Yevrosiyo matematik nazariya va kompyuter fanlari jurnali

Curious students who begin learning modern programming languages with the, regardless of the language, inevitably become acquainted with graphics libraries. The following article provides information on the graphical tools of Borland C++ Builder, as well as tips and instructions for creating multi-form applications.

This article examines new algorithmic solutions and their underlying mathematical foundations. It analyzes complex mathematical principles necessary for understanding and developing advanced algorithms in areas such as machine learning, quantum computing, blockchain technology, optimization, and graph theory. The article highlights the role of linear algebra, calculus, probability theory, and information theory in modern algorithms. It also discusses mathematical approaches to pressing issues such as algorithmic fairness and interpretability. By synthesizing existing research and conceptual ideas, the article demonstrates the inseparable connection between abstract mathematical theory and practical algorithmic innovations, and outlines future research and application directions.

Ushbu ilmiy maqolada sun'iy neyron tarmoqlarning giperparametrlarini avtomatik optimallashtirish muammosi tadqiq etilgan. Giperparametrlarni an’anaviy usullar bilan optimallashtirish ko‘pincha samarasiz va resurs talab qiluvchi jarayon hisoblanadi. Tadqiqotda genetik algoritmlar yordamida neyron tarmoqlarning (o‘rganish tezligi, qatlamlar soni, neyronlar soni, aktivatsiya funksiyasi, bat hajmi kabi) giperparametrlarini avtomatik tanlash va optimallashtirish usuli taklif etilgan. Genetik algoritm populyatsiyasi har bir individ (xromosoma) orqali giperparametrlar to‘plamini ifodalaydi, moslik funksiyasi sifatida esa validatsiya to‘plamidagi aniqlik (accuracy) ishlatilgan. Tanlov, krossover va mutatsiya operatorlari yordamida avlodlar o‘tishi natijasida giperparametrlarning eng yaxshi kombinatsiyasi aniqlangan. MNIST, CIFAR-10 va Iris ma’lumotlar to‘plamlarida olib borilgan eksperimentlar natijasida taklif etilgan usulning an’anaviy usullarga (grid search, random search) qaraganda 15-25% tezroq va aniqroq optimallashtirish imkonini berishi ko‘rsatilgan. Shuningdek, ushbu usul neyron tarmoqlarning ishlash ko‘rsatkichini o‘rtacha 3-8% oshirishi isbotlangan.