124 research outputs found
Formation of bright central galaxies in massive haloes
No full textGalaxy formation is one of the most active and evolving fields of research in observational astronomy and cosmology. While we know today which physical processes qualitatively regulate galaxy evolution, the precise timing and the behaviour of these processes and their relations to host environments remain unclear. Many interesting questions are still debated: What regulates galaxy evolution? When do massive galaxies assemble their stellar mass and how? Where does this mass assembly occur? This thesis studies the formation and evolution of central galaxies in groups and clusters over the last 9 billion years in an attempt to answer these questions.
Two important properties of galaxy clusters and groups make them ideal systems to study cosmic evolution. First, they are the largest structures in the Universe that have undergone gravitational relaxation and virial equilibrium. By comparing mass distributions among the nearby- and early-Universe clusters, we can measure the rate of the structure growth and formation. Second, the gravitational potential wells of clusters are deep enough that they retain all of the cluster material, despite outflows driven by supernovae (SNe) and active galactic nuclei (AGN). Thus, the cluster baryons can provide key information on the essential mechanisms related to galaxy formation, including star formation efficiency and the impact of AGN and SNe feedback on galaxy evolution. This thesis reports the identification of a large sample of galaxy groups including their optical and X-ray properties. It includes several refereed journal articles, of which five have been included here.
In the first article (Gozaliasl et al. 2014a), we study the distribution and the development of the magnitude gap between the brightest group galaxies and their brightest satellites in our well defined mass-selected sample of 129 X-ray galaxy groups at 0.04 < z < 1.23 in XMM-LSS. We investigate the relation between magnitude gap and absolute r-band magnitude of the central group galaxy and its brightest satellite. Our observational results are compared to the predictions by three semi-analytic models (SAMs) based on the Millennium simulation. We show that the fraction of galaxy groups with large magnitude gaps (e.g. fossils) increases significantly with decreasing redshift by a factor of ∼ 2. In contrast to the model predictions, we show that the intercept of the relation between the absolute magnitude of the brightest groups galaxies (BGGs) and the magnitude gap becomes brighter as a function of increasing redshift. We attribute this evolution to the presence of a younger population of the observed BGGs.
In the second article (Gozaliasl et al. 2016), we study the distribution and evolution of the star formation rate (SFR) and the stellar mass of BGGs over the last 9 billion years, using a sample of 407 BGGs selected from X-ray galaxy groups at 0.04 < z < 1.3 in the XMM-LSS, COSMOS, and AEGIS fields. We find that the mean stellar mass of BGGs grows by a factor of 2 from z = 1.3 to present day and the stellar mass distribution evolves towards a normal distribution with cosmic time. We find that the BGGs are not completely inactive systems as the SFR of a considerable number of BGG ranges from 1 to 1000 M_sun/yr.
In the third article (Gozaliasl et al. 2014b), we study the evolution of halo mass, magnitude gap, and composite (stacked) luminosity function of galaxies in groups classified by the magnitude gap (as fossils, normal/non-fossils, and random groups) using the Guo et al. (2011) SAM. We find that galaxy groups with large magnitude gaps, i.e. fossils (∆M1,2 ≥ 2 mag), form earlier than the non-fossil systems. We measure the evolution of the Schechter function parameters, finding that M∗ for fossils grows by at least +1 mag in contrast to non-fossils, decreasing the number of massive galaxies with redshift. The faint-end slope (α) of both fossils and non-fossils remains constant with redshift. However, φ∗ grows significantly for both type of groups, changing the number of galaxies with cosmic time. We find that the number of dwarf galaxies in fossils shows no significant evolution in comparison to non-fossils and conclude that the changes in the number of galaxies (φ∗) of fossils are mainly due to the changes in the number of massive (M∗) galaxies. Overall, these results indicate that the giant central galaxies in fossils form by multiple mergers of the massive galaxies.
In the fourth article (Khosroshahi et al. 2014), we analyse the observed X-ray, optical, and spectroscopic data of four optically selected fossil groups at z ∼ 0.06 in 2dFGRS to examine the possibility that a galaxy group, which hosts a giant luminous elliptical galaxy with a large magnitude gap, can be associated with diffuse X-ray radiation, similar to that of fossil groups. The X-ray and optical properties of these groups indicate the presence of extended X-ray emission from the hot intra-group gas. We find that one of them is a fossil group, and the X-ray luminosity of two groups is close to the defined threshold for fossil groups. One of the groups is ruled out due to the optical contamination in the input sample.
In the fifth paper (Khosroshahi et al. 2015), we analyse data from the multiwavelength observations of galaxy groups to probe statistical predictions from the SAMs. We show that magnitude gap can be used as an observable parameter to study groups and to probe galaxy formation models.ei saavutettav
Machine Learning in Bankruptcy Prediction: A Literature Review
No full textSubstantial research efforts have focused on the topic of bankruptcy prediction. Researchers have analyze bankruptcy and default events using various statistical and machine learning techniques for risk management. Academics have also employed various data sources and processing techniques. In this thesis, various proposed models since 2017 in the literature are evaluated and compared. A literature review is conducted to compare the use of data and processing techniques. The models are then implemented and compared in a uniform testing environment to determine the most optimal method. The data from 2009 to 2021 on US firms from the Compustat database is used for the experiment. Several features sets, both from the literature and from widely used feature selection methods, are generated and applied in the experiment to determine the most suitable set of features for predicting bankruptcy.
The experiment results have highlighted several insights in relation to the application of machine learning models in bankruptcy prediction. The models trained on Altman’s original features set outperforms those of the the other sets, including recently proposed features sets. This may be due to the diverse set of firms in the experiment, which are from various industries with varying financial conditions. Regarding machine learning models, ensemble methods, random forest, categorical boosting, and gradient boosted decision tree, outperform the other techniques in almost every evaluation metric.
Most of the recently proposed methods show lackluster performances compared to previously employed models. The results encourage future research in a more focused manner, which focuses on firms in a single field or scope, to avoid introducing noises and affecting classifying ability. Furthermore, more research on the interpretability of the models would be beneficial to professionals in the field
Application of Natural Language Processing in Financial News Sentiment Analysis for Stock Price Prediction
No full textThis thesis studies the application of Natural Language Processing (NLP) in the analysis of financial news sentiment and its subsequent impact on stock price prediction. With the increasing complexity of the financial market, the need for advanced computational techniques to predict stock price movement is evident. This study systematically reviews current research findings on the efficacy of NLP methods in analyzing the sentiment of financial news and stock price movements. By conducting a systematic literature review on the research from the past six years, this review emphasizes the development of sentiment analysis and NLP techniques and evaluates their predictive power.
A total number of 33 papers were chosen for this review. Key findings suggest that due to the recent advancement, particularly the introduction of the transformer model, the focus of NLP in stock prediction has shifted from traditional statistical-based feature representation to learning-based embedding methods.
The conclusion addresses the potential of sentiment analysis as a predictive tool and suggests directions for future research. This emphasizes the need for innovative NLP applications in the financial domain to enhance investment strategies and market understanding
Prediction of Building Energy Consumption Using Machine Learning
No full textBuilding energy consumption (BEC), which is responsible for more than 36% of global energy consumption and greenhouse gas emissions, poses a significant challenge to global environmental sustainability. Building energy consumption prediction (BECP) is critical for optimizing building design and improving building management to reduce energy usage and CO2 emissions. Although there are numerous studies on ML-based BECPs, and many literature reviews have emerged from these studies; a comprehensive evaluation of deep reinforcement learning (DRL) methods for BECP is missing in these literature reviews. To fill this gap, our state-of-the-art literature review provides a detailed discussion on the performance and potential of DRL methods applied to BECP, while updating the optimal ML model and effective features from the latest research on ML-based BECP from 2020 to 2024. This study presents a review of nine high-quality articles, each with an average annual citation rate of almost 50, covering 31 ML algorithms, with a focus on the analysis of applications, data properties, and ML methods in the selected articles. This study identifies the kCNN-LSTM model as the most effective ML method, achieving a root mean square error (RMSE) of only 0.0036 and an average computational time of merely 68.33 seconds, demonstrating an excellent balance between prediction accuracy and computational efficiency. This study also highlights the substantial potential of DRL to improve the accuracy of BECP and extend to other fields of prediction, despite its higher computational time compared to some other ML methods
Machine Learning and Deep Learning in Gravitational Wave Detection: Glitch Classification
No full textGravitational waves, which are rippling effects in space and time created by the rapid acceleration of massive objects in the cosmos, can dramatically transform humankind's comprehension of the universe. Nevertheless, gravitational wave detection faces many challenges, amongst which, a prominent issue is noise mitigation. Background noise, also known as `glitches', affects data quality and creates false detection. Therefore, classifying glitches is an important process for mitigating noise. This process could be significantly improved in terms of efficiency and accuracy with implementing a variety of machine learning (ML) and deep learning (DL) methods. The work aims to assess ML and DL methods that have been developed or adapted to the glitch classification task. This is in order to determine if the size of the dataset would affect the model performance; identify the best performing algorithm; and lastly, assess areas requiring more improvement or further research. The thesis has shown that deep transfer learning is the best performing method to date, with the two algorithms ResNet50 and InceptionV3 yielding the highest accuracy of 98.84\%. It is also determined that dataset size produces no effect on model accuracy. Lastly, it has been established that there should be more and better representations of actual gravitational wave signals in the dataset for a better prospect in detecting more types of gravitational waves, such as continuous and burst
Artificial Intelligence for Prostate Cancer Screening and Diagnosis: A State-of-the-art Review
No full textProstate cancer (PCa) is one of the most prevalent forms of cancer affecting men worldwide. This thesis conducts a state-of-the-art review of AI-based approaches for PCa screening and diagnosis, focusing on three major types of input data: prostate-specific antigen (PSA) screening data, magnetic resonance images (MRIs), and histopathology images. The purpose of this thesis is to determine the current extent of research into AI for PCa screening and diagnosis, highlighting recent advancements, future prospects, and remaining challenges.
The selected articles show wide variability in study designs, data sources, model architectures, and evaluation approaches. Key findings suggest that even though recent research mostly focuses on AI for medical image analysis, PSA data still carries valuable information and can be combined with MRI data and other clinical variables to improve diagnostic performance. AI models inputting MRIs have demonstrated diagnostic performance surpassing the PI-RADS (prostate imaging-reporting and data system) on several coarse-level PCa classification tasks. Similarly, AI models employed for Gleason grading of histopathology images have been reported to match pathologists' performance and to improve inter-rater agreement.
In addition to the literature review, an analysis consisting of 11 articles was conducted to identify the relationship between the AUC values, cohort sizes, and model types of models differentiating between clinically significant PCa (csPCa) and non-csPCa using MRIs. The analysis reveals a statistically significant positive correlation (Kendall's tau = 0.673, 95% CI: 0.348-0.944, p = 0.020) between the AUCs and the cohort sizes, suggesting that increasing cohort size might have a positive impact on performance. In addition, permutation tests with the t-test statistic indicate a statistically significant difference (t = 2.297, p = 0.032) between the mean AUC values of traditional machine learning models and deep learning models, suggesting that deep learning models might be capable of achieving better performance than traditional machine learning models on such medical image analysis tasks.
Despite the great potential and significant progress of research into AI for PCa, several challenges remain, including the shortage of data, label noise, and wide variability in study implementation and evaluation. These problems call for the application of novel machine-learning techniques and collaborative research endeavors
Machine Learning Techniques for Stock Market Prediction: A State-of-the-Art Review
No full textMachine learning techniques for stock market prediction have been the subject of research and application for several decades. These techniques are currently undergoing rapid evolution, with researchers constantly refining existing models and developing new approaches to enhance predictive capabilities. Therefore, staying well-informed with the latest developments in this field is an important but also challenging task.
To navigate the extent of current research on machine learning methods for stock market prediction, this thesis conducts a state-of-the-art review on seventeen relevant and highly cited papers, published between 2019 to 2024. The review offers two main contributions: (1) a comprehensive examination of 112 unique feature variables employed and (2) a review of machine learning methods implemented. Additionally, the review provides a yearly distribution of the utilized machine learning methods, a summary of the best machine learning models, and a discussion of the latest developments in the field.
The findings of the review indicate rapid and significant advancements in the field of stock market forecasting over the last six years. The primary trend in the reviewed papers is the dominance of deep learning techniques, demonstrating superior forecasting power over traditional machine learning methods. Deep learning classifiers were frequently utilized for sentiment analysis of textual data. Furthermore, ensemble learning has been integrated into to deep learning models to combine the predictive capabilities of multiple deep learning predictors. In terms of the utilized data, the majority of the employed variables were technical indicators, whereas no fundamental variable was used. Papers published after 2021 generally utilized fewer variables, and there has also been a growing interest in integrating textual data extracted from different media sources
State-of-the-Art Transiting Exoplanet Detection Algorithms
No full textExoplanets are planets outside solar system. Some of them may host life or be habitable, and each tells us something new about our own solar system. Most prominent astronomical method of exoplanet detection is currently transit method, which aims to detect regular planetary eclipses of stars (i.e., transits). Transiting exoplanet surveys such as Kepler and TESS observe thousands of stars over yearslong periods, thus producing a substantial amount of data in need of efficient and reliable automated analysis.
This thesis reviews modern algorithms that aim to distinguish true transits from unrelated events in the data. The algorithms are compared by their architecture, performance and applicable missions to determine the state-of-the-art and identify the best approaches. The results indicate that ExoMiner V1.2 (Valizadegan et al., 2023) and Astronet-Triage-v2 (Tey et al., 2023) outperform other models for Kepler and TESS data, accordingly. In general, deep learning methods, such as convolutional neural networks, are the best tools for the problem, with potential for even further improvement from self-attention-based transformer models. This study mainly used qualitative analysis, and further research could focus on quantitative comparison of performance on newest datasets
Chandra centres for COSMOS X-ray galaxy groups: differences in stellar properties between central dominant and offset brightest group galaxies
No full textWe present the results of a search for galaxy clusters and groups in the ∼2 deg^2 of the COSMOS field using all available X-ray observations from the XMM–Newton and Chandra observatories. We reach an X-ray flux limit of 3×10^(−16) erg cm^(−2)s^(−1) in the 0.5–2 keV range, and identify 247 X-ray groups with M_(200c) = 8×10^(12) -3×10^(14)M⊙ at a redshift range of 0.08 ≤ z < 1.53, using the multiband photometric redshift and the master spectroscopic redshift catalogues of the COSMOS. The X-ray centres of groups are determined using high-resolution Chandra imaging. We investigate the relations between the offset of the brightest group galaxies (BGGs) from halo X-ray centre and group properties and compare with predictions from semi-analytic models and hydrodynamical simulations. We find that BGG offset decreases with both increasing halo mass and decreasing redshift with no strong dependence on the X-ray flux and SNR. We show that the BGG offset decreases as a function of increasing magnitude gap with no considerable redshift-dependent trend. The stellar mass of BGGs in observations extends over a wider dynamic range compared to model predictions. At z < 0.5, the central dominant BGGs become more massive than those with large offsets by up to 0.3 dex, in agreement with model prediction. The observed and predicted log-normal scatter in the stellar mass of both low- and large-offset BGGs at fixed halo mass is ∼0.3 dex
- …
