Συγκριτική μελέτη αλγορίθμων διάχυσης για τη γένεση τεχνητών εικόνων
A comparative study of algorithms for the generation of artistic images

View/ Open
Keywords
Μοντέλα διάχυσης ; Βαθιά γενετική μοντελοποίηση ; Παραγωγή εικόνων ; Μηχανική μάθηση ; Επεξεργασία φυσικής γλώσσας ; Νευρωνικά δίκτυα ; Υπολογιστική όραση ; Γενετικά μοντέλα ; Μοντέλα διάχυσηςAbstract
Diffusion models are a class of generative models that simulate the gradual diffusion of noise into data and learn to reverse this process to create new high-quality data. Their basic idea is based on a two-step process: progressively adding noise to the training data (forward process) to turn it into random noise, and reversing this process (reverse process), where the model is trained to gradually remove the noise, producing samples that resemble the original data. This approach makes diffusion models highly stable during training and capable of generating realistic and diverse data. They have wide application in areas such as image generation, audio processing, and natural language processing, with Stable Diffusion and DALL-E being well-known examples.