905 resultados para Reinforcement Learning,resource-constrained devices,iOS devices,on-device machine learning


Relevância:

100.00% 100.00%

Publicador:

Resumo:

A Bayesian optimisation algorithm for a nurse scheduling problem is presented, which involves choosing a suitable scheduling rule from a set for each nurse's assignment. When a human scheduler works, he normally builds a schedule systematically following a set of rules. After much practice, the scheduler gradually masters the knowledge of which solution parts go well with others. He can identify good parts and is aware of the solution quality even if the scheduling process is not yet completed, thus having the ability to finish a schedule by using flexible, rather than fixed, rules. In this paper, we design a more human-like scheduling algorithm, by using a Bayesian optimisation algorithm to implement explicit learning from past solutions. A nurse scheduling problem from a UK hospital is used for testing. Unlike our previous work that used Genetic Algorithms to implement implicit learning [1], the learning in the proposed algorithm is explicit, i.e. we identify and mix building blocks directly. The Bayesian optimisation algorithm is applied to implement such explicit learning by building a Bayesian network of the joint distribution of solutions. The conditional probability of each variable in the network is computed according to an initial set of promising solutions. Subsequently, each new instance for each variable is generated by using the corresponding conditional probabilities, until all variables have been generated, i.e. in our case, new rule strings have been obtained. Sets of rule strings are generated in this way, some of which will replace previous strings based on fitness. If stopping conditions are not met, the conditional probabilities for all nodes in the Bayesian network are updated again using the current set of promising rule strings. For clarity, consider the following toy example of scheduling five nurses with two rules (1: random allocation, 2: allocate nurse to low-cost shifts). In the beginning of the search, the probabilities of choosing rule 1 or 2 for each nurse is equal, i.e. 50%. After a few iterations, due to the selection pressure and reinforcement learning, we experience two solution pathways: Because pure low-cost or random allocation produces low quality solutions, either rule 1 is used for the first 2-3 nurses and rule 2 on remainder or vice versa. In essence, Bayesian network learns 'use rule 2 after 2-3x using rule 1' or vice versa. It should be noted that for our and most other scheduling problems, the structure of the network model is known and all variables are fully observed. In this case, the goal of learning is to find the rule values that maximize the likelihood of the training data. Thus, learning can amount to 'counting' in the case of multinomial distributions. For our problem, we use our rules: Random, Cheapest Cost, Best Cover and Balance of Cost and Cover. In more detail, the steps of our Bayesian optimisation algorithm for nurse scheduling are: 1. Set t = 0, and generate an initial population P(0) at random; 2. Use roulette-wheel selection to choose a set of promising rule strings S(t) from P(t); 3. Compute conditional probabilities of each node according to this set of promising solutions; 4. Assign each nurse using roulette-wheel selection based on the rules' conditional probabilities. A set of new rule strings O(t) will be generated in this way; 5. Create a new population P(t+1) by replacing some rule strings from P(t) with O(t), and set t = t+1; 6. If the termination conditions are not met (we use 2000 generations), go to step 2. Computational results from 52 real data instances demonstrate the success of this approach. They also suggest that the learning mechanism in the proposed approach might be suitable for other scheduling problems. Another direction for further research is to see if there is a good constructing sequence for individual data instances, given a fixed nurse scheduling order. If so, the good patterns could be recognized and then extracted as new domain knowledge. Thus, by using this extracted knowledge, we can assign specific rules to the corresponding nurses beforehand, and only schedule the remaining nurses with all available rules, making it possible to reduce the solution space. Acknowledgements The work was funded by the UK Government's major funding agency, Engineering and Physical Sciences Research Council (EPSRC), under grand GR/R92899/01. References [1] Aickelin U, "An Indirect Genetic Algorithm for Set Covering Problems", Journal of the Operational Research Society, 53(10): 1118-1126,

Relevância:

100.00% 100.00%

Publicador:

Resumo:

The recently discovered abilities to synthesize single-walled carbon nanotubes and prepare single layer graphene have spurred interest in these sp2-bonded carbon nanostructures. In particular, studies of their potential use in electronic devices are many as silicon integrated circuits are encountering processing limitations, quantum effects, and thermal management issues due to rapid device scaling. Nanotube and graphene implementation in devices does come with significant hurdles itself. Among these issues are the ability to dope these materials and understanding what influences defects have on expected properties. Because these nanostructures are entirely all-surface, with every atom exposed to ambient, introduction of defects and doping by chemical means is expected to be an effective route for addressing these issues. Raman spectroscopy has been a proven characterization method for understanding vibrational and even electronic structure of graphene, nanotubes, and graphite, especially when combined with electrical measurements, due to a wealth of information contained in each spectrum. In Chapter 1, a discussion of the electronic structure of graphene is presented. This outlines the foundation for all sp2-bonded carbon electronic properties and is easily extended to carbon nanotubes. Motivation for why these materials are of interest is readily gained. Chapter 2 presents various synthesis/preparation methods for both nanotubes and graphene, discusses fabrication techniques for making devices, and describes characterization methods such as electrical measurements as well as static and time-resolved Raman spectroscopy. Chapter 3 outlines changes in the Raman spectra of individual metallic single-walled carbon nantoubes (SWNTs) upon sidewall covalent bond formation. It is observed that the initial degree of disorder has a strong influence on covalent sidewall functionalization which has implications on developing electronically selective covalent chemistries and assessing their selectivity in separating metallic and semiconducting SWNTs. Chapter 4 describes how optical phonon population extinction lifetime is affected by covalent functionalization and doping and includes discussions on static Raman linewidths. Increasing defect concentration is shown to decrease G-band phonon population lifetime and increase G-band linewidth. Doping only increases G-band linewidth, leaving non-equilibrium population decay rate unaffected. Phonon mediated electron scattering is especially strong in nanotubes making optical phonon decay of interest for device applications. Optical phonon decay also has implications on device thermal management. Chapter 5 treats doping of graphene showing ambient air can lead to inadvertent Fermi level shifts which exemplifies the sensitivity that sp2-bonded carbon nanostructures have to chemical doping through sidewall adsorption. Removal of this doping allows for an investigation of electron-phonon coupling dependence on temperature, also of interest for devices operating above room temperature. Finally, in Chapter 6, utilizing the information obtained in previous chapters, single carbon nanotube diodes are fabricated and characterized. Electrical performance shows these diodes are nearly ideal and photovoltaic response yields 1.4 nA and 205 mV of short circuit current and open circuit voltage from a single nanotube device. A summary and discussion of future directions in Chapter 7 concludes my work.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Due to increasing integration density and operating frequency of today's high performance processors, the temperature of a typical chip can easily exceed 100 degrees Celsius. However, the runtime thermal state of a chip is very hard to predict and manage due to the random nature in computing workloads, as well as the process, voltage and ambient temperature variability (together called PVT variability). The uneven nature (both in time and space) of the heat dissipation of the chip could lead to severe reliability issues and error-prone chip behavior (e.g. timing errors). Many dynamic power/thermal management techniques have been proposed to address this issue such as dynamic voltage and frequency scaling (DVFS), clock gating and etc. However, most of such techniques require accurate knowledge of the runtime thermal state of the chip to make efficient and effective control decisions. In this work we address the problem of tracking and managing the temperature of microprocessors which include the following sub-problems: (1) how to design an efficient sensor-based thermal tracking system on a given design that could provide accurate real-time temperature feedback; (2) what statistical techniques could be used to estimate the full-chip thermal profile based on very limited (and possibly noise-corrupted) sensor observations; (3) how do we adapt to changes in the underlying system's behavior, since such changes could impact the accuracy of our thermal estimation. The thermal tracking methodology proposed in this work is enabled by on-chip sensors which are already implemented in many modern processors. We first investigate the underlying relationship between heat distribution and power consumption, then we introduce an accurate thermal model for the chip system. Based on this model, we characterize the temperature correlation that exists among different chip modules and explore statistical approaches (such as those based on Kalman filter) that could utilize such correlation to estimate the accurate chip-level thermal profiles in real time. Such estimation is performed based on limited sensor information because sensors are usually resource constrained and noise-corrupted. We also took a further step to extend the standard Kalman filter approach to account for (1) nonlinear effects such as leakage-temperature interdependency and (2) varying statistical characteristics in the underlying system model. The proposed thermal tracking infrastructure and estimation algorithms could consistently generate accurate thermal estimates even when the system is switching among workloads that have very distinct characteristics. Through experiments, our approaches have demonstrated promising results with much higher accuracy compared to existing approaches. Such results can be used to ensure thermal reliability and improve the effectiveness of dynamic thermal management techniques.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Jerne's idiotypic network theory postulates that the immune response involves inter-antibody stimulation and suppression as well as matching to antigens. The theory has proved the most popular Artificial Immune System (AIS) model for incorporation into behavior-based robotics but guidelines for implementing idiotypic selection are scarce. Furthermore, the direct effects of employing the technique have not been demonstrated in the form of a comparison with non-idiotypic systems. This paper aims to address these issues. A method for integrating an idiotypic AIS network with a Reinforcement Learning based control system (RL) is described and the mechanisms underlying antibody stimulation and suppression are explained in detail. Some hypotheses that account for the network advantage are put forward and tested using three systems with increasing idiotypic complexity. The basic RL, a simplified hybrid AIS-RL that implements idiotypic selection independently of derived concentration levels and a full hybrid AIS-RL scheme are examined. The test bed takes the form of a simulated Pioneer robot that is required to navigate through maze worlds detecting and tracking door markers.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

A Bayesian optimisation algorithm for a nurse scheduling problem is presented, which involves choosing a suitable scheduling rule from a set for each nurse's assignment. When a human scheduler works, he normally builds a schedule systematically following a set of rules. After much practice, the scheduler gradually masters the knowledge of which solution parts go well with others. He can identify good parts and is aware of the solution quality even if the scheduling process is not yet completed, thus having the ability to finish a schedule by using flexible, rather than fixed, rules. In this paper, we design a more human-like scheduling algorithm, by using a Bayesian optimisation algorithm to implement explicit learning from past solutions. A nurse scheduling problem from a UK hospital is used for testing. Unlike our previous work that used Genetic Algorithms to implement implicit learning [1], the learning in the proposed algorithm is explicit, i.e. we identify and mix building blocks directly. The Bayesian optimisation algorithm is applied to implement such explicit learning by building a Bayesian network of the joint distribution of solutions. The conditional probability of each variable in the network is computed according to an initial set of promising solutions. Subsequently, each new instance for each variable is generated by using the corresponding conditional probabilities, until all variables have been generated, i.e. in our case, new rule strings have been obtained. Sets of rule strings are generated in this way, some of which will replace previous strings based on fitness. If stopping conditions are not met, the conditional probabilities for all nodes in the Bayesian network are updated again using the current set of promising rule strings. For clarity, consider the following toy example of scheduling five nurses with two rules (1: random allocation, 2: allocate nurse to low-cost shifts). In the beginning of the search, the probabilities of choosing rule 1 or 2 for each nurse is equal, i.e. 50%. After a few iterations, due to the selection pressure and reinforcement learning, we experience two solution pathways: Because pure low-cost or random allocation produces low quality solutions, either rule 1 is used for the first 2-3 nurses and rule 2 on remainder or vice versa. In essence, Bayesian network learns 'use rule 2 after 2-3x using rule 1' or vice versa. It should be noted that for our and most other scheduling problems, the structure of the network model is known and all variables are fully observed. In this case, the goal of learning is to find the rule values that maximize the likelihood of the training data. Thus, learning can amount to 'counting' in the case of multinomial distributions. For our problem, we use our rules: Random, Cheapest Cost, Best Cover and Balance of Cost and Cover. In more detail, the steps of our Bayesian optimisation algorithm for nurse scheduling are: 1. Set t = 0, and generate an initial population P(0) at random; 2. Use roulette-wheel selection to choose a set of promising rule strings S(t) from P(t); 3. Compute conditional probabilities of each node according to this set of promising solutions; 4. Assign each nurse using roulette-wheel selection based on the rules' conditional probabilities. A set of new rule strings O(t) will be generated in this way; 5. Create a new population P(t+1) by replacing some rule strings from P(t) with O(t), and set t = t+1; 6. If the termination conditions are not met (we use 2000 generations), go to step 2. Computational results from 52 real data instances demonstrate the success of this approach. They also suggest that the learning mechanism in the proposed approach might be suitable for other scheduling problems. Another direction for further research is to see if there is a good constructing sequence for individual data instances, given a fixed nurse scheduling order. If so, the good patterns could be recognized and then extracted as new domain knowledge. Thus, by using this extracted knowledge, we can assign specific rules to the corresponding nurses beforehand, and only schedule the remaining nurses with all available rules, making it possible to reduce the solution space. Acknowledgements The work was funded by the UK Government's major funding agency, Engineering and Physical Sciences Research Council (EPSRC), under grand GR/R92899/01. References [1] Aickelin U, "An Indirect Genetic Algorithm for Set Covering Problems", Journal of the Operational Research Society, 53(10): 1118-1126,

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Jerne's idiotypic network theory postulates that the immune response involves inter-antibody stimulation and suppression as well as matching to antigens. The theory has proved the most popular Artificial Immune System (AIS) model for incorporation into behavior-based robotics but guidelines for implementing idiotypic selection are scarce. Furthermore, the direct effects of employing the technique have not been demonstrated in the form of a comparison with non-idiotypic systems. This paper aims to address these issues. A method for integrating an idiotypic AIS network with a Reinforcement Learning based control system (RL) is described and the mechanisms underlying antibody stimulation and suppression are explained in detail. Some hypotheses that account for the network advantage are put forward and tested using three systems with increasing idiotypic complexity. The basic RL, a simplified hybrid AIS-RL that implements idiotypic selection independently of derived concentration levels and a full hybrid AIS-RL scheme are examined. The test bed takes the form of a simulated Pioneer robot that is required to navigate through maze worlds detecting and tracking door markers.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Natural language processing has achieved great success in a wide range of ap- plications, producing both commercial language services and open-source language tools. However, most methods take a static or batch approach, assuming that the model has all information it needs and makes a one-time prediction. In this disser- tation, we study dynamic problems where the input comes in a sequence instead of all at once, and the output must be produced while the input is arriving. In these problems, predictions are often made based only on partial information. We see this dynamic setting in many real-time, interactive applications. These problems usually involve a trade-off between the amount of input received (cost) and the quality of the output prediction (accuracy). Therefore, the evaluation considers both objectives (e.g., plotting a Pareto curve). Our goal is to develop a formal understanding of sequential prediction and decision-making problems in natural language processing and to propose efficient solutions. Toward this end, we present meta-algorithms that take an existent batch model and produce a dynamic model to handle sequential inputs and outputs. Webuild our framework upon theories of Markov Decision Process (MDP), which allows learning to trade off competing objectives in a principled way. The main machine learning techniques we use are from imitation learning and reinforcement learning, and we advance current techniques to tackle problems arising in our settings. We evaluate our algorithm on a variety of applications, including dependency parsing, machine translation, and question answering. We show that our approach achieves a better cost-accuracy trade-off than the batch approach and heuristic-based decision- making approaches. We first propose a general framework for cost-sensitive prediction, where dif- ferent parts of the input come at different costs. We formulate a decision-making process that selects pieces of the input sequentially, and the selection is adaptive to each instance. Our approach is evaluated on both standard classification tasks and a structured prediction task (dependency parsing). We show that it achieves similar prediction quality to methods that use all input, while inducing a much smaller cost. Next, we extend the framework to problems where the input is revealed incremen- tally in a fixed order. We study two applications: simultaneous machine translation and quiz bowl (incremental text classification). We discuss challenges in this set- ting and show that adding domain knowledge eases the decision-making problem. A central theme throughout the chapters is an MDP formulation of a challenging problem with sequential input/output and trade-off decisions, accompanied by a learning algorithm that solves the MDP.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Many different photovoltaic technologies are being developed for large-scale solar energy conversion such as crystalline silicon solar cells, thin film solar cells based on a-Si:H, CIGS and CdTe. As the demand for photovoltaics rapidly increases, there is a pressing need for the identification of new visible light absorbing materials for thin-film solar cells. Nowadays there are a wide range of earth-abundant absorber materials that have been studied around the world by different research groups. The current thin film photovoltaic market is dominated by technologies based on the use of CdTe and CIGS, these solar cells have been made with laboratory efficiencies up to 19.6% and 20.8% respectively. However, the scarcity and high cost of In, Ga and Te can limit in the long-term the production in large scale of photovoltaic devices. On the other hand, quaternary CZTSSe which contain abundant and inexpensive elements like Cu, Zn, Sn, S and Se has been a potential candidate for PV technology having solar cell efficiency up to 12.6%, however, there are still some challenges that must be accomplished for this material. Therefore, it is evident the need to find the alternative inexpensive and earth abundant materials for thin film solar cells. One of these alternatives is copper antimony sulfide(CuSbS2) which contains abundant and non-toxic elements which has a direct optical band gap of 1.5 eV, the optimum value for an absorber material in solar cells, suggesting this material as one among the new photovoltaic materials. This thesis work focuses on the preparation and characterization of In6Se7, CuSbS2 and CuSb(S1-xSex)2 thin films for their application as absorber material in photovoltaic structures using two stage process by the combination of chemical bath deposition and thermal evaporation.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Part 16: Performance Measurement Systems

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Previous work has shown that robot navigation systems that employ an architecture based upon the idiotypic network theory of the immune system have an advantage over control techniques that rely on reinforcement learning only. This is thought to be a result of intelligent behaviour selection on the part of the idiotypic robot. In this paper an attempt is made to imitate idiotypic dynamics by creating controllers that use reinforcement with a number of different probabilistic schemes to select robot behaviour. The aims are to show that the idiotypic system is not merely performing some kind of periodic random behaviour selection, and to try to gain further insight into the processes that govern the idiotypic mechanism. Trials are carried out using simulated Pioneer robots that undertake navigation exercises. Results show that a scheme that boosts the probability of selecting highly-ranked alternative behaviours to 50% during stall conditions comes closest to achieving the properties of the idiotypic system, but remains unable to match it in terms of all round performance.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Regulated Transformer Rectifier Units contain several power electronic boards to facilitate AC to DC power conversion. As these units become smaller, the number of devices on each board increases while their distance from each other decreases, making active cooling essential to maintaining reliable operation. Although it is widely accepted that liquid is a far superior heat transfer medium to air, the latter is still capable of yielding low device operating temperatures with proper heat sink and airflow design. The purpose of this study is to describe the models and methods used to design and build the thermal management system for one of the power electronic boards in a compact, high power regulated transformer rectifier unit. Maximum device temperature, available pressure drop and manufacturability were assessed when selecting the final design for testing. Once constructed, the thermal management system’s performance was experimentally verified at three different power levels.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

An important parameter in integrated optical device is the propagation loss of the waveguide. Its characterization gives the information of the fabrication quality as well as the information of other passive devices on the chip as it is the basic building block of the passive devices. Although, over the last three decades many methods have been developed, there is not a single standard present yet. This paper presents a comparative analysis of the methods existing from the past as well as methods developed very recently in order to provide a complete picture of the pros and cons of different types of methods and from this comparison the best method is suggested according to the authors opinion. To support the claim, apart from the analytical comparison, this paper also presents a comparison performed with the experimental results between the suggested best method which is recently proposed by Massachusetts Institute of Technology (MIT) researchers based on undercoupled all-pass microring structure and the popular cut-back method.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Selling devices on retail stores comes with the big challenge of grabbing the customer’s attention. Nowadays people have a lot of offers at their disposal and new marketing techniques must emerge to differentiate the products. When it comes to smartphones and tablets, those devices can make the difference by themselves, if we use their computing power and capabilities to create something unique and interactive. With that in mind, three prototypes were developed during an internship: a face recognition based Customer Detection, a face tracking solution with an Avatar and interactive cross-app Guides. All three revealed to have potential to be differentiating solutions in a retail store, not only raising the chance of a customer taking notice of the device but also of interacting with them to learn more about their features. The results were meant to be only proof of concepts and therefore were not tested in the real world.

Relevância:

100.00% 100.00%

Publicador:

Resumo:

Combinatorial optimization problems are typically tackled by the branch-and-bound paradigm. We propose to learn a variable selection policy for branch-and-bound in mixed-integer linear programming, by imitation learning on a diversified variant of the strong branching expert rule. We encode states as bipartite graphs and parameterize the policy as a graph convolutional neural network. Experiments on a series of synthetic problems demonstrate that our approach produces policies that can improve upon expert-designed branching rules on large problems, and generalize to instances significantly larger than seen during training.