Métodes d'aprenentatge per reforç basats en la política i la seva aplicació a la robótica

Projecte de recerca elaborat a partir d’una estada al Robot Locomotion Group del Massachusetts Institute of Technology, Estats Units, entre març i agost del 2006. Es descriu la feina portada a terme en el camp de l'aprenentatge per reforç (RL), una metodologia molt utilitzada en aprenentatge artificial. En RL, un agent intenta maximitzar un valor escalar (càstig o premi) obtingut com a resultat de la seva interacció amb l'entorn. L'objectiu d'un sistema basat en RL és el de trobar una política d'actuació òptima que relaciona l'estat de l'entorn amb una acció determinada que maximitzi la suma de reforços futurs. El principal avantatge és que no utilitza cap base de dades conegudes, així que l'agent no rep informació sobre quina decisió triar, com succeeix en molts tipus d'aprenentatge, sinó que ha de triar per descobrir aquelles accions que tenen un valor més alt, sent molt adient en robòtica aplicada. Els principals desavantatges són uns temps de convergència sovint elevats i la manca de generalització quan tractem variables contínues. Principalment, el treball s’ha centrat en l'estudi de noves i més complexes metodologies basades en RL que combinessin dos tipus d'algorismes: els basats en funcions de valor i els representats únicament per una política d'actuació. Posteriorment s'analitzà la seva aplicabilitat en aplicacions robòtiques reals. En tots els estudis i les simulacions s’ha utilitzat un braç robòtic dissenyat i contruït al laboratori. El tipus de robot, anomenat Acrobot, és un banc de proves molt utilitzat en els camps de teoria de control i aprenentatge.

Venda de kiwis online: www.prosmokiwi.com

Avui en dia la venda de productes, mitjançant les possibilitats que ens ofereix Internet, es troba en ple creixement. Aquest projecte pretén posar en funcionament una pàgina Web dedicada a la venda de fruita, concretament kiwis. Des de fa un temps, la població comença a ser conscient del desequilibri entre l'agent productor i l'agent comercial. Com passa també en altres sectors, el productor ven a un preu molt inferior respecte al que després es ficarà de cara al comprador final. En el cas de la fruita, el client acaba comprant un producte més car i normalment de menys qualitat. L'objectiu principal d'aquest projecte és promoure la venda online a partir d'una mercaderia de qualitat i més econòmica, aconseguint un major benefici tant per part del productor com del client.

H-Link simulator

Aquest projecte intenta donar solució al problema amb el qual s’ha trobat la empresa Hitachi Air Conditioning Products SA a la hora de fer probes durant el desenvolupament del seu principal producte, el CSNet Web. El simulador que s’ha realitzat en aquest projecte pretén reproduir la topologia d’una xarxa d’unitats d’aire condicionat per a que el CSNet Web les interpreti com si aquestes màquines fossin reals. Per a aconseguir-ho, el simulador reproduirà les comunicacions entre aquestes màquines i el CSNet Web.

Which subgroup of patients with rheumatoid arthritis benefits from switching to rituximab versus alternative anti-tumour necrosis factor (TNF) agents after previous failure of an anti-TNF agent?

BACKGROUND: Patients with rheumatoid arthritis (RA) with an inadequate response to TNF antagonists (aTNFs) may switch to an alternative aTNF or start treatment from a different class of drugs, such as rituximab (RTX). It remains unclear in which clinical settings these therapeutic strategies offer most benefit. OBJECTIVE: To analyse the effectiveness of RTX versus alternative aTNFs on RA disease activity in different subgroups of patients. METHODS: A prospective cohort study of patients with RA who discontinued at least one aTNF and subsequently received either RTX or an alternative aTNF, nested within the Swiss RA registry (SCQM-RA) was carried out. The primary outcome, longitudinal improvement in 28-joint count Disease Activity Score (DAS28), was analysed using multivariate regression models for longitudinal data and adjusted for potential confounders. RESULTS: Of the 318 patients with RA included; 155 received RTX and 163 received an alternative aTNF. The relative benefit of RTX varied with the type of prior aTNF failure: when the motive for switching was ineffectiveness to previous aTNFs, the longitudinal improvement in DAS28 was significantly better with RTX than with an alternative aTNF (p = 0.03; at 6 months, -1.34 (95% CI -1.54 to -1.15) vs -0.93 (95% CI -1.28 to -0.59), respectively). When the motive for switching was other causes, the longitudinal improvement in DAS28 was similar for RTX and alternative aTNFs (p = 0.40). These results were not significantly modified by the number of previous aTNF failures, the type of aTNF switches, or the presence of co-treatment with a disease-modifying antirheumatic drug. CONCLUSION: This observational study suggests that in patients with RA who have stopped a previous aTNF treatment because of ineffectiveness changing to RTX is more effective than switching to an alternative aTNF.

Expansió accelerada de l’univers i teories de gravetat modificada a grans escales

Estudi realitzat a partir d’una estada al Physics Department de la New York University, United States, Estats Units, entre 2006 i 2008. Una de les observacions de més impacte en la cosmologia moderna ha estat la determinació empírica que l’Univers es troba actualment en una fase d’Expansió Accelerada (EA). Aquest fenòmen implica que o bé l’Univers està dominat per un nou sector de matèria/energia, o bé la Relativitat General deixa de tenir validesa a escales cosmològiques. La primera possibilitat comprèn els models d’Energia Fosca (EF), i el seu principal problema és que l’EF ha de tenir propietats tan especials que es fan difícils de justificar teòricament. La segona possibilitat requereix la construcció de teories consistents de Gravetat Modificada a Grans Distàncies (GMGD), que són una generalització dels models de gravetat massiva. L’interès fenomenològic per aquestes teories també va resorgir amb l’aparició dels primers exemples de models de GMGD, com ara el model de Dvali, Gabadadze i Porrati (DGP), que consisteix en un tipus de brana en una dimensió extra. Malauradament, però, aquest model no permet explicar de forma consistent l’EA de l’Univers. Un dels objectius d’aquest projecte ha estat establir la viabilitat interna i fenomenològica dels models de GMGD. Des del punt de vista fenomenològic, ens hem centrat en la questió més important a la pràctica: trobar signatures observacionals que permetin distingir els models de GMGD dels d’EF. A nivell més teòric, també hem investigat el significat de les inestabilitats del model DGP.L’altre gran objectiu que ens vam proposar va ser la construcció de noves teories de GMGD. En la segona part d’aquest projecte, hem elaborat i mostrat la consistència del model “DGP en Cascada”, que generalitza el model DGP a més dimensions extra, i representa el segon model consistent i invariant-Lorentz a l’espai pla conegut. L’existència d’altres models de GMGD més enllà de DGP és de gran interès atès que podria permetre obtenir l’EA de l’Univers de forma purament geomètrica.

A balance of questions: what can we ask of climate change economics?

The standard approach to the economics of climate change, which has its best known implementation in Nordhaus's DICE and RICE models (well described in Nordhaus's 2008 book, A Question of Balance) is not well equipped to deal with the possibility of catastrophe, since we are unable to evaluate a risk averse representative agent's expected utility when there is any signi cant probability of zero consumption. Whilst other authors attempt to develop new tools with which to address these problems, the simple solution proposed in this paper is to ask a question that the currently available tools of climate change economics are capable of answering. Rather than having agents optimally choosing a path (that differs from the recommendations of climate scientists) within models which cannot capture the essential features of the problem, I argue that economic models should be used to determine the savings and investment paths which implement climate targets that have been suggested in the physical science literature.

Moral Hazard with Counterfeit Signals

In many moral hazard problems, the principal evaluates the agent's performance based on signals which the agent may suppress and replace with counterfeits. This form of fraud may affect the design of optimal contracts drastically, leading to complete market failure in extreme cases. I show that in optimal contracts, the principal deters all fraud, and does so by two complementary mechanisms. First, the principal punishes signals that are suspicious, i.e. appear counterfeit. Second, the principal is lenient on bad signals that the agent could suppress, but does not.

Market Competition and Lower Tier Incentives

The relationship between competition and performance-related pay has been analyzed in single-principal-single-agent models. While this approach yields good predictions for managerial pay schemes, the predictions fail to apply for employees at lower tiers of a firm's hierarchy. In this paper, a principal-multi-agent model of incentive pay is developed which makes it possible to analyze the effect of changes in the competitiveness of markets on lower tier incentive payment schemes. The results explain why the payment schemes of agents located at low and mid tiers are less sensitive to changes in competition when aggregated firm data is used. Journal of Economic Literature classiffication numbers: D82, J21, L13, L22. Keywords: Cournot Competition, Contract Delegation, Moral Hazard, Entry, Market Size, Wage Cost.

The simultaneous conjugacy problem in groups of piecewise linear functions

Guba and Sapir asked, in their joint paper [8], if the simultaneous conjugacy problem was solvable in Diagram Groups or, at least, for Thompson's group F. We give an elementary proof for the solution of the latter question. This relies purely on the description of F as the group of piecewise linear orientation-preserving homeomorphisms of the unit. The techniques we develop allow us also to solve the ordinary conjugacy problem as well, and we can compute roots and centralizers. Moreover, these techniques can be generalized to solve the same questions in larger groups of piecewise-linear homeomorphisms.

An Empirical Examination of the R&D Boundaries of the Firm - A Problem-Solving Perspective

We consider, both theoretically and empirically, how different organization modes are aligned to govern the efficient solving of technological problems. The data set is a sample from the Chinese consumer electronics industry. Following mainly the problem solving perspective (PSP) within the knowledge based view (KBV), we develop and test several PSP and KBV hypotheses, in conjunction with competing transaction cost economics (TCE) alternatives, in an examination of the determinants of the R&D organization mode. The results show that a firm’s existing knowledge base is the single most important explanatory variable. Problem complexity and decomposability are also found to be important, consistent with the theoretical predictions of the PSP, but it is suggested that these two dimensions need to be treated as separate variables. TCE hypotheses also receive some support, but the estimation results seem more supportive of the PSP and the KBV than the TCE.

Symmetry for a Dirichlet-Neumann problem arising in water waves

"Vegeu el resum a l'inici del document del fitxer adjunt."

The management of iron deficiency in inflammatory bowel disease--an online tool developed by the RAND/UCLA appropriateness method.

BACKGROUND: Iron deficiency is a common and undertreated problem in inflammatory bowel disease (IBD). AIM: To develop an online tool to support treatment choice at the patient-specific level. METHODS: Using the RAND/UCLA Appropriateness Method (RUAM), a European expert panel assessed the appropriateness of treatment regimens for a variety of clinical scenarios in patients with non-anaemic iron deficiency (NAID) and iron deficiency anaemia (IDA). Treatment options included adjustment of IBD medication only, oral iron supplementation, high-/low-dose intravenous (IV) regimens, IV iron plus erythropoietin-stimulating agent (ESA), and blood transfusion. The panel process consisted of two individual rating rounds (1148 treatment indications; 9-point scale) and three plenary discussion meetings. RESULTS: The panel reached agreement on 71% of treatment indications. 'No treatment' was never considered appropriate, and repeat treatment after previous failure was generally discouraged. For 98% of scenarios, at least one treatment was appropriate. Adjustment of IBD medication was deemed appropriate in all patients with active disease. Use of oral iron was mainly considered an option in NAID and mildly anaemic patients without disease activity. IV regimens were often judged appropriate, with high-dose IV iron being the preferred option in 77% of IDA scenarios. Blood transfusion and IV+ESA were indicated in exceptional cases only. CONCLUSIONS: The RUAM revealed high agreement amongst experts on the management of iron deficiency in patients with IBD. High-dose IV iron was more often considered appropriate than other options. To facilitate dissemination of the recommendations, panel outcomes were embedded in an online tool, accessible via http://ferroscope.com/.