SES School of Engineering and Science
The thesis focuses on directed evolution and structure function relationship of phytase from Yersinia mollaretii (Ymphytase). The “key beneficial” mutations identified in the directed evolution have been iteratively combined. Ymphytase variant with 54% improvement in thermostability (58°C for 20 min) and 200 U/mg improved activity was achieved. MD simulations results showed decreased overall Ymphytase flexibility in thermostable variant with slight increase in active site loop flexibility. The decreased flexibility might be due to improved intra-protein interactions like hydrogen bonds (G187S, K289E) and salt-bridge interaction (T77K).
Three conceptually novel methods for protein engineering have been developed in part-III (Chapter 6 – 8). Combinatorial assembly of site saturation test in protein segment (ProCASTing), a sequence independent method was developed for parallel site saturation of more than one consecutive site (4 to 8) in any part of the protein. Multisite combinatorial assembly of site saturation test (OmniCASTing), practically simple method was developed for parallel site saturation of more than one site (five) regardless of positions in the gene. Using OmniCASTing, a variant with improved thermostability, pH stability and activity was obtained. Three properties improvement might be due to cooperative effects between the new combinations of mutations compare to parent combination. Protein consensus based surface engineering (ProCoS) method combining computational analysis and molecular biology tools was developed. The utility of ProCoS method has been demonstrated by surface engineering of Ymphytase that yielded a variant with 34 amino acid substitution (20% negative polar amino acids) and 3.8 fold improvement in pH stability (pH 2.8). Two hypotheses have been proposed in part-IV (chapter 8 – 9) and validated by experimental evidences.
In most developed countries working hours of truck drivers are con- strained by hours of service regulations. When optimizing vehicle routes, trucking companies must consider these constraints in order to assure that drivers can comply with the regulations. This paper studies the combined vehicle routing and truck driver scheduling problem (VRTDSP), which generalizes the well-known vehicle-routing problem with time windows by considering working hour constraints. A branch-and-price algorithm for solving the VRTDSP with U.S. hours of service regulations is presented. This is the first algorithm that solves the VRTDSP to proven optimality.
An iterative neural architecture based on repeated application of the Denoising Autoencoder is introduced. The architecture is placed in the family of other approaches involving networks of simple units and iteration at the exploitation stage. It is shown that repeated feeding of a pattern to a Denoising Autoencoder often yields non-trivial sensible improvements of the pattern. This statement is supported by a classification experiment, in which the data transformed by our architecture is shown to be more linearly separable than the original samples.
This paper studies the revised hours of service regulations for truck drivers in the United States which will enter into force in July 2013. It provides a detailed model of the new regulation and presents and a new simulation-based method to assess the impact of the rule change on operational costs and road safety. Unlike previous methodologies, the new methodology takes into account that carriers can use optimization as a tool to minimize the economic impact of stricter regulations. Simulation experiments are conducted indicating that the monetized safety benefit of reducing the daily driving time limits is on the same order of magnitude compared to the additional operational costs.
There are many attempts in the field of robotics and artificial intelligence towards achieving a well-functioning architecture that can control an agent in a stable and elegant manner, while allowing for mixing of behaviors. However, state of the arts robots are far away from the coordinated, complex and graceful behavior that we can observe in animals. A popular approach towards tackling complex motor control problems is the employment of pattern generators. Several architectures have shown good performance, modulating and switching between different patterns. However, those methods perform gradual changes rather than the fast and yet smooth modulations of a pattern for a short duration of time, characteristic for reflexive actions. In this report I present a practical study of a technique that achieves changes in amplitude, frequency and shift of a periodic signal in the time frame of less than a period.
In this report we present a method of adding a feedback control mechanism to an echo state network (ESN) pattern generator in order to modulate its output patterns with the purpose of tracking slowly varying control targets, e.g. shift, amplitude, or frequency of an oscillatory pattern. A proofof- principle case study is presented where a basic ESN is trained to produce a stable sinewave oscillation with fixed shift, amplitude and frequency. With the controller in place, the system demonstrates that the shift, amplitude and frequency of the produced sine waveform can be modulated simultaneously by suitably generated slow varying control signals inserted into the network. Furthermore, an equilibration procedure is introduced to relearn ESN weights such that the equilibrated ESN pattern generator can approximately reproduce the reservoir dynamics across the controllable range, with the feedback control loop switched off. As a result, when reconnecting the feedback control loop to the equilibrated ESN, the energy of the control signals are many orders of magnitude smaller compared to the native system.
Current advances in reservoir computing have demonstrated that fixed random recurrent networks with only readouts trained often outperform fully-trained recurrent neural networks. While full supervised training of such networks is problematic, intuitively there should also be something better than a random network. In this contribution we investigate a different approach which is in between the two. We use reservoirs derived from recursive self-organizing maps that are trained in an unsupervised way and later tested by training supervised readouts. This approach enables us to train greedy unsupervised hierarchies of such dynamic reservoirs. We demonstrate in a rigorous way the advantage of using the self-organizing reservoirs over the traditional random ones and using hierarchies of such over single reservoirs with a synthetic handwriting-like temporal pattern recognition dataset.
In complex pattern recognition tasks data usually exhibits many local distortions which significantly disturb the recognition process. A method for extracting temporal features from a signal that are invariant to these distortions is presented in this report. The idea is to use Echo State Network to generate a rich high-dimensional representation of data. Temporal features are then extracted by finding projections of the high-dimensional representation that are minimally influenced by the selected distortions while still carrying most of the information about the underlying signal required for the performed task. The algorithm performance is analyzed on synthetic signals as well as on high-dimensional handwriting data for shift and scale distortions. It is shown that the algorithm is capable to extract a low dimensional feature set from a reservoir which is invariant to the selected distortions and relevant to the performed task.
We consider identification of operator families defined via a time-frequency series expansion of the operator spreading function. The identification problem is transformed into an infinite-dimensional linear algebra problem. Our aim is to establish a connection between the identifiability of the operator family and a density measure of the time-frequency index set. In this way, the identification problem can be compared to the classical density condition for existence of Gabor frames. The conclusion is that the relationship between identifiability of such operator families and the critical density is highly intricate because of the presence of additional conditions. Criteria for identifiability are developed for families of time-frequency localized operators defined via time-frequency series expansions of the spreading function based on the Gaussian function.
Constructive design of Gabor frame windows is rare, and most results come from the one-dimensional case. The connection between the geometry of fundamental domains of lattices and Gabor systems was explored first in a series of papers by Han and Wang [HW01], [HW04]. We build upon these results to construct Gabor frames with smooth and compactly supported window functions in higher dimensions. For this purpose we study pairs of lattices with equal density allowing compact and star-shaped fundamental domains. Concrete examples are provided and the results are extended to other special class of lattices. In addition, we make observations on the intricate behavior of Gabor systems with multivariate Gaussian windows.
