A paradigm for active decision support

Active decision support is concerned with developing advanced forms of decision support where the support tools are capable of actively participating in the decision making process, and decisions are made by fruitful callaboration between the human and the machine. It is currently an active and leading area of research within the field of decision support systems. The objective of this paper is to share the details of our research in this area. We present our overall research strategy for exploring advanced forms of decision support and discuss in detail our research prototype called JANUS that implements our ideas. We establish the contributions of our work and discuss our experiences and plans for future.     

Performance Analysis for E-Business: Impact of Long Range Dependence

We consider an e-business web-server system where the network traffic exhibits self-similarity. We demonstrate that traditional techniques are unsuitable for predicting the network performance under such traffic conditions. Instead, we propose and demonstrate a novel decomposition approximation technique that helps predict delays more accurately and thus is better suited for capacity planning and network design when compared to traditional queueing network analyzers. We also consider several strategies for mitigating the effect of self-similarity, and conclude that admission control holds the greatest potential for improving service. We provide an approximation technique for computing the admission control parameter values. Numerical results and suggestions for future work are discussed.     

Recent Advances in Hierarchical Reinforcement Learning

Reinforcement learning is bedeviled by the curse of dimensionality: the number of parameters to be learned grows exponentially with the size of any compact encoding of a state. Recent attempts to combat the curse of dimensionality have turned to principled ways of exploiting temporal abstraction, where decisions are not required at each step, but rather invoke the execution of temporally-extended activities which follow their own policies until termination. This leads naturally to hierarchical control architectures and associated learning algorithms. We review several approaches to temporal abstraction and hierarchical organization that machine learning researchers have recently developed. Common to these approaches is a reliance on the theory of semi-Markov decision processes, which we emphasize in our review. We then discuss extensions of these ideas to concurrent activities, multiagent coordination, and hierarchical memory for addressing partial observability. Concluding remarks address open challenges facing the further development of reinforcement learning in a hierarchical setting.     

Small bowel image classification based on Fourier-Zernike moment features and canonical discriminant analysis

In this paper, we propose a novel method for the classification of small bowel images into normal or abnormal class for automatic detection of cancers. We extract the Fourier features from the input small bowel image, and then the Zernike moment features are computed from the Fourier features. We then use the canonical discriminant analysis (CDA) to classify the small bowel images to normal or abnormal class. Experimental results show that the proposed method achieves the highest correct classification rate 100% for this problem. Our method is computationally very efficient. It can be used to automate the classification of capsule endoscopic images and to reduce the cost of interpreting those images that are acquired in clinical setting.     

Energy expenditure of cycle rickshaw pullers in different places in India

The cycle rickshaw is a popular transportation device. The aim of the study was to assess workload of cycle rickshaw pullers--physiological and subjective at four different places in India. Subjects were instructed to pull a cycle rickshaw with two passengers, for 20 min. Working and recovery heart rates were recorded. The mean values of energy expenditure of pulling a cycle rickshaw varied from 23.5 +/- 2.66 to 25.35 +/- 1.51 kJ/min. Relative cardiac strain and cardiac cost indicated that the job is 'heavy' to 'very heavy'. Subjective assessment of workload was 'heavy' to 'very heavy'. The combined workload assessed from physiological parameters and subjective assessment indicated that the job could be categorised as 'heavy' to 'very heavy' at all the places studied and needs to be reduced by redesigning the structural and functional components of the cycle rickshaw. The rickshaw pullers carry out the jobs many times per day, sometimes without proper rest pauses between trips. The outcome of the research project is beneficial for cycle rickshaw pullers, health administrators as well as manufacturers of cycle rickshaws. Primarily, the manufacturers would be able to use the data for producing a newer model of cycle rickshaw, which would require less energy to drive. The health administrators would be able to take policy decisions for administering better health care for the unorganised and underprivileged workers. The pullers could be guided into taking care of their health by improving their work practice, i.e. taking a sufficient rest pause between trips.     

Grid-enabled Virtual Screening Against Malaria

WISDOM is an international initiative to enable a virtual screening pipeline on a Grid infrastructure. Its first attempt was to deploy large scale in silico docking on a public Grid infrastructure. Protein–ligand docking is about computing the binding energy of a protein target to a library of potential drugs using a scoring algorithm. Previous deployments were either limited to one cluster, to Grids of clusters in the tightly protected environment of a pharmaceutical laboratory or to desktop Grids. The first large scale docking experiment ran on the EGEE Grid production service from 11 July 2005 to 19 August 2005 against targets relevant to research on malaria and saw over 41 million compounds docked for the equivalent of 80 years of CPU time. Up to 1,700 computers were simultaneously used in 15 countries around the world. Issues related to the deployment and the monitoring of the in silico docking experiment as well as experience with Grid operation and services are reported in the paper. The main problem encountered for such a large scale deployment was the Grid infrastructure stability. Although the overall success rate was above 80%, a lot of monitoring and supervision was still required at the application level to resubmit the jobs that failed. But the experiment demonstrated how Grid infrastructures have a tremendous capacity to mobilize very large CPU resources for well targeted goals during a significant period of time. This success leads to a second computing challenge targeting avian flu neuraminidase N1.     

Temperature-programmed desorption vs. N2 desorption in determining pore-size distribution of mesoporous silica molecular sieves

Two mesoporous silica molecular sieves, one synthesized by cationic template method and another by neutral template route, were characterized for their pore-size distribution by a novel Temperature-Programmed Desorption (TPD) method and the widely used N₂ desorption method. The pore-size distributions determined by the two methods agree quite well and are within experimental errors. For example, TPD method gave a pore size distribution (radius) centered at 14 Å while N₂ desorption method showed a peak centered at 13.3 Å. for the mesoporous silica prepared by cationic template route. The new TPD method based on thermogravimetric analysis is a viable option for mesopore characterization of silica-based materials.     

Determination of Flaw-Size Distribution by the Crack-Off Process

The crack-off process is a new alternative to diamond sawing for the cutting of brittle materials. The process involves exposing brittle rods (or plates) to biaxial fluid pressure. Fluid penetration into surface flaws causes fracture at pressures strongly dependent on the severity of the flaws. A method for calculating the probability-density distribution of surface flaws in a batch of brittle rods, using the crack-off pressures for a random sample, is presented here. The scatter in the pressure data is a consequence of the statistical nature of surface flaws. Parameters defining the flawsize distribution can hence be deduced from the crack-off pressure distribution. The method has been tested on two batches of glass rods that have undergone different treatments, with satisfactory results.