Browsing by Author "Lei, Yu"

Now showing 1 - 3 of 3
  • Thumbnail Image
    A Combinatorial Approach to Hyperparameter Optimization
    Khadka, Krishna; Chandrasekaran, Jaganmohan; Lei, Yu; Kacker, Raghu N.; Kuhn, D. Richard (ACM, 2024-04-14)
    In machine learning, hyperparameter optimization (HPO) is essential for effective model training and significantly impacts model performance. Hyperparameters are predefined model settings which fine-tune the model’s behavior and are critical to modeling complex data patterns. Traditional HPO approaches such as Grid Search, Random Search, and Bayesian Optimization have been widely used in this field. However, as datasets grow and models increase in complexity, these approaches often require a significant amount of time and resources for HPO. This research introduces a novel approach using 𝑡-way testing—a combinatorial approach to software testing used for identifying faults with a test set that covers all 𝑡-way interactions—for HPO. 𝑇 -way testing substantially narrows the search space and effectively covers parameter interactions. Our experimental results show that our approach reduces the number of necessary model evaluations and significantly cuts computational expenses while still outperforming traditional HPO approaches for the models studied in our experiments.
  • Thumbnail Image
    Functional Regression and Adaptive Control
    Lei, Yu (Virginia Tech, 2012-09-13)
    The author proposes a novel functional regression method for parameter estimation and adaptive control in this dissertation. In the functional regression method, the regressors and a signal which contains the information of the unknown parameters are either determined from raw measurements or calculated as the functions of the measurements. The novel feature of the method is that the algorithm maps the regressors to the functionals which are represented in terms of customized test functions. The functionals are updated continuously by the evolution laws, and only an infinite number of variables are needed to compute the functionals. These functionals are organized as the entries of a matrix, and the parameter estimates are obtained using either the generalized inverse method or the transpose method. It is shown that the schemes of some conventional adaptive methods are recaptured if certain test function designs are employed. It is proved that the functional regression method guarantees asymptotic convergence of the parameter estimation error to the origin, if the system is persistently excited. More importantly, in contrast to the conventional schemes, the parameter estimation error may be expected to converge to the origin even when the system is not persistently excited. The novel adaptive method are also applied to the Model Reference Adaptive Controller (MRAC) and adaptive observer. It is shown that the functional regression method ensures asymptotic stability of the closed loop systems. Additionally, the studies indicate that the transient performance of the closed loop systems is improved compared to that of the schemes using the conventional adaptive methods. Besides, it is possible to analyze the transient responses a priori of the closed loop systems with the functional regression method. The simulations verify the theoretical analyses and exhibit the improved transient and steady state performances of the closed loop systems.
  • Thumbnail Image
    Resource Management with Smart Antenna in CDMA Systems
    Lei, Yu (Virginia Tech, 2001-11-12)
    Third generation (3G) mobile communication systems will provide services supporting high-speed data network and multimedia applications in addition to voice applications. The Smart antenna technique is one of the leading technologies that helps to meet the requirement by such services to radio network capacity. Resource management schemes such as power control, handoff and channel reservation/assignment are also essential for providing the seamless services with high quality. Smart antenna techniques will help to enhance the capability of resource management through more efficient and flexible use of resources. In this thesis, adaptive array and switched beam antenna techniques are compared in terms of algorithm, performance, complexity and hardware requirements. Based on these comparisons, sub-optimal code gate algorithm are most likely the suitable algorithms for next generation code division multiple access (CDMA) systems due to its good performances, robustness, and low complexity. A multi-cell CDMA simulator is developed for investigating the gain from smart antenna techniques in both bit error rate (BER) performance improvement and enhancement to resource management schemes. Our study shows that smart antenna techniques can significantly improve the performance of the system and help to build more powerful and flexible resource management schemes. With eight array elements, the system capacity can be increased by a factor of four. Power control command rates can be reduced through the tradeoff with the interference reduction by smart antennas. Smart antennas will also reduce handover failure rates and further increase the system capacity by reducing the resources reserved for soft handover.