ABSTRACT
Normalization-based realization of adaptive techniques is presented to locate the exon segments in gene sequences. Based on filter fluctuations at its input and output, adjustment of coefficients of tap is done consequently. For this reason, normalization of AEPs can be used to daze the issues related to strengthening of gradient noise for large input information vector. With this, normalizing the adjustment applied to the coefficient of filter weight vector is done in every iteration based on input vector squared norm. At this time, two types of normalization, namely, data and error normalizations, are considered. The resulting techniques include normalized LMS (NLMS) and error-normalized LMS (ENLMS) algorithms. For further improving the performance, normalized variants of LMF and VSLMS algorithms including normalized LMF (NLMF) and variable step size normalized LMS (VNLMS) algorithms are deliberated. All these techniques are extended to sign-based algorithms so as to further reduce the computational complexity. Adaptive filters with larger length are required in real time as gene sequences are typically with more length. Here, LMS algorithm is computationally expensive to implement. Therefore, all their maximum versions are deliberated to significantly reduce computational complexity.
These techniques are further extended to sign-based realizations to decrease the complexity of computations. In addition to filtering techniques by using adaptation, this chapter also provides a performance analysis on the basis of various metrics such as specificity, sensitivity, precision, computational complexity issues, and convergence analysis. Also, a detailed discussion of various normalization-based algorithms considered for genomic sequence analysis is presented.
