(2005). Neurofuzzy knowledge processing in intelligent learning environments for improved student diagnosis. Information Sciences 170 (2-4) 273-307. This is an author-produced version of a paper published in Information Sciences . This version has been peer-reviewed but does not include the final publisher proof corrections, published layout or pagination.All articles available through Birkbeck ePrints are protected by intellectual property law, including copyright law. Any use made of the contents should comply with the relevant law. Abstract. In this paper, a neural network implementation for a fuzzy logic-based model of the diagnostic process is proposed as a means to achieve accurate student diagnosis and updates of the student model in Intelligent Learning Environments. The neuro-fuzzy synergy allows the diagnostic model to some extent "imitate" teachers in diagnosing students' characteristics, and equips the intelligent learning environment with reasoning capabilities that can be further user to drive pedagogical decisions depending on the student learning style. The neuro-fuzzy implementation helps to encode both structured and non-structured teachers' knowledge: when teachers' reasoning is available and well defined, it can be encoded in the form of fuzzy rules; when teachers' reasoning is not well defined but is available through practical examples illustrating their experience, then the networks can be trained to represent this experience. The proposed approach has been tested in diagnosing aspects of student's learning style in a discovery-learning environment that aims to help students to construct the concepts of vectors in physics and mathematics. The diagnosis outcomes of the model have been compared against the recommendations of a group of five experienced teachers, and the results produced by two alternative soft computing methods. The results of our pilot study show that the neuro-fuzzy model successfully manages the inherent uncertainty of the diagnostic process; especially for marginal cases, i.e. where it is very difficult, even for human tutors, to diagnose and accurately evaluate students by directly synthesizing subjective and, some times, conflicting judgments.2

No abstract

Objectives. In this work, a simple technique to assess the image quality characteristics of the postprocessed image is developed and an easy to use figure of image quality (FIQ) is introduced. This FIQ characterizes images in terms of resolution and noise. In addition information capacity, defined within the context of Shannon's information theory, was used as an overall image quality index. Materials and Methods. A digital mammographic image was postprocessed with three digital filters. Resolution and noise were calculated via the Modulation Transfer Function (MTF), the coefficient of variation, and the figure of image quality. In addition, frequency dependent parameters such as the noise power spectrum (NPS) and noise equivalent quanta (NEQ) were estimated and used to assess information capacity. Results. FIQs for the “raw image” data and the image processed with the “sharpen edges” filter were found 907.3 and 1906.1, correspondingly. The information capacity values were 60.86 × 103 and 78.96 × 103 bits/mm2. Conclusion. It was found that, after the application of the postprocessing techniques (even commercial nondedicated software) on the raw digital mammograms, MTF, NPS, and NEQ are improved for medium to high spatial frequencies leading to resolving smaller structures in the final image.

The paper investigates retraining options and the performance of pre-trained Convolutional Neural Networks (CNNs) for sound classification. CNNs were initially designed for image classification and recognition, and, at a second phase, they extended towards sound classification. Transfer learning is a promising paradigm, retraining already trained networks upon different datasets. We selected three ‘Image’- and two ‘Sound’-trained CNNs, namely, GoogLeNet, SqueezeNet, ShuffleNet, VGGish, and YAMNet, and applied transfer learning. We explored the influence of key retraining parameters, including the optimizer, the mini-batch size, the learning rate, and the number of epochs, on the classification accuracy and the processing time needed in terms of sound preprocessing for the preparation of the scalograms and spectrograms as well as CNN training. The UrbanSound8K, ESC-10, and Air Compressor open sound datasets were employed. Using a two-fold criterion based on classification accuracy and time needed, we selected the ‘champion’ transfer-learning parameter combinations, discussed the consistency of the classification results, and explored possible benefits from fusing the classification estimations. The Sound CNNs achieved better classification accuracy, reaching an average of 96.4% for UrbanSound8K, 91.25% for ESC-10, and 100% for the Air Compressor dataset.

No abstract