On-line policy optimisation of Bayesian spoken dialogue systems via human interaction

A partially observable Markov decision process has been proposed as a dialogue model that enables robustness to speech recognition errors and automatic policy optimisation using reinforcement learning (RL). However, conventional RL algorithms require a very large number of dialogues, necessitating a user simulator. Recently, Gaussian processes have been shown to substantially speed up the optimisation, making it possible to learn directly from interaction with human users. However, early studies have been limited to very low dimensional spaces and the learning has exhibited convergence problems. Here we investigate learning from human interaction using the Bayesian Update of Dialogue State system. This dynamic Bayesian network based system has an optimisation space covering more than one hundred features, allowing a wide range of behaviours to be learned. Using an improved policy model and a more robust reward function, we show that stable learning can be achieved that significantly outperforms a simulator trained policy. © 2013 IEEE.

Assessment of dry weight by monitoring changes in blood volume during hemodialysis using Crit-Line.

BACKGROUND: Routine assessment of dry weight in chronic hemodialysis patients relies primarily on clinical evaluation of patient fluid status. We evaluated whether measurement of postdialytic vascular refill could assist in the assessment of dry weight. METHODS: Twenty-eight chronic, stable hemodialysis patients were studied during routine treatment sessions using constant dialysate temperature and dialysate sodium concentration, and relative changes in blood volume were monitored using Crit-Line III monitors throughout this study. The study was divided into three phases. Phase 1 studies evaluated the time-dependence of vascular compartment refill after completion of hemodialysis. Phase 2 studies evaluated the relationships in patient subgroups between intradialytic changes in blood volume and the presence of postdialytic vascular compartment refill during that last 10 minutes of hemodialysis after stopping ultrafiltration. Phase 3 studies evaluated the extent of dry weight changes following the application of a protocol for blood volume reduction, postdialytic vascular compartment refill, and correlation with clinical evidence of intradialytic hypovolemia and/or postdialytic fatigue. Phase 3 included anywhere from three to five treatments. RESULTS: Phase 1 studies demonstrated that despite interpatient variability in the magnitude of postdialytic vascular compartment refill, when significant refill was evident, it always continued for at least 30 minutes. However, the majority of refill took place within 10 minutes postdialysis. Phase 2 studies identified 3 groups of patients: those who exhibited intradialytic reductions in blood volume but not postdialytic vascular compartment refill (group 1), those who exhibited intradialytic reductions in blood volume and postdialytic vascular compartment refill (group 2), and those whose blood volume did not change substantially during hemodialysis treatment (group 3). In phase 3 studies, use of an ultrafiltration protocol for blood volume reduction and monitoring of postdialytic vascular compartment refill combined with clinical assessment of hypovolemia and postdialytic fatigue demonstrated that patients often had a clinical dry weight assessment which was too low or too high. In all 28 patients studied, dry weight was either increased or decreased following use of this protocol. CONCLUSION: Determination of the extent of both intradialytic decreases in blood volume and postdialytic vascular compartment refill, combined with clinical assessment of intradialytic hypovolemia and postdialytic fatigue, can help assess patient dry weight and optimize volume status while reducing dialysis associated morbidity. The number of hospital admissions due to fluid overload may be reduced.

Theoretical study of line narrowing in a sub-terahertz monolithic dual laser source with an integrated reflector

A monolithic design is proposed for low-noise sub-THz signal generation by integrating a reflector onto a dual laser source. The reflectivity and the position of such a reflector can be adjusted to obtain constructive feedback from the reflector to both lasers, thus causing a Vernier feedback effect. As a result, 10-fold line narrowing, the narrowing being limited by the resolution of the simulation, is predicted using a transmission line model. Finally, a simple control scheme using an electrical feedback loop to adjust laser biases is proposed to maintain the line narrowing performance. This line narrowing technique, comprising a passive integrated reflector, could allow the development of a low-cost, compact and energy-efficient solution for high-purity sub-THz signal generation. © The Institution of Engineering and Technology 2014.