هاشم پروين

Given the increasing trend of household electrical energy consumption and its direct impact on household costs and grid sustainability, the development of new energy management frameworks for smart homes has become an inevitable necessity. The increasing proliferation of smart home technologies and dynamic electricity pricing also requires attention to optimizing home energy consumption through intelligent scheduling of appliances. Traditional home energy management systems (HEMS) are often ill-equipped to handle the complexities of dynamic electricity pricing and changing user needs. This complexity requires advanced solutions that can intelligently plan device usage to optimize for multiple and often conflicting goals. Advanced reinforcement learning algorithms such as deep Q-networks offer promising solutions by learning optimal policies through interaction with the environment. These systems can dynamically adjust device schedules based on real-time data, such as energy prices and user behavior patterns, without the need for explicit programming for each scenario. In this study, a deep Q-network model for a home energy management system is presented, which aims to optimize the scheduling of home appliances using deep reinforcement learning algorithms and compare their performance with baseline methods. For this purpose, a set of electricity tariff data over a ten-week period is generated and simulated, along with a dataset of user preferences including three selected time slots and, in order, the priority for turning on four scheduleable home appliances, including: dishwasher, vacuum cleaner, iron, and washing machine, are given as input to the simulated environment. Four different algorithms were designed and implemented for appliance scheduling: random slot selection based on user preferences, fixed first-priority selection, greedy algorithm based on lowest-cost slot selection, and a deep reinforcement learning-based method called Cost-Convenience Optimized Deep Q-Network (CCO-DQN) algorithm that enables dynamic and adaptive learning by utilizing neural networks and reward policies. Simulation results show that the proposed algorithm is able to significantly reduce household energy costs while maintaining user comfort levels. Compared to traditional methods and heuristic algorithms, the proposed framework has been able to provide better performance in both reducing the total energy cost and increasing system flexibility. The main contributions of this thesis in relation to the design of the neural network model of the proposed algorithm are the design of a dedicated architecture of a deep Q-neural network that is able to simultaneously model the impact of variable energy tariffs and user preferences. In this architecture, multidimensional input vectors including dynamic tariffs, time priority score, and device performance constraints are considered. The second is the dual-objective reward design (cost-comfort), where through an innovative mechanism, the reward is shaped to simultaneously reduce the electricity cost and maintain user comfort. This function establishes a stable balance between the conflicting objectives with adaptive weighting. The third is the optimization of the state-action representation, where the state space is modeled to include the instantaneous energy prices and the user preference vector, and the action space reflects the flexibility of the device scheduling. This design results in more efficient exploration and faster convergence of the model. By combining multi-objective optimization, network architecture customization, and learning stability enhancement techniques, these contributions propose the CCO-DQN algorithm as a more efficient alternative to stochastic and greedy baseline methods in home energy management.

Text-to-image synthesis, a fundamental task in generative artificial intelligence, seeks to produce realistic images that match natural language descriptions. This study proposes a dual-attention adversarial generative network that uses an innovative multi-stage architecture to enhance the issues of visual realism, semantic coherence, and fine-grained detail generation. The DA2GAN model uses two distinct phases, a self-directed phase to generate low-resolution drafts, and an alignment phase to enhance these drafts to high-resolution outputs. The dual-attention technique allows DA2GAN to focus on both textual and visual features, ensuring accurate representation of descriptive elements and producing coherent, high-quality images. We evaluate DA2GAN using the CUB and Oxford-102 datasets and evaluate the competitive results on well-known metrics such as IS, FID, and RP. The proposed model outperforms previous models in both qualitative and quantitative comparisons. The qualitative comparisons further emphasize the capacity of the model to produce clearer and more realistic images with better agreement with text descriptions. The innovations demonstrated in DA2GAN demonstrate its potential as an advanced framework for text-to-image synthesis that can be used in creative content development, automated design, and image enhancement

Sentiment analysis is an essential task in numerous domains, necessitating effective dimensionality reduction and feature extraction techniques. This study introduces MultiObjective Manifold Representation for Opinion Mining (MOMR). This novel approach combines deep global and local manifold feature extraction to reduce dimensions while capturing intricate data patterns efficiently. Additionally, incorporating a self-attention mechanism further enhances MOMR's capability to focus on relevant parts of the text, resulting in improved performance in sentiment analysis tasks. MOMR was evaluated against established techniques such as Long Short-Term Memory (LSTM), Naive Bayes (NB), Support Vector Machines (SVM), Recurrent Neural Networks (RNN), and Convolutional Neural Networks (CNN), as well as recent state-of-the-art models across multiple datasets including IMDB, Fake News, Twitter, and Yelp. Therefore, our comparative analysis underscores MOMR's efficacy in sentiment analysis tasks across diverse datasets, highlighting its potential and applicability in real-world sentiment analysis applications. On the IMDB dataset, MOMR achieved an accuracy of 99.7% and an F1 score of 99.6%, outperforming other methods such as LSTM, NB, SMSR, and various SVM and CNN models. For the Twitter dataset, MOMR attained an accuracy of 88.0% and an F1 score of 88.0%, surpassing other models, including LSTM, CNN, BiLSTM, Bi-GRU, NB, and RNN. In the Fake News dataset, MOMR demonstrated superior performance with an accuracy of 97.0% and an F1 score of 97.6%, compared to techniques like RF, RNN, BiLSTM+CNN, and NB. For the Yelp dataset, MOMR achieved an accuracy of 80.0% and an F1 score of 80.0%, proving its effectiveness alongside other models such as Bidirectional Encoder Representations from Transformers (BERT), aspect-sentence graph convolutional neural network (ASGCN), Multi-layer Neural Network, LSTM, and bidirectional recurrent convolutional neural network attention (BRCAN).

Our brain is capable of describing and categorizing the images that appear before us. But how can a computer process an image and identify it with an appropriate and accurate description? This seemed unattainable a few years ago, but with the advancement of machine vision algorithms and deep learning, as well as the availability of datasets and suitable artificial intelligence models, creating an appropriate description generator for an image has become easier. Image captioning is also a growing industry worldwide. The process of generating image captions involves converting images into a series of words using a series of pixels. Image captioning can be seen as an end-to-end challenge in the form of a sequence-to-sequence challenge. To achieve this goal, it is necessary to process both words and images. In this thesis, first, an explanation of image captioning and its applications in various fields is presented, and then, the evolutionary course of image captioning methods is examined. Various methods that have been proposed over time for image captioning have been comprehensively reviewed. This coherent classification helps us to gain a deeper understanding of the techniques and methods available in image captioning. Also, recent articles in the field of image captioning have been reviewed in this thesis. Based on the results obtained from the review of recent articles, the necessity of continuing research in the field of image captioning has been emphasized. These researches can lead to significant improvements in existing methods for image captioning and also the discovery of newer and more advanced methods. In this thesis, an encoder-decoder method based on attention has been used. Unlike previous methods where attention was only applied to one of the sections, the attention mechanism has been applied to both the image and the text. This is a new idea in this field, and the final caption is generated word by word. The FLICKR8K dataset has been used, and the evaluation metrics used are BLEU (1,2,3,4), ROUGE, and METEOR. The results are 51, 49, 48, 44, 52, and 37.5 respectively. These results indicate an improvement over previous method.

Breast cancer is one of the most common cancers among women, which endangers the patient's life if not diagnosed and predicted on time. A variety of medical imaging methods as well as biopsies help doctors to diagnose breast cancer. Since biopsy is invasive, therefor using medical imaging methods is safe for patients. A physician can diagnose the presence of a tumor by examining breast images. Computer science used in medical image processing and disease diagnosis. Deep learning is a subset of artificial intelligence that has achieved promising results in processing all kinds of images, especially medical images. In this research a proposed convolution neural network, a proposed canvolutioanl auto-encored, a pre-trained ResNet-50, a trained from scratch ResNet-50, a pre-trained Inception v3 and a trained from scratch Inception v3 were developed to classify ultrasound breast images. There are two scenarios for images classification. In the first scenario, images were classified into two classes of benign cancer and malignant cancer. In the second images were classified into three classes of benign cancer, malignant cancer and healthy control. The dataset contains 780 images in three classes: healthy or normal (133 images), malignant cancer (210 images) and benign cancer (487 images). The number of images increased to 6,413 images (including 2,185 benign cancer images, 2,100 malignant cancer images, and 2,128 healthy images) using data augmentation methods such as 5 degrees rotating, flipping horizontally, and flipping vertically. In the next step, the size of the images was 150*125 pixeles. The highest accuracy of 97% was obtained by pre-trained Inception V3 to classify images into two classes. The highest precision of 100% was obtained by trained from scratch ResNet-50 in benign calss. Pre-trained Inception v3 achieved the precision of 98% in malignant cancer calss. For sensetivity in benign calss and malignant class, the highest value of 98% and 99% were obtained by from scratch trained Inception V3 and in common canvolutioanl auto-encored and trained from scratch ResNet-50, respectively. The highest F1 score of 99% and 98% were obtained by the pre-trained Inception v3 in both classes of benign and malignant. For three classes classification, the pre-trained Inception v3 achived the highest accuracy of 96%. Also, the highest precision, senstivity and F1 score of 100%, 96% and 98% were obtained by pre-trained Inception v3 in benign class. In healty control group, the highest F1 score of 98%, 99% and 98% was obtained by convolution neural network, convolution neural network and pre-trained Inception v3.

سیستم های توصیه گر یکی از موفق ترین ابزارها برای مقابله با سرریز داده ها شناخته می شوند که روزبه روز استفاده از آن ها گسترده تر می شود. این سیستم ها یک نوع ویژه از سیستم های پالایش اطلاعات هستند که آیتم ها را بر اساس جذابیت آن ها برای کاربر از یک مجموعه بزرگ از آیتم ها پالایش می کنند. این سیستم ها سعی دارند، بر اساس عملکرد، سلیقه های شخصی، رفتارهای کاربر و بسته به زمینه ای که در آن مورد استفاده قرار می گیرند به هر کاربر پیشنهادهایی را ارائه دهند که با تمایلات شخصی وی تطابق داشته و کاربر را در فرایند تصمیم گیری یاری نمایند. سیستم های توصیه گر پالایش گروهی یکی از پرکاربردترین و مؤثرترین روش های توصیه محسوب می شوند که با بررسی انتخاب های کاربران در گذشته، الگوهایی را در داده ها پیدا می کنند که با توجه به آن الگوهای رفتاری برای هر کاربر توصیه مناسب را ارائه می دهند. روش های مبتنی بر پالایش گروهی معمولاً از سه مشکل اصلی رنج می برند که شامل: شروع سرد، پراکندگی داده ها و مقیاس پذیری کم می باشند. در راستای برطرف نمودن چالش های گفته شده، این پایان نامه، دو روش جدید مبتنی بر تجزیه نامنفی ماتریس و اطلاعات اعتماد کاربران در شبکه های اجتماعی، ارائه می شود. چون روش تجزیه نامنفی ماتریس یکی از کاراترین روش های مبتنی بر مدل برای سیستم های توصیه پالایش گروهی است، لذا از اطلاعات اعتماد و عدم اعتماد بین کاربران برای کمک به تجزیه درست و دقیق ماتریس رتبه بندی، استفاده می کنیم تا رتبه های نامشخص برای کاربران و آیتم هایی که شروع سرد دارند با دقت مناسبی پیش بینی شود و در برخورد با داده های پراکنده، بتوان با دقت مناسبی سلیقه کاربر هدف را تخمین زد. همچنین برای افزایش مقیاس پذیری و کاهش پیچیدگی الگوریتم ها از روش های بهینه سازی کارآمد برای حل تابع هدف نهایی استفاده می شود. بعلاوه، سیستم های توصیه گر به دلیل پویا بودن محیط و افزایش سریع اطلاعات، روزبه روز فضای مسئله آن ها بزرگ تر می شود، لذا الگوریتم های دقیق (ریاضی) در حالات مختلفی نمی توانند جواب بهینه را تولید کنند و دنبال کردن جواب دقیق در اکثر مواقع خیلی سخت و پرهزینه است. بعلاوه، در اکثر موارد جواب تقریبی و نزدیک به جواب واقعی، می تواند برای ما رضایت بخش باشد. بنابرین، دو روش پیشنهادی جدید مبتنی بر الگوریتم های فراابتکاری ارائه می شود که