| 1. |
- Brännvall, Rickard, 1975-
(författare)
-
Machine learning based control of small-scale autonomous data centers
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The low-latency requirements of 5G are expected to increase the demand for distributeddata storage and computing capabilities in the form of small-scale data centers (DC)located at the edge, near the interface between mobile and wired networks. These edgeDC will likely be of modular and standardized designs, although configurations, localresource constraints, environments and load profiles will vary and thereby increase theDC infrastructure diversity. Autonomy and energy efficiency are key objectives for thedesign, configuration and control of such data centers. Edge DCs are (by definition)decentralized and should continue operating without human intervention in the presenceof disturbances, such as intermittent power failures, failing components and overheating.Automatic control is also required for efficient use of renewable energy, batteries and theavailable communication, computing and data storage capacity.These objectives demand data-driven models of the internal thermal and electricprocesses of an autonomous edge DC, since the resources required to manually defineand optimize the models for each DC would be prohibitive. In this thesis machinelearning methods that are implemented in a modular design are evaluated for thermalcontrol of such modular DCs. Experiments with small server clusters are presented, whichwere performed in order to investigate what parameters that are important in the designof advanced control strategies for autonomous edge DC. Furthermore, recent transferlearning results are discussed to understand how to develop data driven models thatcan be deployed to modular DC in varying configurations and environmental contextswithout training from scratch.The first study demonstrates how a data driven thermal model for a small clusterof servers can be calibrated to sensor data and used for constructing a model predictivecontroller for the server cooling fan. The experimental investigations of cooling fancontrol continues in the next study which explores operational sweet-spots and energyefficient holistic control strategies. The machine learning based controller from the firststudy is then re-purposed to maintain environmental conditions in an exhaust chamberfavourable for drying apples, as part of a practical study how excess heat produced bycomputation can be used in the food processing industry. A fourth study describes theRISE EDGE lab - a test bed for small data centers - built with the intention to exploreand evaluate related technologies for micro-grids with renewable energy and batteries,5G connectivity and coolant storage. Finally the last work presented develops the modelfrom the first study towards an application for thermal based load balancing.
|
|
| 2. |
- Brännvall, Rickard, et al.
(författare)
-
Modular and Transferable Machine Learning for Heat Management and Reuse in Edge Data Centers
- 2023
-
Ingår i: Energies. - : MDPI. - 1996-1073. ; 16:5
-
Tidskriftsartikel (refereegranskat)abstract
- This study investigates the use of transfer learning and modular design for adapting a pretrained model to optimize energy efficiency and heat reuse in edge data centers while meeting local conditions, such as alternative heat management and hardware configurations. A Physics-Informed Data-Driven Recurrent Neural Network (PIDD RNN) is trained on a small scale-model experiment of a six-server data center to control cooling fans and maintain the exhaust chamber temperature within safe limits. The model features a hierarchical regularizing structure that reduces the degrees of freedom by connecting parameters for related modules in the system. With a RMSE value of 1.69, the PIDD RNN outperforms both a conventional RNN (RMSE: 3.18), and a State Space Model (RMSE: 2.66). We investigate how this design facilitates transfer learning when the model is fine-tuned over a few epochs to small dataset from a second set-up with a server located in a wind tunnel. The transferred model outperforms a model trained from scratch over hundreds of epochs.
|
|
| 3. |
- Jönsson, L, et al.
(författare)
-
Analyzing overall survival in randomized controlled trials with crossover and implications for economic evaluation
- 2014
-
Ingår i: Value in Health. - : Wiley: No OnlineOpen / Elsevier. - 1098-3015. ; 17:6, s. 707-713
-
Tidskriftsartikel (refereegranskat)abstract
- Background: Offering patients in oncology trials the opportunity to cross over to active treatment at disease progression is a common strategy to address ethical issues associated with placebo controls but may lead to statistical challenges in the analysis of overall survival and cost-effectiveness because crossover leads to information loss and dilution of comparative clinical efficacy. Objectives: We provide an overview of how to address crossover, implications for risk-effect estimates of survival (hazard ratios) and cost-effectiveness, and how this influences decisions of reimbursement agencies. Two case studies using data from two phase III sunitinib oncology trials are used as illustration. Methods: We reviewed the literature on statistical methods for adjusting for crossover and recent health technology assessment decisions in oncology. Results: We show that for a trial with a high proportion of crossover from the control arm to the investigational arm, the choice of the statistical method greatly affects treatment-effect estimates and cost-effectiveness because the range of relative mortality risk for active treatment versus control is broad. With relatively frequent crossover, one should consider either the inverse probability of censoring weighting or the rank-preserving structural failure time model to minimize potential bias, with choice dependent on crossover characteristics, trial size, and available data. A large proportion of crossover favors the rank-preserving structural failure time model, while large sample size and abundant information about confounding factors favors the inverse probability of censoring weighting model. When crossover is very infrequent, methods yield similar results. Conclusions: Failure to correct for crossover may lead to suboptimal decisions by pricing and reimbursement authorities, thereby limiting an effective drug's potential.
|
|
| 4. |
|
|
| 5. |
|
|
| 6. |
- Redig, Josefine, et al.
(författare)
-
Real-world cost-effectiveness of targeted therapy in metastatic renal cell carcinoma in Sweden : a population-based retrospective analysis
- 2019
-
Ingår i: Cancer Management and Research. - Auckland : Dove medical press. - 1179-1322. ; 11, s. 1289-1297
-
Tidskriftsartikel (refereegranskat)abstract
- Objective: To explore cost-effectiveness of targeted therapies (TTs) in the treatment of metastatic renal cell carcinoma (mRCC) in a real-world context using a nationwide population-based approach.Methods: Data on patients diagnosed with mRCC between 2002 and 2012 were extracted from Swedish national health data registers. To facilitate comparisons of patients diagnosed before and after TT introduction to the market, three cohorts were derived: pre-TT introduction (preTT), patients diagnosed 2002-2005; early TT introduction (TTi), patients diagnosed 2006-2008; and late TT introduction (TTii), which was limited to patients diagnosed 2009-2010 to ensure availability of total health care resource utilization (HCRU) data. Patients were followed until end of 2012. The value of TTs across cohorts was estimated using mean HCRU costs per life-year (LY) gained. Data on HCRU were obtained through national health registers for dispensed medication and inpatient and outpatient care, and the associated costs were estimated using the Lin method to account for censoring. LYs gained were defined as the difference in mean survival over the study period.Results: The preTT, TTi, and TTii cohorts consisted of 1,366, 1,158, and 806 patients, respectively. Mean survival in years from mRCC diagnosis was 1.45 in the preTT cohort, 1.62 in the TTi cohort, and 1.83 in the TTii cohort. The respective mean total HCRU cost per patient over the study period was US$16,894, US$29,922, and US$30,037. The cost per LY gained per cohort was US$78,656 for TTi vs preTT, US$34,132 for TTii vs preTT, and US$523 for TTii vs TTi.Conclusion: Given common willingness-to-pay per LY gained thresholds, this study in a real-world population suggests the use of TTs in the Swedish mRCC population is increasingly cost-effective over time.
|
|
| 7. |
- Sandin, Olle, 1989-
(författare)
-
Predicting Sheared Edge Characteristics of High Strength Steels
- 2022
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- An efficient way of reducing CO2 emissions from the transportation sector is to reduce the vehicle weights, i.e. lightweighting. A common strategy for lightweighting of vehicles is to replace the steels used to build structural parts of the vehicle, usually manufactured by metallic sheets, with stronger, advanced high strength steel (AHSS) grades. Using stronger steel grades enables down-gauging while the structural integrity of the parts remain unchanged. However, the increase in strength of AHSS typically comes with a loss of ductility, affecting their forming properties. A common AHSS manufacturing defect is edge cracking occurring when a sheared edge (damaged by the shearing operation) is bent or stretched. It is known in the sheet metal forming industry that the shear cutting process introduces damage, in terms of micro-cracks and notches, to sheared edges from which edge cracks can grow. Conventional forming analyses do not include the effects from sheared edge damage and therefore can not predict edge cracking during forming. Numerical modelling of the shear cutting process can aid the understanding of sheared edge damage and how it affects the AHSS edge cracking phenomena.This thesis presents experimental and numerical methods for calibration of acommercial damage- and failure model, intended for shear cutting simulations. Crack initiation and propagation govern the shear cutting process of AHSS sheets. Therefore, a commercial numerical damage- and failure model was studied regarding its ability to predict shear edge damage. The investigation shows that the damage and failure model has limitations concerning prediction of crack initiation, thus concluding that modelling of processes including formation of cracks using the said damage- and failure model risks to generate erroneous results. This phenomena was also seen in modelling of shear cutting, where the crack-driven fracture process following burnish formation was delayed. Through sensitivity analysis of uncalibrated areas on the failure locus could accurate correlation between numerical and experimental cut edge morphology be obtained. Such results show that additional calibration experiments are necessary, but also the need for development of stress-state dependent failure modelling of AHSS that includes the effect from cracks.
|
|
| 8. |
- Sandin, Rickard
(författare)
-
Heterogeneity in oligopoly : theories and tests
- 1996
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- This dissertation consists of five separate chapters, of which the three first three are empirical and the last two are theoretical. Chapter 1: Competition in Interrelated Markets: An Empirical Study. (Co-authored with Marcus Asplund.)This chapter studies competition in small, concentrated and interrelated markets. Our data set consists of price information from 543 driving schools in 250 local markets in Sweden, which gives a large sample to test hypotheses on how market structure influences competition. The results show that if prices in nearby markets are low, and the distances to them are short, it reduces prices, as suggested in models of spatial competition. Moreover, we find that prices in closely located markets are interdependent. It is also shown that prices are increasing in firm concentration within a market, as most theories of oligopoly predict. Chapter 2: Estimating the Number of Firms and Capacity in Small Markets. (Co-authored with Marcus Asplund.)Many oligopoly theories predict that there will be a positive correlation between market size and the equilibrium number of firms, and some also imply that competition is more intense in larger markets. We test these predictions with a sample of 535 driving schools in 249 markets. With an ordered Probit, a Tobit, and a Poisson model we estimate the relation between the number of firms, capacity, and market size. We find a strong positive correlation between market size and the number of firms. The results show that the per firm market size is increasing in the number of firms in the market. The market size per capacity unit is smaller in large markets. Since the industry produces a fairly homogenous good, we argue that this is evidence that competition is increasing in market size. Chapter 3: The Survival of New Products. (Co-authored with Marcus Asplund.)We study the product turnover in an industry and, in particular, the survival of new products. The data set consists of monthly sales of all products sold in the Swedish beer market over the time period of 1989-1995. The death rates of newly introduced products are high - out of 199 products an estimated 25 percent were withdrawn within 18 months and 50 percent within approximately 48 months. We use parametric duration models with time varying covariates to estimate survival functions. Our results show that products with low and decreasing market shares have higher hazard rates. Moreover, the hazard rates are dependent on the characteristics of the producer. Products from firms with a large number of other products, and (to a lesser extent) the largest market shares are more likely to be withdrawn. Chapter 4: Uniform Subsidy Reductions in International Oligopoly.This chapter studies the effect of production subsidies used as strategic instruments by two rivalling countries whose firms differ in production efficiency. In particular, it examines the welfare effects of a uniform subsidy reduction from the Cournot-Nash equilibrium under different assumptions regarding technology and taste. It is found that the net exporter (usually the efficient country) gains while the net importer (usually the inefficient country) loses from a uniform subsidy reduction. Results show that a non-linear demand function or marginal cost functions with different slopes across countries is necessary to obtain an increase in total welfare.Chapter 5: Subsidies in Oligopoly Markets: A Welfare Comparison between Symmetric and Asymmetric Costs. (Co-authored with Stephen F. Hamilton.)This chapter studies welfare effects of uniform production subsidies in oligopoly markets, comparing cases of symmetric and asymmetric costs. Cost asymmetry reduces the impact relative to the symmetric-cost case if the deman function is concave and magnifies the impact if demand is convex. The welfare difference increases with the degree of market power and the cost differential in the industry.
|
|
