The Fast-evolving Landscape of Deep Learning: What’s New and What’s Next | Chapter 3 | Engineering Research: Perspectives on Recent Advances Vol. 10

 

Deep learning is rapidly evolving with transformative breakthroughs in foundation models (GPT-4, Gemini), generative AI (diffusion models, video synthesis), and self-supervised learning, reducing reliance on labelled data. Key advances in efficient AI (edge computing, model compression) and reinforcement learning (robotics, autonomous systems) are expanding practical applications. Emerging frontiers like neurosymbolic AI and AGI research highlight both progress and unresolved challenges. This chapter examines these cutting-edge developments, offering insights into deep learning’s current state and future trajectory.

 

 

Author(s) Details

K. Sridhar
Department of Computer Science and Engineering, Malla Reddy (MR) Deemed to be University, Maisammaguda (H), Gundlapochampally (V), Medchal - Malkajgiri District, Telangana, India.

 

Ravikumar Thallapalli
Department of Computer Science and Engineering, Vaageswari College of Engineering (Autonomous) Accredited by NAAC A+, Beside LMD Police Station, Karimnagar, Telangana, India.

 

M. Srinivas
Department of Computer Science and Engineering (AI&ML), Vaageswari College of Engineering (Autonomous) Accredited by NAAC A+, Beside LMD Police Station, Karimnagar, Telangana, India.

 

P. Venkateshwarlu
Department of Computer Applications, Vaageswari College of Engineering (Autonomous) Accredited by NAAC A+, Beside LMD Police Station, Karimnagar, Telangana, India.

 

Please see the book here :-https://doi.org/10.9734/bpi/erpra/v10/6079

The Role of Machine Learning in Modern Modeling | Chapter 6 | Mathematics and Computer Science: Contemporary Developments Vol. 7

This chapter highlights the shift from traditional mathematical modeling, grounded in theoretical principles and differential equations, to data-driven approaches that leverage machine learning and empirical data. While conventional models offer structured frameworks for understanding systems, they can be limited in flexibility and scalability. In contrast, data-driven models uncover patterns from large datasets and handle complex, non-linear systems without relying solely on theoretical assumptions. By integrating machine learning with traditional models, accuracy and adaptability improve significantly. Different machine learning techniques, including supervised and reinforcement learning, extract valuable insights, especially in cases where traditional models falter. Hybrid models combining physics-based approaches with data-driven techniques enhance prediction capabilities, such as in energy consumption forecasts for smart grids. The chapter also addresses challenges like data quality and model transparency, emphasizing how hybrid models improve interpretability and predictive power. Case studies demonstrate the benefits of integrating machine learning with traditional models in enhancing model robustness and accuracy. In summary, the fusion of machine learning and traditional methods creates more reliable models, especially for complex systems where conventional approaches face limitations. [1, 2, 3, 4, 5].

 

Author (s) Details

 

Ghada Awad Elkarim Mohammed Ahmed Ahmed
Department of Mathematics, Faculty of Science, Al-Baha University, Alaqiq 65799, Saudi Arabia.

 

Please see the book here:- https://doi.org/10.9734/bpi/mcscd/v7/2649

Study about Intelligent Virtual Agent: Learning How to Make Compromises | Chapter 11 | New Visions in Science and Technology Vol. 9

 The goal of the study is to provide the learning agent more control over the path to achieving a goal. The study's precise goals are to investigate the prospect of intelligent virtual agents learning to make acceptable choices only when all user-defined prerequisites for how to attain a goal are met. To fulfil these goals, this chapter presents a modification of the Q-learning algorithm. It is hoped that by doing so, it will be able to achieve aims such as modelling shopping therapy, knowing others' preferences, determining whether a shopping habit is becoming a problem, recognising cognitive memory problems, and modelling behaviour specific to different age groups. The use of a measurements model (a model of environmental criteria and/or emotional models), represented as a new memory matrix, is introduced to help the Q-learning agent identify the best approach to the objective by meeting specific difficult criteria. The learning agent can compromise a criterion if the goal cannot be achieved by following the pre-set criteria. To achieve the goal, the agent makes the fewest possible choices and suitable concessions. If the criteria are grouped according to their importance, the agent will be able to choose a greater number of acceptable compromises rather than unacceptable ones. Three separate intelligent learning agents were trained using the updated algorithm: a shopping cart agent, a gift-shopping agent, and a broker. Their behaviour has improved as a result of the exams.

Author(S) Details

Dilyana Budakova
Technical University of Sofia, Plovdiv Branch, Bulgaria.

Veselka Petrova-Dimitrova
Technical University of Sofia, Plovdiv Branch, Bulgaria.

Lyudmil Dakovski
European Polytechnical University, Pernik, Bulgaria.

View Book:- https://stm.bookpi.org/NVST-V9/article/view/4790