Competitive training in hidden Markov models

The use of hidden Markov models is placed in a connectionist framework, and an alternative approach to improving their ability to discriminate between classes is described. Using a network style of training, a measure of discrimination based on the a posteriori probability of state occupation is proposed, and the theory for its optimization using error back-propagation and gradient ascent is presented. The method is shown to be numerically well behaved, and results are presented which demonstrate that when using a simple threshold test on the probability of state occupation, the proposed optimization scheme leads to improved recognition performance.

MMI training for continuous phoneme recognition on the TIMIT database

This paper reports our experiences with a phoneme recognition system for the TIMIT database which uses multiple mixture continuous density monophone HMMs trained using MMI. A comprehensive set of results are presented comparing the ML and MMI training criteria for both diagonal and full covariance models. These results using simple monophone HMMs show clear performance gains achieved by MMI training, and are comparable to the best reported by others including those which use context-dependent models. In addition, the paper discusses a number of performance and implementation issues which are crucial to successful MMI training.

Rolling of thin strip and foil: Application of a tribological model for "mixed" lubrication

A mechanical model of cold rolling of foil is coupled with a sophisticated tribological model. The tribological model treats the "mixed" lubrication regime of practical interest, in which there is "real" contact between the roll and strip as well as pressurized oil between the surfaces. The variation of oil film thickness and contact ratio in the bite is found by considering flattening of asperities on the foil and the build-up of hydrodynamic pressure through the bite. The boundary friction coefficient for the contact areas is taken from strip drawing tests under similar tribological conditions. Theoretical results confirm that roll load and forward slip decrease with increasing rolling speed due to the decrease in contact ratio and friction. The predictions of the model are verified using mill trials under industrial conditions. For both thin strip and foil, the load predicted by the model has reasonable agreement with the measurements. For rolling of foil, forward slip is overestimated. This is greatly improved if a variation of friction through the bite is considered.

A mixed culture trail of mud crab (Scylla serrata Forskal) with tilapia ( Oreocromis niloticus L.)

A ten-month study on mixed culture of mud crab Scylla serrata with tilapia (Oreochromis niloticus) was performed in brackish water earthen ponds using live tilapia fry as the only feed for crab. The monthly growth rate varied from 9.07-19.llg among four treatments. Treatment T 1, cw: 0.68cm (±0.72) and bw: 19.11 gm (±12.97) showed highest performance which was followed by T 3, cw: 0.62cm (±0.60) and bw: 13.42 gm (±10.51), T4, cw: 0.65cm (±0.64) and bw: 13.20 gm (±9.89) and T2, cw: 0.36cm (± 0.25) and bw: 9.07 gm (±8.05). Highest survivability of crabs was also recorded in T 1 (21.5%) which was followed by T 2 (15.65%), T4 (14.95%) and T 3 (14.15%). In terms of survivability, significant differences (p<0.05) were observed among the treatments whereas these were recorded as insignificant difference (p<0.05) in final weight, weight gain and production of crabs and tilapia. Mixed culture of mud crab with tilapia could make more rewarding than crab monoculture but the study suggests that only tilapia fry can not fulfill the feed requirement of crabs in respects of survivability, final body weight and weight gain. Besides, existed salinity level of 4-12 ppt during experimental period might be the another key factor for low survivability and weight gain.

The potential of mixed culture of freshwater giant prawn Macrobrachium rosenbergii de Man and tiger shrimp Penaeus monodon Fab. at Khulna region, Bangladesh

The freshwater giant prawn (golda), Macrobrachium rosenbergii and tiger shrimp (bagda), Penaeus monodon were stocked together with or without fin fishes at different stocking rates in semi-saline waters at Khulna region and their growth, survival, yield and costreturn analysis were made. Survival rate of golda and bagda ranged from 23.0 to 36.8% and 8.2 to 24%, respectively. The both species were significantly affected by their own stocking density. The average final weight of golda and bagda ranged from 62.4 to 73.3 g and 32.0 to 66.4 g. The bivariate analysis of average final weight of both golda and bagda revealed that golda positively and bagda negatively influenced by the total stocking density. However, the results of the individual sizes of both golda and bagda showed an increase in the proportions of smaller animals and a decrease in the proportion of larger ones with increasing stocking rates. The harvesting weights of all animals in the experimental ghers were in marketable sizes although their prices varied with the individual size. The total production comprised of both golda and bagda ranged from 514.6 to 952.8 kg ha·1, over a culture period of 10 months. Return on investment ranged from 51.0 to 125.7%.

Over-wintering performance of mixed sex and monosex tilapia (Oreochromis niloticus L.) in Bangladesh

An experiment was conducted for six months in 6 experimental ponds (each size 80 of m2) to assess the over-wintering performance between mixed sex and monosex tilapia, Oreochromis niloticus. The experiment was carried out with two treatments each with three replicates. In the first treatment (T1), mixed sex tilapia were stocked in 3 ponds with a mean initial of 4.80±0.18 g. In the second treatment (T2), monosex tilapia were stocked in another 3 ponds with a mean initial weight of 4.81 ±0.20 g. Each pond was stocked with 250 fingerlings. Fish were fed at the rate of 6% of fish body weight at the beginning. The feeding rate was gradually reduced to 2% for the third month and finally increased to 3% for rest of the period. Water quality was monitored fortnightly and the ranges were: temperature17.86-29.10°C, dissolved oxygen 4.25-6.10 mg/1, pH 6.97-7.20 and transparency 24.10-36.50 cm. After 6 months of rearing monosex tilapia attained a significantly (pmixed sex tilapia. Monosex tilapia also resulted in significantly (Pmixed sex tilapia. However, there was no significant (pmixed sex and monosex tilapia. The production of monosex tilapia (3723.10 kg/ha) was about 32% higher than that of mixed sex tilapia (2776.28 kg/ha). The net profit/ha generated from the 6 months culture period was calculated as Tk. 43,311.45 and. 69,277.32/- for mixed sex and monosex tilapia respectively. The results of the present study suggested that it is possible to successfully culture tilapia during the winter period and the culture of monosex tilapia is more profitable due to its higher growth rate than that of mixesex tilapia.

Performance of genetically improved farmed tilapia (GIFT) under mono and mixed cui ture with silver barb (Barbodes gonionotus) in south-west Bangladesh

A participatory on-farm trial was carried out to evaluate the production performance of GIFT (genetically improved farmed tilapia) strain of Oreochmis sp., either alone or with silver barb (Barbodes gonionotus), in six rain-fed freshwater ponds of coastal area. There vvere two treatments; (i) GIFT alone at a stocking density of 24,700/ha (T1) and (ii) 1:1 combination of GIFT and silver barb (T1). Each of the treatments had three replications. A significantly (p

Foot contact detection for sprint training

We introduce a new algorithm to automatically identify the time and pixel location of foot contact events in high speed video of sprinters. We use this information to autonomously synchronise and overlay multiple recorded performances to provide feedback to athletes and coaches during their training sessions. The algorithm exploits the variation in speed of different parts of the body during sprinting. We use an array of foreground accumulators to identify short-term static pixels and a temporal analysis of the associated static regions to identify foot contacts. We evaluated the technique using 13 videos of three sprinters. It successfully identifed 55 of the 56 contacts, with a mean localisation error of 1.39±1.05 pixels. Some videos were also seen to produce additional, spurious contacts. We present heuristics to help identify the true contacts. © 2011 Springer-Verlag Berlin Heidelberg.

Training needs of traditional marine fishermen in Kerala

Training needs of 80 fishermen in 25 subject areas revealed a mean training need score of 23.0l; 95% wanted to get trained. The training needs were fairly strong in all subject areas, with the highest demand being for fishery engineering. Training need was also high for areas related to fishery technology. Most of the fishermen preferred to have the training at their own village, and in the months of June or July for an average period of 20.85 days. Education and income were positively related to intensity of training needs whereas age, number of family members, number of employed family members and experience in fishing were negatively correlated with it. These six variables explained 27 of the variance in training need intensity.