Optimal sample selection for batch-mode reinforcement learning

in Proceedings of the 3rd International Conference on Agents and Artificial Intelligence (ICAART 2011) (2011)

We introduce the Optimal Sample Selection (OSS) meta-algorithm for solving discrete-time Optimal Control problems. This meta-algorithm maps the problem of finding a near-optimal closed-loop policy to the ... [more ▼]

We introduce the Optimal Sample Selection (OSS) meta-algorithm for solving discrete-time Optimal Control problems. This meta-algorithm maps the problem of finding a near-optimal closed-loop policy to the identification of a small set of one-step system transitions, leading to high-quality policies when used as input of a batch-mode Reinforcement Learning (RL) algorithm. We detail a particular instance of this OSS metaalgorithm that uses tree-based Fitted Q-Iteration as a batch-mode RL algorithm and Cross Entropy search as a method for navigating efficiently in the space of sample sets. The results show that this particular instance of OSS algorithms is able to identify rapidly small sample sets leading to high-quality policies [less ▲]

Optimal synthesis of mechanisms using time-varying dimensions and natural coordinates

in Herskowitz, José (Ed.) Proceedings of the 6th World Congress of Structural and Multidisciplinary Optimization (WCSMO6) (2005, June)

This paper presents a simple approach to optimize the dimensions and the positions of 2D mechanisms for path or function-generator synthesis. The proposed method is particularly adapted to assembled ... [more ▼]

This paper presents a simple approach to optimize the dimensions and the positions of 2D mechanisms for path or function-generator synthesis. The proposed method is particularly adapted to assembled mechanisms since time-varying dimensions always satisfy the assembly conditions which may represent a real difficulty when dealing with closed-loop mechanisms. The objective is to minimize the strain energy of the bars - considered as flexible - of the mechanism when this one follows perfectly the desired path. Two optimization strategies are developed and criticized. The first one is based on separated optimizations of design parameters and point coordinates. The second one is more global and is performed in two stages : multiple local synthesis are needed first to find the initial point coordinates, and then a global synthesis stage is undertaken to find both the best dimensions and coordinates. The use of natural coordinates is also particularly interesting since the only non linear functions to optimize are distance functions, and the objective function is rather well-conditioned for a gradient-based optimizer. The question of finding the global optimum is addressed and discussed. Since a standard genetic algorithm may fail to find it, a different approach is proposed: exploring the design space to find several local optima among which the designer will choose the most relevant one taking other design constraints into account. A simple technique is applied which consists in running multiple optimization processes starting from uniformly-distributed initial dimensions (full-factorial design of experiments) across the parameter space. Three applications are presented: a simple four-bar path synthesis to illustrate the optimization strategies, a four-bar steering linkage synthesis for function generation – Ackermann relation to highlight the limits of both strategies, and eventually a six-bar steering mechanism to explore the design space and find different local optima. [less ▲]

Optimal T2 control chart with double sampling scheme - an alternative to the MEWMA chart

in Quality & Reliability Engineering International (2012), 28(7), 751-760

Recent studies have shown that the double sampling (DS) scheme yields rapid detection of out of control situations, but the economic consequences of applying the proposed method are not discussed in the ... [more ▼]

Recent studies have shown that the double sampling (DS) scheme yields rapid detection of out of control situations, but the economic consequences of applying the proposed method are not discussed in the literature yet. In this paper, the economic statistical design of the DS T2 control chart is designed to address this issue. In this regard, upon the Lorenzen and Vance (1986)’s economic model, the problem is formulized and then the cost function is minimized using the genetic algorithm search method to obtain the optimal design parameters. Besides, we assumed that the length of the time that process remains in control is exponentially distributed. Through an illustrative example we show that by applying the proposed method relatively large benefits can be achieved in a comparison with the classical T2 and the statistical DS T2 charts. Furthermore the performance of the ESD DS T2 charts is compared to the MEWMA and other variable ratio sampling (VRS) T2 control charts in the literature. [less ▲]

Optimale experimentele studies voor functionele MRI

in STAtOR (2012), 13(1), 23-26

Optimalisation de l'expression orale en langue étrangère. Formation pour l'I.F.A.P.M.E.

Learning material (2008)

Cette formation destinée aux formateurs de langues étrangères de l'IFAPME portait sur l'optimalisation de l'expression orale en langues étrangères. Quatre grands volets constituaient cette communication ... [more ▼]

Cette formation destinée aux formateurs de langues étrangères de l'IFAPME portait sur l'optimalisation de l'expression orale en langues étrangères. Quatre grands volets constituaient cette communication. Dans les deux premiers volets, nous avons établi une distinction entre l'expression et l'interaction orales, puis avons présenté la typologie des productions langagières (descriptif/informatif ; injonctif/incitatif ; argumentatif ; narratif) proposée dans le document "Familles de tâches en langues modernes" (CFB). Dans le troisième volet, nous avons travaillé sur le genre textuel "débat". A partir d'un exemple concret proposé aux participants, nous avons montré ce que cette production textuelle à dominante argumentative mobilise comme savoirs au niveau du lexique, des fonctions langagières et de la grammaire, ainsi que sur le plan des stratégies de communication. Un mode particulier d'organisation de la classe - "l'aquarium"- a été présenté, puis testé par les participants. Enfin, dans le quatrième volet, nous avons montré comment ce débat pouvait être inséré dans le canevas didactique "traditionnel" et dans le canevas par "situation-problème". [less ▲]

Optimalisation de la prise en charge du patient diabetique de type 2: resultats de l'etude DREAM" en medecine generale."

in Revue Médicale de Liège (2003), 58(3), 139-46

This paper reports the results of the "DREAM" ("Diabetes REinforcement of Adequate Management") study performed among 42 general practitioners. A treatment algorithm was proposed to intensify the ... [more ▼]

This paper reports the results of the "DREAM" ("Diabetes REinforcement of Adequate Management") study performed among 42 general practitioners. A treatment algorithm was proposed to intensify the management of type 2 diabetic patients in case of poor control (HbA1C > 7.5% and fasting blood glucose > 140 mg/dl) before asking for an advice from a diabetologist. A total of 163 patients participated in the study: 84 men and 79 women, 59 +/- 10 years old; 5.2 +/- 6.1 years of known duration of diabetes; BMI: 30.7 +/- 5.8 kg/m2. Almost two thirds of patients received an antihypertensive agent, one third a lipid-lowering drug, and three quarters at least one antidiabetic agent: 22% a sulfonylurea, 18% metformin, 25% a combined oral therapy, 10% insulin (alone or in combination). At baseline, mean HbA1C averaged 7.1 +/- 2.2% (N: 4-6%), and 34% of patients had HbA1C > 7.5%. From these 163 patients, 100 were evaluated after 12 months, 57% being well controlled vs 36% at baseline. Among the 32 patients whose hypoglycaemic treatment was intensified, HbA1C level significantly decreased by almost 0.8% (p < 0.05). Despite precise guidelines, 55% of patients with HbA1C > 7.5% had no treatment adjustment, essentially in the group already on oral bitherapy or on insulin. Among the 65 patients evaluated after 24 months, the percentage of well controlled patients decreased to 44% (vs 51% after 12 months). The decision of switching to insulin was often delayed (no patient after 1 year and 2 patients only after 2 years). In conclusion, the "DREAM" study demonstrates the feasibility, but also the difficulty, of optimising the management of type 2 diabetic patients in general practice. [less ▲]

Optimalisation de la production du 73Se par cyclotron. Synthèses de nouveaux composés radiopharmaceutiques marqués par émetteurs gamma et positoniques (75Se, 73Se, 11C et 18F)

Doctoral thesis (1990)

Optimalisation des apports en acides aminés pour des taurillons BBB en complément à une ration fermière

Conference (2001, December 12)

Optimalisation des techniques prothetiques de rehabilitation orale grace a l'orthodontie: analyse d'un cas clinique.

in Revue Médicale de Liège (2008), 63(10), 609-14

This clinical case underlines the importance of a multidisciplinary approach, prosthetic and orthodontic, for the achievement of an oral rehabilitation treatment plan. Preliminary orthodontic treatment ... [more ▼]

This clinical case underlines the importance of a multidisciplinary approach, prosthetic and orthodontic, for the achievement of an oral rehabilitation treatment plan. Preliminary orthodontic treatment has significantly improved, the position of some natural teeth, which can serve or not as a prosthetic abutment. This has produced a direct prosthetic benefit by limiting tissue reduction for preparation and by giving a better biomechanical and functional context to the restoration.Thus, after an orthodontic treatment step, lasting a year and a half, using both removable appliance and fixed appliance, prosthetic rehabilitation was possible. This included a fixed prosthesis and a removable prosthesis that gave a very satisfying an esthetic result and good masticatory function. [less ▲]

Optimaliser les processus électoraux en RDC: transparence et efficience démocratique

Report (2010)