Aun Neow Poo

Evaluation of simple performance measures for tuning SVM hyperparameters

Choosing optimal hyperparameter values for support vector machines is an important step in SVM design. This is usually done by minimizing either an estimate of generalization error or some other related performance measure. In this paper, we empirically study the usefulness of several simple performance measures that are inexpensive to compute (in the sense that they do not require expensive matrix operations involving the kernel matrix). The results point out which of these measures are adequate functionals for tuning SVM hyperparameters. For SVMs with L1 soft-margin formulation, none of the simple measures yields a performance uniformly as good as k-fold cross validation; Joachims’ Xi-Alpha bound and the GACV of Wahba et al. come next and perform reasonably well. For SVMs with L2 soft-margin formulation, the radius margin bound gives a very good prediction of optimal hyperparameter values.

A Fast Dual Algorithm for Kernel Logistic Regression

This paper gives a new iterative algorithm for kernel logistic regression. It is based on the solution of a dual problem using ideas similar to those of the Sequential Minimal Optimization algorithm for Support Vector Machines. Asymptotic convergence of the algorithm is proved. Computational experiments show that the algorithm is robust and fast. The algorithmic ideas can also be used to give a fast dual algorithm for solving the optimization problem arising in the inner loop of Gaussian Process classifiers.

Adaptive iso-planar tool path generation for machining of free-form surfaces

The iso-planar (Cartesian) tool path generation method has been used for several decades. However, it suffers an inherent drawback: in the region where the direction of the surface normal is close to that of the parallel intersecting planes, the intersecting plane intervals have to be reduced because of the influence of surface slopes. This causes redundant tool paths in the associated flatter regions and results in lower machining efficiency. This paper presents an algorithm that overcomes the disadvantage of the iso-planar method while keeping its advantages of robustness and simplicity. In this algorithm, the concept of isophote is applied to partition the surface into different regions. In each region the tool path side steps are adaptive to the surface features. Therefore redundant tool paths are avoided. By applying the region-by-region or global-local machining strategy, the machining efficiency is increased.

Oriented bounding box and octree based global interference detection in 5-axis machining of free-form surfaces

Global interference detection is a critical problem in 5-axis NC machining of free-form surfaces. Based on the hierarchical oriented bounding box (OBB) which is used in virtual reality to detect spatial collisions between 3D objects, a new global interference detection method is developed in this paper. In this method, in order to simplify the computation process of updating tool positions and orientations in 5-axis machining, the cutter and cutter holder are modeled by a hierarchical OBB structure, whereas the workpiece surfaces are approximated by an octree. Interference detection is conducted between the tool OBBs and the gray octants of the surface octree with the separating axis theorem. With the hierarchical structure of octree, if interference is found in one octant, its sub-octants are further processed to locate the exact colliding leaf nodes and the discretized surface points contained in these leaf nodes are tested with a conventional vector calculation method for exact interference detection; if no interference is detected, all the sub-octants are then considered as interference free and are not processed further. Meanwhile, with the hierarchical structure of the tool OBBs, should interference occur between octants and the OBBs in the first level of the hierarchical structure, the sub-OBBs in the second level would be further tested. Otherwise it could be determined with certainty that there is no interference between the tool and the octant.

Focal length calibration from two views: method and analysis of singular cases

We consider the problem of estimating the focal length of a camera from two views while the focal length is not varied during the motion of the camera. An approach based on Kruppas equations is proposed. Specifically, we derive two linear and one quadratic equations to solve the problem. Although the three equations are interdependent in general, each one may be singular for different configurations. We study in detail the generic singularities of the problem and the actual singularities of the individual calibration equations. Results of our experiments using synthetic and real data underline the effect that singular configurations may have on self-calibration. However, these results are stable once the singularities are avoided.

Multi-category classification by soft-max combination of binary classifiers

In this paper, we propose a multi-category classification method that combines binary classifiers through soft-max function. Posteriori probabilities are also obtained. Both, one-versus-all and one-versus-one classifiers can be used in the combination. Empirical comparison shows that the proposed method is competitive with other implementations of one-versus-all and one-versus-one methods in terms of both classification accuracy and posteriori probability estimate.

A uniform biped gait generator with offline optimization and online adjustable parameters

This paper presents the Genetic Algorithm Optimized Fourier Series Formulation (GAOFSF) method for stable gait generation in bipedal locomotion. It uses a Truncated Fourier Series (TFS) formulation with its coefficients determined and optimized by Genetic Algorithm. The GAOFSF method can generate human-like stable gaits for walking on flat terrains as well as on slopes in a uniform way. Through the adjustment of only a single or two parameters, the step length and stride-frequency can easily be adjusted online, and slopes of different gradients are accommodated. Dynamic simulations show the robustness of the GAOFSF, with stable gaits achieved even if the step length and stride frequency are adjusted by significant amounts. With its ease of adjustments to accommodate different gait requirements, the approach lends itself readily for control of walking on a rough terrain and in the presence of external perturbations.

Adjustable Bipedal Gait Generation using Genetic Algorithm Optimized Fourier Series Formulation

This paper presents a method for optimally generating stable bipedal walking gaits, based on a truncated Fourier series formulation with coefficients tuned by genetic algorithm. It also provides a way to adjust the stride-frequency, step-length or walking pattern in real-time. The proposed approach to gait synthesis is not limited by the robot kinematic structure and can be used to satisfy various motion assumptions. It is also easy to generate optimal gaits on terrains of different slopes or on stairs under different motion requirements. Dynamic simulation results show the validity and robustness of the approach. The gaits generated resulted in human-like motions optimized for stability, even walking speed and lower leg-strike velocity of the swing foot

Microassembly Fabrication of Tissue Engineering Scaffolds With Customized Design

This paper presents a novel technique to fabricate scaffold/cell constructs for tissue engineering by robotic assembly of microscopic building blocks (of volume 0.50.50.2 and 60 thickness). In this way, it becomes possible to build scaffolds with freedom in the design of architecture, surface morphology, and chemistry. Biocompatible microparts with complex 3-D shapes were first designed and mass produced using MEMS techniques. Semi-automatic assembly was then realized using a robotic workstation with four degrees of freedom integrating a dedicated microgripper and two optical microscopes. Coarse movement of the gripper is determined by pattern matching in the microscopes images, while the operator controls fine positioning and accurate insertion of the microparts. Successful microassembly was demonstrated using SU-8 and acrylic resin microparts. Taking advantage of parts distortion and adhesion forces, which dominate at micro-level, the parts cleave together after assembly. In contrast to many current scaffold fabrication techniques, no heat, pressure, electrical effect, or toxic chemical reaction is involved, a critical condition for creating scaffolds with biological agents.

Robotic micro-assembly of scaffold/cell constructs with a shape memory alloy gripper

Describes an integrated approach to design and fabricate scaffold/cell constructs for tissue engineering. With this approach it becomes possible to produce scaffolds with controlled distribution of living cells and growth factors, a critical condition for successful grafting. Our idea consists of building a scaffold/cell construct by robotic micro-assembly of microscopic polymer building blocks. The paper introduces the rationale and concept of this interdisciplinary project and presents some realized steps. A 3D contact FEM simulation has been carried out to study the forces involved on the scaffold elements and micro-gripper during assembly. An error analysis has been performed to evaluate the accumulated error when building a scaffold/cell construct. A dedicated monolithic shape memory alloy micro-gripper has been realized and tested, which is able to handle parts in the range of 50-100 microns.