d'experiència està desenvolupant i quin sentit ecològic te per a ell. L'Educador, el seu model del món, la seva persona, està compromès en el procés d'ajuda. No és només un tècnic que aplica recursos objectius. El seu model de comunicació, el seu pensament, les seves expectatives i anticipacions, tenen un pes considerable en el tipus d'intervenció pedagògica que durà a terme i, d'escreix, en el tipus d' aprenentatge que facilita. En la intervenció, l'Educador parteix d'una avaluació intencionadament positiva, centrada en els recursos i les solucions, en la metacognició i l'autoregulació dels processos, a partir de premisses que pressuposen l'èxit personal.

New strings for old : the role of similarity processing in an incidental learning task

This paper reports three experiments that examine the role of similarity processing in McGeorge and Burton's (1990) incidental learning task. In the experiments subjects performed a distractor task involving four-digit number strings, all of which conformed to a simple hidden rule. They were then given a forced-choice memory test in which they were presented with pairs of strings and were led to believe that one string of each pair had appeared in the prior learning phase. Although this was not the case, one string of each pair did conform to the hidden rule. Experiment 1 showed that, as in the McGeorge and Burton study, subjects were significantly more likely to select test strings that conformed to the hidden rule. However, additional analyses suggested that rather than having implicitly abstracted the rule, subjects may have been selecting strings that were in some way similar to those seen during the learning phase. Experiments 2 and 3 were designed to try to separate out effects due to similarity from those due to implicit rule abstraction. It was found that the results were more consistent with a similarity-based model than implicit rule abstraction per se.

Model for calculating the sustainable building index (SBI) in the kingdom of Bahrain

We have developed a model that allows players in the building and construction sector and the energy policy makers on energy strategies to be able to perceive the interest of investors in the kingdom of Bahrain in conducting Building Integrated Photovoltaic (BIPV) or Building integrated wind turbines (BIWT) projects, i.e. a partial sustainable or green buildings. The model allows the calculation of the Sustainable building index (SBI), which ranges from 0.1 (lowest) to 1.0 (highest); the higher figure the more chance for launching BIPV or BIWT. This model was tested in Bahrain and the calculated SBI was found 0.47. This means that an extensive effort must be made through policies on renewable energy, renewable energy education, and incentives to BIPV and BIWT projects, environmental awareness and promotion to clean and sustainable energy for building and construction projects. Our model can be used internationally to create a "Global SBI" database. The Sustainable building and construction initiative (SBCI), United Nation, can take the task for establishing such task using this model.

EnvironalPlanning: an analytical network process model for environmentally conscious construction planning

Although the construction pollution index has been put forward and proved to be an efficient approach to reducing or mitigating pollution level during the construction planning stage, the problem of how to select the best construction plan based on distinguishing the degree of its potential adverse environmental impacts is still a research task. This paper first reviews environmental issues and their characteristics in construction, which are critical factors in evaluating potential adverse impacts of a construction plan. These environmental characteristics are then used to structure two decision models for environmental-conscious construction planning by using an analytic network process (ANP), including a complicated model and a simplified model. The two ANP models are combined and called the EnvironalPlanning system, which is applied to evaluate potential adverse environmental impacts of alternative construction plans.

A demonstration of attentional bias, using a novel dual task paradigm, towards clinically salient material in recovering alcohol abuse patients?

Background. This study examined whether alcohol abuse patients are characterized either by enhanced schematic processing of alcohol related cues or by an attentional bias towards the processing of alcohol cues. Method. Abstinent alcohol abusers (N = 25) and non-clinical control participants (N = 24) performed a dual task paradigm in which they had to make an odd/even decision to a centrally presented number while performing a peripherally presented lexical decision task. Stimuli on the lexical decision task comprised alcohol words, neutral words and non-words. In addition, participants completed an incidental recall task for the words presented in the lexical decision task. Results. It was found that, in the presence of alcohol related words, the performance of patients on the odd/even decision task was poorer than in the presence of other stimului. In addition, patients displayed slower lexical decision times for alcohol related words. Both groups displayed better recall for alcohol words than for other stimuli. Conclusions. These results are interpreted as supporting neither model of drug cravings. Rather, it is proposed that, in the presence of alcohol stimuli, alcohol abuse patients display a breakdown in the ability to focus attention.

Parallel performance and scalability experiments with the Danish Eulerian Model on the EPCC supercomputers

The Danish Eulerian Model (DEM) is a powerful air pollution model, designed to calculate the concentrations of various dangerous species over a large geographical region (e.g. Europe). It takes into account the main physical and chemical processes between these species, the actual meteorological conditions, emissions, etc.. This is a huge computational task and requires significant resources of storage and CPU time. Parallel computing is essential for the efficient practical use of the model. Some efficient parallel versions of the model were created over the past several years. A suitable parallel version of DEM by using the Message Passing Interface library (AIPI) was implemented on two powerful supercomputers of the EPCC - Edinburgh, available via the HPC-Europa programme for transnational access to research infrastructures in EC: a Sun Fire E15K and an IBM HPCx cluster. Although the implementation is in principal, the same for both supercomputers, few modifications had to be done for successful porting of the code on the IBM HPCx cluster. Performance analysis and parallel optimization was done next. Results from bench marking experiments will be presented in this paper. Another set of experiments was carried out in order to investigate the sensitivity of the model to variation of some chemical rate constants in the chemical submodel. Certain modifications of the code were necessary to be done in accordance with this task. The obtained results will be used for further sensitivity analysis Studies by using Monte Carlo simulation.

Intelligent agents and distributed models for cooperative task support

Purpose – To describe some research done, as part of an EPSRC funded project, to assist engineers working together on collaborative tasks. Design/methodology/approach – Distributed finite state modelling and agent techniques are used successfully in a new hybrid self-organising decision making system applied to collaborative work support. For the particular application, analysis of the tasks involved has been performed and these tasks are modelled. The system then employs a novel generic agent model, where task and domain knowledge are isolated from the support system, which provides relevant information to the engineers. Findings – The method is applied in the despatch of transmission commands within the control room of The National Grid Company Plc (NGC) – tasks are completed significantly faster when the system is utilised. Research limitations/implications – The paper describes a generic approach and it would be interesting to investigate how well it works in other applications. Practical implications – Although only one application has been studied, the methodology could equally be applied to a general class of cooperative work environments. Originality/value – One key part of the work is the novel generic agent model that enables the task and domain knowledge, which are application specific, to be isolated from the support system, and hence allows the method to be applied in other domains.

Internal model control of nonlinear systems through the inversion of recurrent neural networks

Recurrent neural networks can be used for both the identification and control of nonlinear systems. This paper takes a previously derived set of theoretical results about recurrent neural networks and applies them to the task of providing internal model control for a nonlinear plant. Using the theoretical results, we show how an inverse controller can be produced from a neural network model of the plant, without the need to train an additional network to perform the inverse control.

A multinomial model of applying recognition to judge between multiple alternatives

Proponents of the “fast and frugal” approach to decision-making suggest that inferential judgments are best made on the basis of limited information. For example, if only one of two cities is recognized and the task is to judge which city has the larger population, the recognition heuristic states that the recognized city should be selected. In preference choices with >2 options, it is also standard to assume that a “consideration set”, based upon some simple criterion, is established to reduce the options available. A multinomial processing tree model is outlined which provides the basis for estimating the extent to which recognition is used as a criterion in establishing a consideration set for inferential judgments.