The rapid advancement of Machine Learning (ML) and Deep Learning (DL) has revolutionized various domains, necessitating efficient and cost-effective ML inference capabilities. Function-as-a-Service (FaaS) has emerged as a promising approach for hosting ML inference services, providing a serverless computing environment that streamlines development cycles and offers scalability and simplified infrastructure management. However, existing autoscaling strategies employed by popular FaaS platforms often overlook critical factors such as response time and tail latency. Additionally, Python's Global Interpreter Lock (GIL) poses challenges for parallel computing in high-request traffic scenarios. This thesis addresses the need for efficient and cost-effective Machine Learning (ML) inference capabilities by exploring batching and autoscaling strategies for Serverless Inference instances. The study proposes a prototype FaaS framework that provides adaptive request batching, reactive autoscaling policies, and SLO monitoring, thus allowing Serverless Inference workloads to meet their SLO targets even during peak traffic. The proposed approach aims to optimize resource utilization, mitigate tail latency, and improve overall system performance. / Master of Science / Machine Learning (ML) and Deep Learning (DL) are advanced techniques that allow computers to learn from data and make predictions or decisions without being explicitly programmed. This has led to significant advancements in various fields. Inference refers to the process of applying a trained ML model to new data to make predictions or extract insights. In the context of ML, there is a growing need for efficient and cost-effective inference capabilities. A new approach called Function-as-a-Service (FaaS) has emerged that can address this need. FaaS is a way of abstracting the server infrastructure away from the developers. This means developers can focus on writing the ML code without worrying about managing the underlying infrastructure. FaaS offers benefits such as scalability, simplified infrastructure management, and faster development cycles. However, existing FaaS platforms face challenges in ensuring fast response times and handling high levels of incoming requests. This thesis aims to address these challenges by proposing a prototype FaaS framework. The framework incorporates adaptive request batching, reactive autoscaling policies, and Service-Level Objectives (SLOs) monitoring. Request batching allows the framework to process multiple requests together, improving efficiency. Autoscaling policies ensure the system dynamically adjusts its resources based on the incoming workload. Monitoring SLOs helps track and meet performance targets, even during peak traffic. By optimizing resource utilization, reducing delays in processing requests, and improving overall system performance, the proposed approach seeks to provide efficient and cost-effective ML inference capabilities in a serverless environment.

Magnetorheological (MR) fluids offer solutions to many engineering challenges. The success of MR fluid is apparent in many disciplines, ranging from the automotive and civil engineering communities to the biomedical engineering community. This well documented success of MR fluids continues to motivate current and future applications of MR fluid. One such application that has been considered recently is MR fluid devices for use in impact and other high velocity applications. In such applications, the fluid environment within the device may be well beyond the scope of our understanding for these fluids. To date, little has been done to explore the suitability of MR fluids in such high velocity and high shear applications. While future applications may expose the fluid to adverse flow conditions, we must also consider current and existing applications which expose the fluid to extreme flow environments. Consider, for example, an MR damper intended for automotive primary suspensions, in which shear rates may exceed 10^5 s^-1. Flow conditions within these dampers far exceed existing fluid behavior characterization. The aim of the current study is to identify the behavior of the fluid under these extreme operating conditions. Specifically, this study intends to identify the behavior of MR fluid subject to high rates of shear and high flow velocities. A high shear rheometer is built which allows for the high velocity testing of MR fluids. The rheometer is capable of fluid velocities ranging from 1 m/s to 37 m/s, with corresponding shear rates ranging from 0.14x10^5 s^-1 to 2.5x10^5 s^-1. Fluid behavior is characterized in both the off-state and the on-state. The off-state testing was conducted in order to identify the high shear viscosity of the fluid. Because the high shear behavior of MR fluid is largely governed by the behavior of the carrier fluid, the carrier fluid behavior was also identified at high shear. Experiments were conducted using the high shear rheometer and the MR fluid was shown to exhibit nearly Newtonian post-yield behavior. A slight thickening was observed for growing shear rates. This slight thickening can be attributed to the behavior of the carrier fluid, which exhibited considerable thickening at high shear. The purpose of the on-state testing was to characterize the MR effect at high flow velocities. As such, the MR fluid was run through the rheometer at various flow velocities and a number of magnetic field strengths. The term "dwell time" is introduced and defined as the amount of time the fluid spends in the presence of a magnetic field. Two active valve lengths were considered, which when coupled to the fluid velocities, generated dwell times ranging from 12 ms to 0.18 ms. The yield stress was found from the experimental measurements and the results indicate that the magnitude of the yield stress is sensitive to fluid dwell time. As fluid dwell times decrease, the yield stress developed in the fluid decreases. The results from the on-state testing clearly demonstrate a need to consider fluid dwell times in high velocity applications. Should the dwell time fall below the response time of the fluid, the yield stress developed in the fluid may only achieve a fraction of the expected value. These results imply that high velocity applications may be subject to diminished controllability for falling dwell times. Results from this study may serve to aid in the design of MR fluid devices intended for high velocity applications. Furthermore, the identified behavior may lead to further developments in MR fluid technology. In particular, the identified behavior may be used to develop or identify an MR fluid well suited for high velocity and high shear applications. / Ph. D.

<p dir="ltr">Multistage adaptive testing (MST) has become one of the most popular test designs for large-scale testing. However, it has some weaknesses such as a larger estimation bias compared to computerized adaptive testing (CAT). On-the-fly multistage adaptive testing (OMST) can balance the advantages and limitations of CAT and MST. Several CAT item selection methods that incorporate response time have been proposed. However, incorporating response time into OMST to select items was rarely studied. The study plans to explore the possibility of applying OMST with response time to the Programme for International Student Assessment to solve the issue of large estimation bias and improve test efficiency.</p>

Thesis (M Sport Sc (Sport Science))--University of Stellenbosch, 2010 / ENGLISH ABSTRACT: The purpose of this study was to determine if there was a difference in mean VMRT between top-level men and women participating in selected ball sports compared to either a normative sample or to a non-athlete sample. VMRT was measured using a new 40-light protocol on the Sport Vision Trainer (SVT). The SVT is a board consisting 80 circular lights controlled by a computer program. The SVT is designed to test visual-motor response time in participants. Data from top-level ball sport players were collected for both men (rugby and cricket) and women (netball and hockey). No significant differences in VMRT were found between the rugby players (n=24) and either the normal sample (n=81) or the non-athlete men (n=24). No significant difference in VMRT were found between the cricket players (n=10) and the non-athlete men. However, the cricket players were found to have significantly slower VMRT than the normal sample of men (n=81). No significant differences in VMRT were found between the netball players (n=19), the hockey players (n=14) and either the normal sample of women (n=84) or the non-athlete women (n=26). The conclusions drawn from this study support the position that VMRT may not be a key performance indicator in top-level ball sport performance and that the expert advantage may be located in other variables, such as anticipation and visual search. / AFRIKAANSE OPSOMMING: Die doel van die studie was om the ondesoek of daar verkille was in die gemiddelde visuele-motoriese reaksie tyd (VMRT) tussen top-vlak bal sport atlete en òf „n normale steekproef òf „n steekproef van nie-atlete. VMRT was gemeet met „ nuwe 40-lig protokol op die Sport Vision Trainer. Inligting van die top-vlak bal sport atlete was ingesamel vir beide mans (rugby en krieket) en dames (netabl en hokkkie) Geen statisties beduidende verskille was gevind vir VMRT tussen die rugby spelers (n=24) en beide van die normale (n=81) of nie-atleet mans steekproef (n=24). Geen statisties beduidende verskille was gevind tussen die krieket spelers (n=10) en die nie-atleet mans nie. Alhoewel die krieket spelers het „n statistie beduidende stadiger VMRT as die normale steekproef mans gehad (n=81). Geen beduidende verskille in VMRT was gevind tussen die netbal spelers (n=19), die hokkie spelers (n=14) en beide van die normale steekproef dames (n=84) of die nie-atletiese dames nie (n=26). Die gevolgtrekking wat gemaak kan word uit die studie ondersteun die standpunt dat VMRT nie „n sleutel prestasie voorspeller in top-vlak bal sportsport prestasie is nie en dat die topvlak speler voordeel deur ander visuele veranderlikes soos antispiasie en “visual search” ondersoek kan word.

Response time (RT) data are able to provide unique insight into both items and examinees regarding speededness and time-demand and should be incorporated into test development practice. To allow test developers to utilize RT information, item RT needs to be summarized into point estimate(PE)(s) that can be understood by content specialists and saved into the item pool. The recent expansion of online testing in K-12 achievement assessments brings opportunities and challenges for measurement experts to investigate and utilize RT information in a context different from that in the majority of literature, in which licensing and certification tests, graduate admission tests, and other applications that incorporate computer-adaptive testing. Using empirical data from four tests in two grade levels of a K-12 standardized achievement assessment, this study explored the empirical distributions of item RT and their fit to five probability distributions, the characteristics of four item RTPEs, and the relationships between item RTPEs and eight item attributes. Based on the principal findings across tests and grades, the empirical distributions of item RT presented widely variable shapes and did not fit any of the five proposed probability distributions; the 90th quantile showed its important capability of capturing and avoiding speededness issues; and the associations between item RTPEs and item attributes proved to be mixed. The generally idiosyncratic findings of this study call for a different perspective and approach to explore RT data and call for more empirical studies to enlighten test development practice in the K-12 standardized achievement assessment field.

The focal objective of this thesis was to examine the potential advantage of introducing the self-schema concept, indexed by response time, into personality assessment. The basic rationale for the use of response time is that a self-schema facilitates response time for self-referent information as it permits people to make assessments easier and automatic. A self-schema is a cognitive structure containing the generic knowledge that people have about themselves, influencing all aspects of the processing of self-relevant information in order to organize, summarize and explain their behavior. Paper I examined the self-schema proposition that the relation between personality score and response time for the Big Five personality factors is curvilinear in accordance with the inverted-U effect. Using more appropriate statistical methods than in previous studies, Study 1 and Study 2 confirmed the existence of the inverted-U effect for all Big Five factors. Thus the results provided support for the self-schema perspective as people scoring low or high on the Big Five traits responded faster than those scoring in the middle. Previous research has shown that the Big Five personality factors Openness to Experience and Agreeableness are powerful predictors of prejudice. The main question in Paper II was whether this prediction could be improved by including a measure of self-schema (schematicity). The results of Study 3 demonstrated that response time significantly improved the prediction of generalized prejudice from the mentioned personality factors and disclosed both an additive and a moderating effect. Thus, the relation between personality trait score and generalized prejudice is moderated by how schematic a person is. Paper III examined the potential linkage between heritability and self-schema. In Study 4, 5, and 6, the relation between heritability and response time for the Big Five personality facets (subfactors) was examined. The results revealed that personality response time is related to personality heritability so that shorter response times are associated with higher heritabilities. Putting the present results into the context of self-schemas, this means that Big Five personality facets with a large heritability on the average would have higher schematicity than those with small heritability estimates. The results of the present thesis extend previous work in the area of self-schema. The findings suggest that self-schema, measured by response time, may be a useful additional tool to fine-tune personality assessment. Also, the findings put emphasis on the importance of considering possible curvilinear relationships and interaction effects in order to better comprehend the rationale underlying self-schemata processing.  Finally, the results imply that the heritability of personality traits should be taken into account when we construct theories and models in personality psychology. The implications of these results are discussed and directions for future research are highlighted.

Är införande och nyttjande av förstainsatsperson (FIP) samhällsekonomiskt lönsamt? En FIP är, i normalfallet, en deltidsbrandman som inte behöver inställa sig vid brandstationen när en olycka har inträffat då denne har ett eget utryckningsfordon och kan på så sätt fortare komma fram till olycksplats. Den här uppsatsen gör en utvärdering av fyra stationsområden i Jönköpings kommun som för FIP-verksamhet. Tidigare studier framställer att FIP har kortare responstid än nästkommande räddningsenhet. Om så är fallet är följden att räddningsarbetet kan påbörjas i ett tidigare skede och således kan det slutliga räddade värdet i en olyckssituation vara av större omfattning än vad som annars vore möjligt. Enligt teorin bör det finnas ett sätt att värdera denna effekt, vilket sedan kan ställas i förhållande till merkostnaden av införande av FIP. Är fördelarna av FIP-verksamhet av större magnitud än nackdelarna är projektet samhällsekonomiskt lönsamt. Metoden använd för att utvärdera FIP i Jönköpings kommun är kostnadsnyttoanalys (CBA), vari analysunderlaget består av 1150 insatser där FIP-rapport ifyllts. Resultatet visar att den årliga nyttan per FIP-enhet, det vill säga per stationsområde, är mellan 242,8 och 652,0 TSEK (ej extrapolerat värde). Den årliga kostnaden är cirka 76,5 TSEK och nettonuvärdet uppgår därför till mellan 166,2 och 509,1 TSEK. Nyttokostnadskvoten är mellan 3,17 och 8,52. Slutsatsen är att införande och nyttjande av FIP är samhällsekonomiskt lönsamt. / Is introduction of first responder economically beneficial to society? A first responder is, most commonly, a part-time firefighter who can directly head for the scene of an accident since he has his own emergency vehicle. This means that the first responder does not have to first go by the fire department, like other part-time firefighters, and thus will be able to help people, or other, in an early stage of the situation. This thesis evaluates four fire departments (areas) in the municipality of Jönköping, which have adopted the concept of first responders into their organizations. Previous studies have shown that first responders have, on average, a shorter response time than the next emergency unit called to a rescue scene. If this is a fact, the rescue work can begin at an earlier stage and thus the final value saved during a rescue operation will be of greater scope than otherwise would be possible. In theory there should be a way to evaluate this effect, which then can be compared with the additional cost of introducing first responders. If the benefits of first responders are of greater magnitude than the disadvantages, the project is economically beneficial to society. The method used to evaluate the project of first responders in the municipality of Jönköping is a cost-benefit analysis (CBA), where the analytical framework consists of 1150 emergency rescues where a first-responder report has been completed. The results show that the annual benefits per first-responder unit, i.e. per station area, is between SEK 242.8 thousand and SEK 652.0 thousand (with no extrapolation). The annual cost is approximately SEK 76.5 thousand. The net present value therefore is between SEK 166.2 thousand and SEK 509.1 thousand. The cost-benefit ratio is between 3.17 and 8.52. The con-clusion is that introduction and use of first responders in the municipality of Jönköping is economic-ally beneficial to society. / Effektsamband i samhället

Next generation data centres are expected to support both advance resource reservation and on-demand access, but the system performance for such a computing environment has not been well-investigated. A reservation request is characterized by a start time, duration, and resource requirement. Discrete event simulation is used to study the performance characteristics of reservation systems. The basic strategy is to accept a request if resources are available and reject the request otherwise. The performance metrics considered are resource utilization and blocking probability. Results showing the impact of input parameters on these performance metrics are presented. It is found that the resource utilization is quite low. Two strategies that can be used to improve the performance for advance reservation are evaluated. The first strategy allows the start time to be delayed up to some maximum value, while the second allows the possibility of non-uniform resource allocation over the duration of the reservation. Simulation results showing the performance improvements of these two strategies are presented. Resources not used by advance reservation are used to support on-demand access. The performance metrics of interest is the mean response time. Simulation results showing the impact of resource availability and its variation over time on the mean response time are presented. These results provide valuable insights into the performance of systems with time-varying processing capacity. They can also be used to develop guidelines for the non-uniform resource allocation strategy for advance reservation in case the reserved resources are used for interactive access.

Next generation data centres are expected to support both advance resource reservation and on-demand access, but the system performance for such a computing environment has not been well-investigated. A reservation request is characterized by a start time, duration, and resource requirement. Discrete event simulation is used to study the performance characteristics of reservation systems. The basic strategy is to accept a request if resources are available and reject the request otherwise. The performance metrics considered are resource utilization and blocking probability. Results showing the impact of input parameters on these performance metrics are presented. It is found that the resource utilization is quite low. Two strategies that can be used to improve the performance for advance reservation are evaluated. The first strategy allows the start time to be delayed up to some maximum value, while the second allows the possibility of non-uniform resource allocation over the duration of the reservation. Simulation results showing the performance improvements of these two strategies are presented. Resources not used by advance reservation are used to support on-demand access. The performance metrics of interest is the mean response time. Simulation results showing the impact of resource availability and its variation over time on the mean response time are presented. These results provide valuable insights into the performance of systems with time-varying processing capacity. They can also be used to develop guidelines for the non-uniform resource allocation strategy for advance reservation in case the reserved resources are used for interactive access.

The purpose of this thesis is to develop and understand dance balance characteristics on various kinds of dance, related positions and shoe types which contribute to dance performance and to understand different balance controls in various groups. The first study was conducted to examine balance into ballet 2nd position between Ballet shoes and Pointe shoes. Eight dancers performed five different conditions in ballet 2nd position (Ballet Flat, Ballet Demi, Pointe Flat, Pointe Demi, and Pointe Toe) and Centre of Pressure (COP) was used to assess balance. The second study was testing balance control and response to perturbations whilst standing on double leg stance dance positions using a moving platform. Four dance positions were performed by eight subjects (Normal Flat, Turnout Flat, Normal Demi and Turnout Demi) and the platform was moved in two different directions (Forward and Backward) at two different speeds (slow and fast). Kinetics, Kinematics and EMG data was taken from this study. The third study was taken to compare balance control and response to perturbations in single leg standing dance positions between eight dancers and eight gymnasts. The subjects performed static single leg balance in Normal Flat, Turnout Flat, Normal Demi and Turnout Demi. Also, perturbed stance trials were collected in anterior, posterior, right and left directions for two dance positions (Normal Flat and Turnout Flat) at two different speeds (slow and fast) on the moving platform. The results from the studies indicate that dancers move in Medial - Lateral direction more than in Anterior - Posterior direction on Demi-pointe and Toe standing by performing plantar flexion during ballet 2nd position. Demi-pointe position may cause longer delay of EMG latencies because CNS is probably sending information already to keep correcting balance on Demi-pointe. Dancers and Gymnasts have different balance controls due to their ways of training in their performance. Dancers generally reacted faster with slow perturbation in Turnout stance than Gymnasts because this is the particular condition which Dancers are training in.