• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 49
  • 11
  • 4
  • 2
  • 2
  • 2
  • 1
  • 1
  • 1
  • 1
  • 1
  • Tagged with
  • 87
  • 87
  • 55
  • 24
  • 24
  • 19
  • 15
  • 13
  • 13
  • 11
  • 10
  • 8
  • 7
  • 7
  • 7
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
21

Meta-learning strategies, implementations, and evaluations for algorithm selection /

Köpf, Christian Rudolf. January 1900 (has links)
Thesis (doctorat)--Universität Ulm, 2005. / Includes bibliographical references (p. 227-248).
22

Αυτοπροσαρμοζόμενος αλγόριθμος για παιχνίδι μερικούς πληροφόρησης

Σώρρος, Νικόλαος 24 October 2012 (has links)
Στη παρούσα διπλωματική εργασία παρουσιάζεται η ανάπτυξη κώδικα σε γλώσσα προγραμματισμού Python με σκοπό να παίζει το παιχνίδι Bluff. Αναλυτικότερα το Bluff ανήκει στη κατηγορία των παιχνιδιών μερικούς πληροφόρησης και εκδοχές του περιλαμβάνουν το στοιχείο της τύχης άρα είναι και στοχαστικό. Στην ίδια κατηγορία παιχνιδιών εντάσεται και το πόκερ στο οποίο διεξάγεται εντονη ερευνητική δραστηριότητα αυτή τη περίοδο. Οι δυσκολίες που παρουσιάζει το εγχείρημα της κατασκευης ενός τέτοιου αλγόριθμου εγκειται στο μεγάλο χώρο καταστασης του παιχνιδιού και στην αδυναμια εφαρμογης της τεχνικής min max λόγω της δομής του παιχνιδιού. Επίσης ενας επιτυχημένος παίχτης bluff θα πρέπει να αναγνωρίζει ποτε ο αντίπαλος μπλοφάρει καθώς και να μπλοφάρει ο ίδιος. Τέλος όπως και στο ποκερ για να γίνεις μετρ στο παιχνίδι θα πρέπει να μεταβάλεις τη στρατηγική σου ανάλογα με τον αντίπαλο, θα πρέπει να εκμεταλεύεσαι τα λάθη του και ταυτόχρονα να μη γίνεσαι προβλέψιμος. Ο κώδικας μας εχει 3 versions. Στη πρώτη version ενας απλος μηχανισμός που στηρίζεται στους κανονες του παιχνιδίου υλοποιείται και εξετάζεται η επιτυχια του. Στη δευτερη εκδοση εισαγουμε το στοιχειο της μπλόφας ενώ στη τρίτη αφου μοντελοποιήσουμε τον αντίπαλο, λαμβάνουμε αποφάσεις με βάση αυτη τη μοντελοποίηση. / This diploma thesis deals with the problem of developing an algorithm that can play the game of Bluff. The programming language that is used is Python. Concretely the game of bluff belongs into the category of partial information games and some variations involve luck which makes it also stochastic. Intense research is conducted in poker which belongs to the same family of games. The main difficulty is the huge state space of these games due to uncertainty and the deficit of the min-max method. In addition a succesfull bluff player must recognize when the opponent is bluffing and must make bluffs on his own. One last thing that this game requires is to have dynamic strategies which means being able to change your strategy according to the opponent in order to maximize your wining by exploiting his errors. The algorithm builded has 3 versions. The first one simulated a beginner that sticks to the rules, makes no bluffs and raises according to probabilities. The second version introduces bluffing. The final version includes opponent modeling and making decision based on that.
23

Text Classification of Legitimate and Rogue online Privacy Policies : Manual Analysis and a Machine Learning Experimental Approach

Rekanar, Kaavya January 2016 (has links)
No description available.
24

A Boosted-Window Ensemble

Elahi, Haroon January 2014 (has links)
Context. The problem of obtaining predictions from stream data involves training on the labeled instances and suggesting the class values for the unseen stream instances. The nature of the data-stream environments makes this task complicated. The large number of instances, the possibility of changes in the data distribution, presence of noise and drifting concepts are just some of the factors that add complexity to the problem. Various supervised-learning algorithms have been designed by putting together efficient data-sampling, ensemble-learning, and incremental-learning methods. The performance of the algorithm is dependent on the chosen methods. This leaves an opportunity to design new supervised-learning algorithms by using different combinations of constructing methods. Objectives. This thesis work proposes a fast and accurate supervised-learning algorithm for performing predictions on the data-streams. This algorithm is called as Boosted-Window Ensemble (BWE), which is invented using the mixture-of-experts technique. BWE uses Sliding Window, Online Boosting and incremental-learning for data-sampling, ensemble-learning, and maintaining a consistent state with the current stream data, respectively. In this regard, a sliding window method is introduced. This method uses partial-updates for sliding the window on the data-stream and is called Partially-Updating Sliding Window (PUSW). The investigation is carried out to compare two variants of sliding window and three different ensemble-learning methods for choosing the superior methods. Methods. The thesis uses experimentation approach for evaluating the Boosted-Window Ensemble (BWE). CPU-time and the Prediction accuracy are used as performance indicators, where CPU-time is the execution time in seconds. The benchmark algorithms include: Accuracy-Updated Ensemble1 (AUE1), Accuracy-Updated Ensemble2 (AUE2), and Accuracy-Weighted Ensemble (AWE). The experiments use nine synthetic and five real-world datasets for generating performance estimates. The Asymptotic Friedman test and the Wilcoxon Signed-Rank test are used for hypothesis testing. The Wilcoxon-Nemenyi-McDonald-Thompson test is used for performing post-hoc analysis. Results. The hypothesis testing suggests that: 1) both for the synthetic and real-wrold datasets, the Boosted Window Ensemble (BWE) has significantly lower CPU-time values than two benchmark algorithms (Accuracy-updated Ensemble1 (AUE1) and Accuracy-weighted Ensemble (AWE). 2) BWE returns similar prediction accuracy as AUE1 and AWE for synthetic datasets. 3) BWE returns similar prediction accuracy as the three benchmark algorithms for the real-world datasets. Conclusions. Experimental results demonstrate that the proposed algorithm can be as accurate as the state-of-the-art benchmark algorithms, while obtaining predictions from the stream data. The results further show that the use of Partially-Updating Sliding Window has resulted in lower CPU-time for BWE as compared with the chunk-based sliding window method used in AUE1, AUE2, and AWE.
25

Evaluation of Machine Learning Algorithms for Classification of Short-Chain Dehydrogenase/Reductase Protein Sequences / Evaluering av Maskininlärningsalgoritmer för Klassificering av Short-Chain Dehydrogenase/Reductase Proteinsekvenser

Olandersson, Sandra January 2003 (has links)
The classification of protein sequences is a subfield in the area of Bioinformatics that attracts a substantial interest today. Machine Learning algorithms are here believed to be able to improve the performance of the classification phase. This thesis considers the application of different Machine Learning algorithms to the classification problem of a data set of short-chain dehydrogenases/reductases (SDR) proteins. The classification concerns both the division of the proteins into the two main families, Classic and Extended, and into their different subfamilies. The results of the different algorithms are compared to select the most appropriate algorithm for this particular classification problem. / Klassificeringen av proteinsekvenser är ett område inom Bioinformatik, vilket idag drar till sig ett stort intresse. Maskininlärningsalgoritmer anses här kunna förbättra utförandet av klassificeringsfasen. Denna uppsats rör tillämpandet av olika maskininlärningsalgoritmer för klassificering av ett dataset med short-chain dehydrogenases/reductases (SDR) proteiner. Klassificeringen rör både indelningen av proteinerna i två huvudklasser, Classic och Extended, och deras olika subklasser. Resultaten av de olika algoritmerna jämförs för att välja ut den mest lämpliga algoritmen för detta specifika klassificeringsproblem. / Sandra Olandersson Blåbärsvägen 27 372 38 Ronneby home: 0457-12084
26

Empirical Evaluation of Machine Learning Algorithms based on EMG, ECG and GSR Data to Classify Emotional States

Pandey, Amare Ketsela Tesfaye and Amrit January 2013 (has links)
The peripheral psychophysiological signals (EMG, ECG and GSR) of 13 participants were recorded in the well planned Cognition and Robotics lab at BTH University and 9 participants data were taken for further processing. Thirty(30) pictures of IAPS were shown to each participant individually as stimuli, and each picture was displayed for five-second intervals. Signal preprocessing, feature extraction and selection, models, datasets formation and data analysis and interpretation were done. The correlation between a combination of EMG, ECG and GSR signal and emotional states were investigated. 2- Dimensional valence-arousal model was used to represent emotional states. Finally, accuracy comparisons among selected machine learning classification algorithms have performed. Context: Psychophysiological measurement is one of the recent and popular ways to identify emotions when using computers or robots. It can be done using peripheral signals: Electromyography (EMG), Electrocardiography (ECG) and Galvanic Skin Response (GSR). The signals from these measurements are considered as reliable signals and can produce the required data. It is further carried out by preprocessing of data, feature selection and classification. Classification of EMG, ECG and GSR data can be conducted with appropriate machine learning algorithms for better accuracy results. Objectives: In this study, we investigate and analyzed with psychophysiological (EMG, ECG and GSR) data to find best classifier algorithm. Our main objective is to classify those data with appropriate machine learning techniques. Classifications of psychophysiological data are useful in emotion recognition. Therefore, our ultimate goal is to provide validated classified psychological measures for the automated adoption of human robot performance. Methods: We conducted a literature review in order to answer RQ1. The sources used are Inspec/ Compendex, IEEE, ACM Digital Library, Google Scholar and Springer Link. This helps us to identify suitable features required for the classification after reading the articles and papers that are peer reviewed as well as lie relevant to the area. Similarly, this helps us to select appropriate machine learning algorithms. We conducted an experiment in order to answer RQ2 and RQ3. A pilot experiment, then after main experiment was conducted in the Cognition and Robotics lab at the university. An experiment was conducted to take measures from EMG, ECG and GSR signal. Results: We obtained different accuracy results using different sets of datasets. The classification accuracy result was best given by the Support Vector Machine algorithm, which gives up to 59% classified emotional states correctly. Conclusions: The psychophysiological signals are very inconsistent with individual participant for specific emotion. Hence, the result we got from the experiment was higher with a single participant than all participants were together. Although, having large number of instances are good to train the classifier well. / The thesis is focused to classify emotional states from physiological signals. Features extraction and selection of the physiological signal was done, which was used for dataset formation and then classification of those emotional states. IAPS pictures were used to elicit emotional/affective states. Experiment was conducted with 13 participants in cognition and Robotics lab using biosensors EMG, ECG and GSR at BTH University. Nine participants data were taken for further preprocessing. We observed in our thesis the classification of emotions which could be analyzed by a combination of psychophysiological signal as Model A and Model B. Since signals of subjects are different for same emotional state, the accuracy was better for single participant than all participants together. Classification of emotional states is useful for HCI and HRI to manufacture emotional intelligence robot. So, it is essential to provide best classifier algorithms which can be helpful to detect emotions for developing emotional intelligence robots. Our work contribution lies in providing best algorithms for emotion recognition for psychophysiological data and selected features. Most of the results showed that SVM performed best with classification accuracy up to 59 % for single participant and 48.05 % for all participants together. For a single dataset and single participant, we found 60.17 % accuracy from MLP but it consumed more time and memory than other algorithms during classification. The rest of the algorithms like BNT, Naive Bayes, KNN and J48 also gave competitive accuracy to SVM. We conclude that SVM algorithm for emotion recognition from a combination of EMG, ECG and GSR is capable of handling and giving better classification accuracy among others. Tally between IAPS pictures with SAM helped to remove less correlated signals and to obtain better accuracies. Still the obtained results are small in percentage. Therefore, more participants are probably needed to get a better accuracy result over the whole dataset. / amarehenry@gmail.com ; Mobile: 0767042234 amrit.pandey111@gmail.com ; Mobile : 0704763190
27

Designing machine learning ensembles : a game coalition approach

Alzubi, Omar A. January 2013 (has links)
No description available.
28

Nanoscale modeling of materials: post deposition morphological evolution of fcc metal surfaces

Karim, Altaf January 1900 (has links)
Doctor of Philosophy / Department of Physics / Talat S. Rahman / This dissertation is an extensive study of several issues related to post deposition morphological evolution of fcc metal surfaces. These studies were carried out by probing the energetics and the dynamics of underlying atomistic mechanisms responsible for surface diffusion. An important aspect is the determination of relative probability of competing atomistic mechanisms and their contribution to controlling shapes and step edge patterns of nano structures on surfaces. In this scenario, the descent of adatoms from Ag islands on Ag(111) surface is examined. It shows an exchange mechanism to dominate over hopping and the process to favor the formation of (100)-microfacetted steps (A-type) over the (111)-microfacetted ones (B-type). Molecular dynamics simulations support these results at low temperature while at high temperature B-type step formation dominates. This change in the trend could happen if these processes leading to the formation of the A and B type steps have different values of their diffusion prefactors. This difference is confirmed on the basis of our calculations of the diffusion coefficients. Further, to understand the macroscopic properties of a system on the basis of its atomic scale information, spatial and temporal fluctuations of step edges on vicinal Cu(1 1 13) and Cu(1 1 19) surfaces is studied using kinetic Monte Carlo (KMC) simulations. These results show excellent agreement with experimental data, highlighting the role of mass transport along step edges, and also showing the validity of tools like KMC which aims at bridging the gap in length and time scales at which a range of interesting phenomena take place. To facilitate unbiased modeling of material properties, a novel way of performing KMC simulations is presented. In this approach the lists of diffusion processes are automatically collected during the simulation using a saddle-point search method in the potential energy landscape. The speed of the simulations is thus enhanced along with a substantial gain in reliability. Using this method the diffusion and coalescence of two-dimensional Cu and Ag adatom-island on Cu(111) and Ag(111) is studied. Together with input from molecular dynamics simulations, new processes involving the concerted motion of smaller islands are revealed. A significant difference in the scaling of the effective diffusion barriers with island size is observed for the sets of smaller (less than 10 atoms) and larger islands. In particular, the presence of concerted island motion leads to an almost linear increase in the effective diffusion barrier with size, while its absence accounts for strong size-dependent oscillations and anomalous behavior for trimers and heptamers. A crossover from diffusion due to the collective motion of the smaller island to a regime in which the island diffuses through the periphery dominated mass transport (large islands, 19 to 100 atoms) is predicted. For islands containing 19 to 100 atoms the scaling exponent is found to be in good agreement with that found in previous studies.
29

Robust Distribution-Free Learning Of Logic Expressions

Rajaraman, K 02 1900 (has links) (PDF)
No description available.
30

A Deep Learning Application for Traffic Sign Recognition

Kondamari, Pramod Sai, Itha, Anudeep January 2021 (has links)
Background: Traffic Sign Recognition (TSR) is particularly useful for novice driversand self-driving cars. Driver Assistance Systems(DAS) involves automatic trafficsign recognition. Efficient classification of the traffic signs is required in DAS andunmanned vehicles for safe navigation. Convolutional Neural Networks(CNN) isknown for establishing promising results in the field of image classification, whichinspired us to employ this technique in our thesis. Computer vision is a process thatis used to understand the images and retrieve data from them. OpenCV is a Pythonlibrary used to detect traffic sign images in real-time. Objectives: This study deals with an experiment to build a CNN model which canclassify the traffic signs in real-time effectively using OpenCV. The model is builtwith low computational cost. The study also includes an experiment where variouscombinations of parameters are tuned to improve the model’s performance. Methods: The experimentation method involve building a CNN model based onmodified LeNet architecture with four convolutional layers, two max-pooling layersand two dense layers. The model is trained and tested with the German Traffic SignRecognition Benchmark (GTSRB) dataset. Parameter tuning with different combinationsof learning rate and epochs is done to improve the model’s performance.Later this model is used to classify the images introduced to the camera in real-time. Results: The graphs depicting the accuracy and loss of the model before and afterparameter tuning are presented. An experiment is done to classify the traffic signimage introduced to the camera by using the CNN model. High probability scoresare achieved during the process which is presented. Conclusions: The results show that the proposed model achieved 95% model accuracywith an optimum number of epochs, i.e., 30 and default optimum value oflearning rate, i.e., 0.001. High probabilities, i.e., above 75%, were achieved when themodel was tested using new real-time data.

Page generated in 0.1004 seconds