Mancia Anguita

Analog CMOS implementation of a discrete time CNN with programmable cloning templates

An analog CMOS implementation of cells for building discrete-time cellular neural networks (DT-CNNs), based on current-mode multipliers and capacitive storage for the analog initial states and cloning templates, is presented. Since the cloning templates are programmable, the circuit could be used for several applications, or it could be reconfigured to perform different tasks on the initial input data sequentially. A chip prototype containing to DT-CNN cells has been designed and manufactured, and some experimental results showing the functionality of the circuit are provided. >

Optimization Strategies for High-Performance Computing of Optical-Flow in General-Purpose Processors

In this paper, we describe the high-performance implementation of an optical-flow algorithm that takes advantage of the processors architecture. Tuning the code, i.e., adapting it to take full advantage of the processor, is challenging, time consuming, and requires efficient programming at different levels but can lead to significant improvements in performance. The optimized implementation presented here is highly interesting for a number of applications since it delivers real-time motion estimations at high-image resolution on a PC or in an embedded system based on a general-purpose processor. In a 2.83 GHz Core 2 Quad PC, it achieves a speedup of 14 compared to our first code version and 2052.7f/s for the well-known 252 times 316 Yosemite sequence, and a speedup of 17.6 and 68.5 f/s for a 1016 times 1280 sequence. But the description of how this high-performance is achieved goes beyond a specific application since the paper presented here illustrates how inherently dense, low-level visual algorithms (pixel-wise computation) can be structured and improved to take full advantage of a standard processor. The implementation is compared with other hardware (based on FPGAs and GPUs) and software (based on clusters, PCs, and special-purpose processors) optical-flow implementations, showing that it outperforms them.

New methodology for the development of adaptive and self-learning fuzzy controllers in real time

Abstract This work proposes a procedure to design adaptive and self-learning fuzzy controllers in real time, requiring only a limited prior knowledge of the plant to be controlled, both in terms of the quantity and precision of this information. The algorithm does not need a mathematical model of the plant, or its approximation by means of a Jacobian matrix. Neither is it necessary to know the response desired at each instant of time, nor need there be previously available data. Auxiliary fuzzy controllers accomplish simultaneously the adaptation of the output scale factor (which is essential in the first steps of the control process) and learning of the parameters within the principal fuzzy controller (fuzzy rules). To verify the validity of the algorithm, real control problems were used: the stabilization of the temperature of a thermostat and level control within a liquid-filled tank. An analysis of the stability and robustness of the proposed algorithm is performed for different initial configurations of the fuzzy systems required by the algorithm and for abrupt changes in the plant to be controlled.

Analysis of the operators involved in the definition of the implication functions and in the fuzzy inference process

Abstract This paper analyzes the performance of some fuzzy implications proposed in the bibliography together with the operators needed for their definition and for the fuzzy inference process. Examining the specialized literature, it is clear that the selection of the best fuzzy implication operator has become one of the main question in the design of a fuzzy system, being occasionally contradictory (at presently there are more than 72 fuzzy implication proposed and investigated). An approach to the problem from a different perspective is given. The question is to determine whether the selection of the fuzzy implication operator is more important with respect to the behaviour of the fuzzy system than the operators (mainly T-norm, T-conorm and defuzzification method) involved in the definition of the implication function and in the rest of the inference process. The relevance and relative importance of the operators involved in the fuzzy inference process are investigated by using a powerful statistical tool, the ANalysis Of the VAriance (ANOVA) [Box et al., Statistics for experiments: an introduction to design, data analysis and model building, Wiley, New York, 1978; Montgomery, Design and Analysis of Experiments, Wiley, New York, 1984].

SCE toolboxes for the development of high-level parallel applications

Users of Scientific Computing Environments (SCE) benefit from faster high-level software development at the cost of larger run time due to the interpreted environment. For time-consuming SCE applications, dividing the workload among several computers can be a cost-effective acceleration technique. Using our PVM and MPI toolboxes, Matlab

Comparison of parallel multi-objective approaches to protein structure prediction

^{\rm {\sc {\textregistered}}}

A Spiking Neural Simulator Integrating Event-Driven and Time-Driven Computation Schemes Using Parallel CPU-GPU Co-Processing: A Case Study

and Octave users in a computer cluster can parallelize their interpreted applications using the native cluster programming paradigm — message-passing. Our toolboxes are complete interfaces to the corresponding libraries, support all the compatible datatypes in the base SCE and have been designed with performance and maintainability in mind. Although in this paper we focus on our new toolbox, MPITB for Octave, we describe the general design of these toolboxes and of the development aids offered to end users, mention some related work, mention speedup results obtained by some of our users and introduce speedup results for the NPB-EP benchmark for MPITB in both SCEs.

A new approach for the design of fuzzy controllers in real time

Protein structure prediction (PSP) is an open problem with many useful applications in disciplines such as medicine, biology and biochemistry. As this problem presents a vast search space and the analysis of each protein structure requires a significant amount of computing time, it is necessary to take advantage of high-performance parallel computing platforms as well as to define efficient search procedures in the space of possible protein conformations. In this paper we compare two parallel procedures for PSP which are based on different multi-objective optimization approaches, i.e. PAES (Knowles and Corne in Proc. Congr. Evol. Comput. 1:98–105, 1999) and NSGA2 (Deb et al. in IEEE Trans. Evol. Comput. 6:182–197, 2002). Although both procedures include techniques to take advantage of known protein structures and strategies to simplify the search space through the so-called rotamer library and adaptive mutation operators, they present different profiles with respect to their implicit parallelism.

Focal-Plane and Multiple Chip VLSI Approaches to CNNs

Time-driven simulation methods in traditional CPU architectures perform well and precisely when simulating small-scale spiking neural networks. Nevertheless, they still have drawbacks when simulating large-scale systems. Conversely, event-driven simulation methods in CPUs and time-driven simulation methods in graphic processing units (GPUs) can outperform CPU time-driven methods under certain conditions. With this performance improvement in mind, we have developed an event-and-time-driven spiking neural network simulator suitable for a hybrid CPU-GPU platform. Our neural simulator is able to efficiently simulate bio-inspired spiking neural networks consisting of different neural models, which can be distributed heterogeneously in both small layers and large layers or subsystems. For the sake of efficiency, the low-activity parts of the neural network can be simulated in CPU using event-driven methods while the high-activity subsystems can be simulated in either CPU (a few neurons) or GPU (thousands or millions of neurons) using time-driven methods. In this brief, we have undertaken a comparative study of these different simulation methods. For benchmarking the different simulation methods and platforms, we have used a cerebellar-inspired neural-network model consisting of a very dense granular layer and a Purkinje layer with a smaller number of cells (according to biological ratios). Thus, this cerebellar-like network includes a dense diverging neural layer (increasing the dimensionality of its internal representation and sparse coding) and a converging neural layer (integration) similar to many other biologically inspired and also artificial neural networks.

VLSI implementations of CNNs for image processing and vision tasks: single and multiple chip approaches

This paper presents a new methodology to achieve real time self tuning and self-learning in fuzzy controllers. The advantage of this approach is that it only requires qualitative information about the plant to be controlled, in terms of the monotony presented by the output with respect to the control signal and delays of the plant. Thus, it is capable of controlling highly nonlinear systems, in a pseudo-optimum way, even when these are time variable. Control is achieved by means of two auxiliary systems: the first one is responsible for adapting the consequences of the main controller to minimize the error arising at the plant output, while the second auxiliary system compiles real input/output data obtained from the plant. The system then learns from these data, adapting both the consequences of the rules and the parameters that define the membership functions, taking into account, not the current state of the plant but rather the global identification performed.