Description:
The choice of state and action representation in Reinforcement Learning (RL) has a significant effect on agent performance for the training task. But its relationship with generalization to new tasks is under-explored. One approach to improving generalization investigated here is the use of language as a representation. We compare vector-states and discreteactions to language representations. We find the agents using language representations generalize better and could solve tasks with more entities, new entities, and more complexity than seen in the training task. We attribute this to the compositionality of language
Description:
The choice of state and action representation in Reinforcement Learning (RL) has a significant effect on agent performance for the training task. But its relationship with generalization to new tasks is under-explored. One approach to improving generalization investigated here is the use of language as a representation. We compare vector-states and discreteactions to language representations. We find the agents using language representations generalize better and could solve tasks with more entities, new entities, and more complexity than seen in the training task. We attribute this to the compositionality of language
Description:
In this paper, a high‐capacity fabricated hollow section is proposed as a lightweight replacement for concrete‐filled composite members which are commonly used as beam‐columns in seismic zones. The lateral cyclic performance of the fabricated hollow beam‐column comprising of mild steel plates (grade 250) and high strength steel (HSS) tubes (grade 750) is investigated. A finite element model is developed and validated against available experimental data. To incorporate the cyclic hardening and softening of both steel materials, a combined plasticity model obtained from low‐cycle material tests is incorporated in the finite element model. A comprehensive sensitivity analysis is undertaken to propose an optimum slenderness design for the thin‐walled steel elements with ductile failure. While the high strength steel tube elements can significantly increase the axial and lateral capacity of the hybrid section, the cyclic failure mechanism is governed by mild steel plate elements, rather than a sudden failure in the high strength steel tubes.
Description:
Water of adequate quality and quantity is the key to health and integrity of the environment and fundamental to good water supply. Achieving water quality and quantity objectives can conflict and has become more complicated with challenges like, climate change, growing populations and changed land uses. Therefore, a multi-objective optimisation strategy is required for achieving optimal water quality and quantity outcomes from a water resources system. This study uses a multi-objective optimisation approach to illustrate the trade-offs occurring when water quantity and quality in a reservoir system are optimised. Taylors Lake, part of the Grampians Reservoir System in Western Victoria, Australia was chosen as the case study for this research as it is quite complex and includes many contemporary water resources challenges seen around the world, such as high turbidity and salinity. The objective functions are set in a way to maximise the water quantity available for supply, while minimising the deviation of quality parameters from the accepted limits. The water system is modelled using eWater Source® modelling platform, while optimisation is undertaken using NSGA-II optimisation technique. Daily time step data over a ten-year period was used in this work. Various optimisation runs were performed with different population sizes and generations to seek out the best trade-off curve. The optimisation results indicate trade-offs between salinity, turbidity, and quantity. Key findings for this case study show that through optimisation, stored water never exceeded 19,000 ML even though the storage capacity was 27,000 ML indicating a significant loss of water to improve quality, or alternatively, a potential asset re-design opportunity.
Description:
Being a significant stakeholder in the graduates' employment outcomes, it is vital to understand employers' perceptions of graduates' work-readiness. However, existing information systems (IS) literature focuses mainly on the perceptions of students or universities. This paper aims to fill this gap by analysing scoping interviews conducted with graduate recruiters and industry experts in Australia regarding attributes that can improve graduates' employment prospects in the information and communication technology industry. A preliminary investigation based on grounded theory identified three emergent themes from the data: behaviors, skills, and knowledge levels. Based on the findings, this study proposes an IS graduate work-readiness framework that can help universities to develop academic programs aimed at enhancing desirable skills and attitudes among IS graduates' employment.