Gianni Orlandi

Fast neural networks without multipliers

Multilayer perceptrons (MLPs) with weight values restricted to powers of two or sums of powers of two are introduced. In a digital implementation, these neural networks do not need multipliers but only shift registers when computing in forward mode, thus saving chip area and computation time. A learning procedure, based on backpropagation, is presented for such neural networks. This learning procedure requires full real arithmetic and therefore must be performed offline. Some test cases are presented, concerning MLPs with hidden layers of different sizes, on pattern recognition problems. Such tests demonstrate the validity and the generalization capability of the method and give some insight into the behavior of the learning algorithm.

A systolic circuit for fast Hartley transform

The authors present a systolic circuit for computing a fast algorithm performing the discrete Hartley transform (DHT). The proposed architecture employs a systolic elevator concept and CORDIC processors. The elevator assures local communications in the proposed algorithm, and the CORDIC processor makes it possible to enhance processing speed and exploit parallelism. The architecture appears to be regular and, therefore, very attractive for VLSI realizations. The computational cost necessary for computing the DFT (discrete Fourier transform) is also discussed with respect to other architectures.<<ETX>>

finite wordlength digital filter design using an annealing algorithm

A very versatile algorithm for the design of finite-wordlength filters is described. It is a simulated annealing algorithm, derived directly from the simulated annealing algorithm for functions of continuous variables which searches over discrete values of the filter coefficients. It has very fast development time and produces filters with good performances. Its only drawback is a very long execution time. Three examples of filter design are included.<<ETX>>

Multi-layer perceptrons with discrete weights

The feasibility of restricting the weight values in multilayer perceptrons to powers of two or sums of powers of two is studied. Multipliers could be thus replaced by shifters and adders on digital hardware, saving both time and chip area, under the assumption that the neuron activation function is computed through a lookup table (LUT) and that a LUT can be shared among many neurons. A learning procedure based on back-propagation for obtaining a neural network with such discrete weights is presented. This learning procedure requires full real arithmetic and therefore must be performed offline. It starts from a multilayer perceptron with continuous weights learned using back-propagation. Then a weight normalization is made to ensure that the whole shifting dynamics is used and to maximize the match between continuous and discrete weights of neurons sharing the same LUT. Finally, a discrete version of BP algorithm with automatic learning rate control is applied up to convergence. Some test runs on a simple pattern recognition problem show the feasibility of the approach

Improved evoked potential estimation using neural network

The possibility of using the multilayer perceptron (MLP) neural network for the processing of the evoked potentials (EPs) is analyzed. In this case, the process can be conceived as deterministic low amplitude signals (damped sine waves), corresponding to the brains response to stimuli, embedded in strongly colored noise, the EEG background activity. Typical values of the signal-to-noise ratio are less than 0 dB. The network, used as a nonlinear filter, is trained using iteratively as the input signal one of a set of available EP ensembles and as the target signal another ensemble of the same set. Experimental results, both on synthetic and real data, show that the method provides good results with very few EP ensembles. Therefore, it allows a noteworthy reduction of the signal nonstationarity and the patients annoyance

Design of multi-layer neural networks with powers-of-two weights

The feasibility of restricting the weight values to powers-of-two or sums of powers-of-two in multilayer neural networks is discussed. A learning procedure based on back-propagation to obtain a neural network with such weights is presented. This learning procedure requires full real arithmetic, and therefore must be performed offline. These neural networks do not require multipliers, and are well suited for high-speed and high-integration digital neural circuits. To show the effectiveness of the approach, tests on a pattern recognition problem are presented.<<ETX>>

An IoT-inspired cloud-based web service architecture for e-Health applications

E-Health services can take advantage of the technological achievements in the area of the Internet of Things (IoT), and of the cost reduction and increasing user-friendliness of health monitoring devices. Homes equipped with environmental sensors, physiological parameters monitoring devices, and home automation devices, could become the “hardware” of an “operating system” for application developers and service providers. The system would expose web services through a unique cloud infrastructure for users data collection and storage, administration and billing, and healthcare service provisioning applications by possibly multiple third parties. We present an instance of a cloud-based web server which relies on a “home system” for the collection of information from an heterogeneous set of devices, providing a high level description of the proposed overall architectural model, of the induced opportunities from the market perspective, and of how it could be used by healthcare applications developers and service providers, including details on how the web server Application Programming Interfaces (API) is implemented in our instance.

Neural networks with self-adaptive topology

A method is presented to dynamically adapt the topology of a neural network using only the information of the learning set. The method, which simply consists of eliminating connections from an initial fully connected network, presents characteristics which can resemble some biological behavior. Several experimental results obtained with recurrent and multilayered networks are reported to demonstrate the capabilities of the proposed method.<<ETX>>

An adaptive neural network filter for evoked potentials

The possibility of using the multilayer perceptron (MLP) neural network for the processing of EEG evoked potentials (EPs) is examined. A structure composed of the cascade of a MLP and a linear combiner is proposed. Experimental results, both on synthetic and real data, show that the method provides good results with very few EP ensembles and without the necessity of prior knowledge of the signal characteristics.<<ETX>>

A clustering approach to multi-source localization in reverberant rooms

Localization of acoustic sources in the presence of reverberation is still a challenging task in audio signal processing. As a matter of fact, commonly adopted models are not adequate to describe real scenarios. Moreover, practical systems should not employ sophisticated and expensive architectures, that require precise synchronization and fast data shuffling among sensors. This work describes a new robust multi-step procedure for speaker localization in reverberant rooms. The proposed approach is based on a disturbed harmonics model of time delays in the frequency domain and employs the well-known ROOT-MUSIC algorithm, after a proper pre-processing of the received signals. Final clustering of raw TDOA estimates gives candidate source positions. Among the appealing features of the proposed approach are the capability of tracking multiple speakers simultaneously and the high accuracy of the closed form TDOA estimator.