Modern neural networks are powerful predictive models. However, when it comes to recognizing that they may be wrong about their predictions and measuring the certainty of beliefs, they perform poorly. For one of the most common activation functions, the ReLU and its variants, even a well-calibrated model can produce incorrect but high confidence predictions. In the related task of action recognition, most current classification methods are based on clip-level classifiers that densely sample a given video for non-overlapping, same sized clips and aggregate the results using an aggregation function - typically averaging - to achieve video level predictions. While this approach has shown to be effective, it is sub-optimal in recognition accuracy and has a high computational overhead. To mitigate both these issues, we propose the confidence distillation framework to firstly teach a representation of uncertainty of the teacher to the student and secondly divide the task of full video prediction between the student and the teacher models. We conduct extensive experiments on three action recognition datasets and demonstrate that our framework achieves state-of-the-art results in action recognition accuracy and computational efficiency. / Thesis / Master of Science (MSc) / We devise a distillation loss function to train an efficient sampler/classifier for video-based action recognition tasks.

京都大学 / 新制・課程博士 / 博士(経済学) / 甲第24374号 / 経博第661号 / 新制||経||302(附属図書館) / 京都大学大学院経済学研究科経済学専攻 / (主査)教授 西山 慶彦, 教授 江上 雅彦, 教授 秋田 祐哉 / 学位規則第4条第1項該当 / Doctor of Economics / Kyoto University / DGAM

Magister Commercii - MCom / The efficient market hypothesis (EMH) is a controversial theory in Finance. Advocates of the EMH argue that it provides a basis for understanding financial markets while critics suggest that the hypothesis is unreasonable in its assumptions of the real function of these markets. Although the EMH may not be perfect, it provides a sufficient baseline against which financial markets may be analysed. Over the past couple of years, academics have broadly examined the EMH in both developing and developed financial markets. However, limited research has been done on African markets. Therefore, this study examines the weak-form EMH of the Johannesburg Stock Exchange (JSE) after 2008 to ascertain the impact the 2008 global financial crisis had on its efficiency. This study analysed the JSE using weekly and monthly returns of the three major indices (RESI 10, FINI 15, INDI 25) as well as the individual companies under these indices from 30th January 2009 to 30th January 2019. Analysis was carried using various statistical tests i.e., runs test, variance ratio test, unit root tests, and a GARCH model which revealed mixed results. Results of the unit root tests (ADF and PP) confirm that the JSE is weak-form efficient when both the weekly and monthly data of the indices and individual companies are analysed. The results of the runs test reveal that all the weekly and monthly data apart from the weekly data of the companies under RESI 10 index exhibit weak-form efficiency. The variance ratio test confirms weak-form inefficiency when weekly data is used while the monthly data confirms weak form efficiency of the JSE and shows that the market moves from periods of efficiency to periods of relative predictability. The results of the GARCH model on the other hand confirm the weak-form efficiency of the JSE when both the weekly and monthly data of the indices are analysed.

No description available.

No description available.

This thesis investigates the price discovery process between the stock and the credit default swap market (CDS). We link the financial theory of efficient markets and the underlying models and conditions involved in CDSs, the stock market and financial crashes. This study uses publicly listed firms and the European market CDS series to construct a matched stock portfolio and uses financial data collected between the years 2019 to 2021. The purpose is to better understand the price discovery process during a potential new type of crisis in modern financial history. It could potentially allow portfolio managers, traders, arbitrageurs and stakeholders who monitor systematic indices to gauge the level of risk in the overall economy. It can also better inform regulators about how the CDS and the stock market reacted to each other during the COVID-19 pandemic. This deductive and quantitative research is based on secondary data gathered from the Eikon financial database. It uses a vector autoregressive model to test a hypothesis regarding the price discovery process between the stock and CDS portfolios.  Our results show that when using only the variables for the CDS and stock market, both variables cause each other, which is to say a feedback effect is present between the CDS Europe index and the matched portfolio of stocks. When adding the three control variables, the stock variable no longer causes the CDS variable, while the CDS variable still causes the stock variable. We conclude that the European credit default swap index leads the matched portfolio of stocks in the price discovery process with our chosen variables.

Due to fundamental physical limitations, conventional digital circuits have not been able to scale at the pace expected from Moore’s law. In addition, computationally intensive applications such as neural networks and computer vision demand large amounts of energy from digital circuits. As a result, energy efficient alternatives are needed in order to provide continued performance scaling. Analog circuits have many well known benefits: the ability to store more information onto a single wire and efficiently perform mathematical operations such as addition, subtraction, and differential equation solving. However, analog computing also comes with drawbacks such as its sensitivity to process variation and noise, limited scalability, programming difficulty, and poor compatibility with digital circuits and design tools. We propose to leverage the strengths of analog circuits and avoid its weaknesses by using digital circuits and time-encoded computation. Time-encoded circuits also operate on continuous data but are implemented using digital circuits. We propose a novel scalable general purpose analog processor using time-encoded circuits that is well suited for emerging applications that require high numeric precision. The processor’s datapath, including time-domain register file and function units are described. We evaluate our proposed approach using an implementation that is simulated with a 0.18µm TSMC process and demonstrate that this approach improves the performance of a scientific benchmark by 4x compared against conventional analog implementations and improves energy consumption by 146x compared against digital implementations.

Today's high performance aircraft are operating in expanded flight envelopes, often maneuvering at high angular rates at high angles-of-attack, even above maximum lift. Current aerodynamic models are inadequate in predicting flight characteristics in the expanded envelope, such as rapid aircraft departures and other unusual motions. Unsteady flows of aircraft are of real concern. The ability to accurately measure aerodynamic loads directly impacts the ability to accurately model and predict flight. Current wind tunnel testing techniques do not adequately address the data fidelity of a test point under the influence of fluctuating loads and moments. Additionally, forced oscillation test techniques, one of the primary tools used to develop dynamic models, do not currently provide estimates of the uncertainty of the results during an oscillation cycle. Further, in testing models across a range of flight conditions, there are frequently parts of the envelope which are well behaved and require few data points to arrive at a sound answer, and other parts of the envelope where the responses are much more active and require a large sample of data to arrive at an answer with statistical significance. Currently, test methods do not factor changes of flow physics into data acquisition schemes, so in many cases data are obtained over more iterations than required, or insufficient data may be obtained to determine a valid estimate. Methods of providing a measure of data integrity for static and forced oscillation test techniques are presented with examples. A method for optimizing required forced oscillation cycles based on decay of uncertainty gradients and balance tolerances is also presented. / Master of Science

No / Sewer and gully flooding have become major causes of pollution particularly in the residential areas majorly caused by blockages in the water system and drainages. An effective way of avoiding this problem will be by deploying some mechanism to monitor gully pot water level at each point in time and escalating unusual liquid levels to the relevant authorities for prompt action to avoid a flooding occurrence. This paper presents a low cost power efficient gully pot liquid level monitoring technique. Passive RFID tags are deployed and signal variation from the Alien Reader Software are used to effectively estimate the level of liquid in the gully pot. The experimental set up is presented and an expository presentation is made of the passive tag design, modelled and simulated and adopted for same application.

No description available.